BEGIN:VCALENDAR VERSION:2.0 PRODID:PHP METHOD:PUBLISH X-WR-CALNAME:MCS Events BEGIN:VEVENT DTSTART;TZID=America/Chicago:20090128T150000 DTEND;TZID=America/Chicago:20090128T160000 DTSTAMP:20130525T020110 LOCATION:Building 221 Conference Room A261, Argonne National Laboratory DESCRIPTION:Carbon emission programs are designed to reduce greenhouse gas emissions by implementing either a carbon tax or a cap-and-trade program. In this talk, we discuss the extent to which a foreign oil producer, such as OPEC, can manipulate cap-and-trade programs by cutting production, resulting in a collapse of some carbon emission markets. This possible manipulation needs to be understood in an international setting with trade among developing countries that have no regulation and other countries that each have their own independent carbon emission program. We analyze a leader-follower computable general equilibrium model to understand this issue that results in mathematical programs with equilibrium constraints that need to be solved. Numerical results providing insights into the possible manipulation of carbon emission programs by foreign oil producers are provided.\n\n SUMMARY:The Manipulation of Carbon Emission Programs by Foreign Oil Producers UID:508 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20090205T160000 DTEND;TZID=America/Chicago:20090205T170000 DTSTAMP:20130525T020110 LOCATION:GCIS W301, University of Chicago DESCRIPTION:The activities of my research group and collaborations focus on the development of bioinformatics systems essential for functional genomics, genetics and phenotypic research. The sequencing of mouse, human and other genomes and the rapid accumulation of very large data sets has resulted in an overwhelming amount of information from multiple sources containing a variety of content and formats. The challenge is to bring all the data together and make it easily accessible to researchers directly and/or for additional computer analysis. Our current research centers on combining bio-ontologies (defined, controlled, structured vocabularies) and database systems to identify molecular elements that contribute to the processes of particular diseases, such as lung cancer. This work is undertaken as part of the Gene Ontology Consortium, a group of 19 model organism databases and genome annotation centers. My group, as part of the Mouse Geneome Informatics Consortiun at The Jackson Laboratory, is responsible for the functional and comparative annotation of mouse genes. SUMMARY:Evidence and Inference: Comparative Biology in the Age of Genomics UID:528 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20090402T150000 DTEND;TZID=America/Chicago:20090402T160000 DTSTAMP:20130525T020110 LOCATION:Building 221 Conference Room A261, Argonne National Laboratory DESCRIPTION:Learn how to prepare an award winning LDRD proposal from the pros! SUMMARY:Hot Tips for preparing DCG LDRD proposals UID:533 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20090225T150000 DTEND;TZID=America/Chicago:20090225T160000 DTSTAMP:20130525T020110 LOCATION:Building 221 Conference Room A261, Argonne National Laboratory DESCRIPTION:In this informal talk we discuss several aspects related to\nthe importance and use of uncertainty in modeling and simulation. Uncertainty has become pervasive in real computations. In this context, the deterministic solution represents just one possible outcome. We briefly discuss the need for uncertainty in simulations and present several computational efficient models for uncertainty representation in reactive flow simulations. Ties between uncertainty quantification and data assimilation - the process of integrating measurements in simulations - will also be addressed.\n SUMMARY:The certain importance of uncertainty in modeling and simulation UID:534 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20090129T103000 DTEND;TZID=America/Chicago:20090129T113000 DTSTAMP:20130525T020110 LOCATION:Building 360, Conference Room L119, Argonne National Laboratory DESCRIPTION:Scientific codes are used on a wide range of hardware architectures and operating systems. The developers and users of these codes need to be sure that the code performs correctly on all of the systems where it is used. The scale of this task can be quite large, especially when you consider that various revisions of the same operating system are best treated separately. To assist in the testing effort, the National Science Foundation (NSF) funds a build and test pool consisting of the hardware and software used by the NSF community. This saves each development group from having to maintain their own build and test systems. In addition, the NSF funds the development of Metronome, a continuous-integration build and test system. Automating the use of the build systems lowers the cost of development by discovering bugs when they are committed.\n\nThis talk will cover the design principles of Metronome, as well as practical experience from using it on a large distributed computing application. It will also compare and contrast the use of Metronome with BuildBot, a popular open-source build tool. SUMMARY:Build and Test for Distributed Computing Applications UID:540 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20090130T130000 DTEND;TZID=America/Chicago:20090130T140000 DTSTAMP:20130525T020110 LOCATION:RI 405 (Research Institutes), University of Chicago DESCRIPTION:High performance computing has become an increasingly important fixture in science, from aiding in the processing of data collected in experiments, to acting as a virtual laboratory in which experiments are done. Thus, high performance computing is creating a third branch of scientific effort. This trend has driven research and development in a variety of different areas from fundamental hardware design to the software that makes the resources useful. With each iteration of this development cycle computational science has become more and more\ncomplex. This effort addresses this complexity in two key interrelated areas: visualization and collaboration.\n\nVisual representation is the key method to simplify the explanation of a complex environment. Consequently, a large research and development community effort has grown to support scientific visualization. It has also spawned research in advanced displays to provide infrastructure\nfor exploring data products. These include immersive displays like the CAVE Automatic Virtual Environment or high resolution displays constructed of multiple individual units like the ActiveMural. This work includes influential contributions in all of these areas.\n\nAt the same time, complex tasks are often simplified by effort sharing. We see that the teams of individuals working together to do this new form of science have become larger and more distributed. Research efforts in collaboration technology have grown to address this problem. Here we describe the Access Grid and tools built for sharing information as part of this effort. As will be seen in this thesis, collaboration technology both relies on visualization technology and supports it in enabling interactions at a distance.\n\nThroughout this work we have taken a user driven iterative approach using real applications from a variety of scientific domains. This end-to-end testbed approach guarantees realistic experimental circumstances with real world stresses and constraints.\n\nThe main contributions of this dissertation are: a) discovery of requirements for the connecting of collaboration and visualization technology to high performance computing; b) development of infrastructure and demonstrations for enabling coupled advanced displays and high performance resources, including the first remote\nconnection of two spatially immersive virtual environments (CAVE to CAVE); and c) development of infrastructure for efficient pixel transport using commodity video codecs to support collaborative scientific visualization. SUMMARY:Visualization and Collaboration Technologies to Support High-Performance Computing Research UID:539 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20090202T103000 DTEND;TZID=America/Chicago:20090202T113000 DTSTAMP:20130525T020110 LOCATION:Building 221 Conference Room A216, Argonne National Laboratory DESCRIPTION:The representational accuracy of the isosurface meshes produced by marching isosurfacing methods is usually defined by the closeness of the produced mesh to the isosurface given by trilinear interpolation. In this seminar, a new metric that evaluates the accuracy of marching isosurfacing methods is introduced. The new metric is an accurate estimate of spatial discrepancy between a produced mesh and the trilinear interpolation isosurface. Computation of the new metric is also described. Experimental results of the accuracy examination of several well-known isosurfacing methods using the new metric are presented. An analysis of the accuracy and rendering cost of the mesh produced by each considered method is also given. In addition, possible future extensions of this research work are discussed. SUMMARY:Evaluation of Accuracy of Marching Isosurfacing Methods UID:548 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20090203T103000 DTEND;TZID=America/Chicago:20090203T113000 DTSTAMP:20130525T020110 LOCATION:Building 221 Conference Room A216, Argonne National Laboratory DESCRIPTION:Very large, software-intensive systems are normally built from a combination of established legacy components and new components contributing new functionality. The cost of building such systems and maintaining them through-life can be reduced by adopting the principles and architectures of Open Systems. These principles and architectures will be discussed. The case for the benefits to be obtained from Open Systems will be argued. \n \nExamples of Open Systems will be drawn from defense applications, from open source applications and from web services, in support of this thesis. It is hoped that the presentation will engender a lively debate, which will allow the speaker to tailor the content to the types of application of interest to the audience. SUMMARY:Systems and Open Architecture – the benefits case UID:550 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20090311T150000 DTEND;TZID=America/Chicago:20090311T160000 DTSTAMP:20130525T020110 LOCATION:Building 221 Conference Room A261, Argonne National Laboratory DESCRIPTION:With the advance in performance and capabilities of modern computers, the drive towards large-scale integrated simulations of complex flow systems is growing. The different flow physics in different components of these systems often calls for integration of different solvers. In this talk, we present a computational methodology that we developed for coupling compressible and low Mach number codes, motivated by the necessity to model systems where flow conditions vary substantially - from supersonic to almost stagnating.\n\nWe describe the details of implementation on structured overlapping meshes, formulate unsteady interface conditions, validate our choice of interface conditions by comparing numerical errors for different formulations, and show the results of numerical experiments performed on a wide range of steady and unsteady laminar and turbulent problems.\n SUMMARY:Code Integration for Large-Scale Systems Simulations: Compressible - Low Mach Number Coupling UID:556 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20090216T143000 DTEND;TZID=America/Chicago:20090216T153000 DTSTAMP:20130525T020110 LOCATION:Ryerson 251, University of Chicago DESCRIPTION:There is a sea change happening in academic research -- a transformation caused by a data deluge that is affecting all disciplines. Modern science increasingly relies on integrated information technologies and computation to collect, process, and analyze complex data. Data-centric science is the \"Fourth Paradigm.\" Tools, technologies, and platforms must seamlessly integrate into standard scientific methodologies and processes. Microsoft External Research is committed to open access, open tools, and interoperability in the heterogeneous world of academic research. This talk will illustrate the far-reaching changes that this new paradigm will have on scientific discovery. SUMMARY:eScience and the Fourth Paradigm UID:557 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20090304T150000 DTEND;TZID=America/Chicago:20090304T160000 DTSTAMP:20130525T020110 LOCATION:Building 221 Conference Room A261, Argonne National Laboratory DESCRIPTION:In machine learning and data mining, data are often represented in numerical form, as a matrix or a tensor. In many cases, the data are onnegative. For example, in text mining, a collection of documents is converted to a term-document matrix whose (i,j) entry is the frequency of term i in document j. In face recognition, a face image in grayscale is represented as a nonnegative matrix, and a set of face images forms a 3-dimensional data tensor.\n\nLow-rank approximation by principal component analysis (for a matrix) or tensor decomposition (for a tensor) is frequently used to filter out noise or capture important features. However, certain properties may be lost due to the introduction of negative values. It has been proven by numerical evidence that adding the nonnegative constraints helps detect the essential features of the data. The resulting technique is called nonnegative matrix factorization (NMF) for matrices or nonnegative tensor factorization (NTF) for tensors.\n\nThe NMF and NTF computation can be formulated as a simple-bound nonconvex optimization problem, to minimize the matrix distance or divergence subject to the nonnegativity constraints. Various alternating algorithms have been proposed to solve it. We show that virtually all alternating algorithms for NMF can be adapted to compute NTF, and draw the relations between the algorithms in the literature. The initialization schemes and sparseness constraints for NMF and NTF will also be discussed.\n SUMMARY:From Nonnegative Matrix Factorization to Nonnegative Tensor Factorization UID:558 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20090218T150000 DTEND;TZID=America/Chicago:20090218T160000 DTSTAMP:20130525T020110 LOCATION:Building 221 Conference Room A261, Argonne National Laboratory DESCRIPTION:CyberSecurity is a growing concern especially in open grids, where attack propagation is easy due to existing collaborations. We consider how to respond optimally to an attack in grid environments.\n\nWe present an optimization model that takes the existing collaborations as input and minimizes the disruption to the grid whilst reducing threat-levels at unaffected sites. Our optimization model outputs which collaborations must be suspended or monitored to reduce threat-levels at unaffected sites.\n\n SUMMARY:Optimal Response to Attacks on The Open Science Grid UID:559 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20090216T103000 DTEND;TZID=America/Chicago:20090216T113000 DTSTAMP:20130525T020110 LOCATION:Bldg: 221, Conference Room A216, Argonne National Laboratory DESCRIPTION:As the types of problems we solve in high-performance computing and other areas become more complex, the amount of data generated and used is growing at a rapid rate. Today many terabytes of data are common; tomorrow petabytes of data will be the norm. One of the challenges in high-performance computing is to provide users with reliable data access in a distributed, heterogeneous environment. In this talk, we will review the existing I/O paradigms in high-performance computing environments and explore better alternatives across both local and wide-area networks. We propose three different techniques to accommodate the I/O requirements of scientific applications. We present a new design for a high-performance, scalable parallel file system that obviates the need for dedicated I/O and metadata servers by utilizing Object-based Storage Devices. We also propose a new remote I/O paradigm that takes advantage of the increasing popularity of high-speed wide-area networks and centralized data repositories to perform I/O over wide-area networks. Lastly, we present a scalable I/O forwarding solution that attempts to bridge the increasing performance gap between the processing power and the I/O subsystems of massively-parallel leadership-class machines such as the IBM Blue Gene/P. SUMMARY:Rethinking I/O in High-Performance Computing Environments UID:560 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20090415T150000 DTEND;TZID=America/Chicago:20090415T160000 DTSTAMP:20130525T020110 LOCATION:Building 221 Conference Room A261, Argonne National Laboratory DESCRIPTION:In this work we improve the existing tools for the recovery and prediction of human decisions based on multiple factors. We use essentially a latent factor method, and obtain the decision-influencing factors from the observed correlations in the available statistical information by singular value decomposition-based principal factor identification. We generalize on widely-used linear representations of decision-making functions by using adaptive high-order polynomial interpolation and applying an iterative and adaptive post-processing to arrive at an estimated probability of every possible outcome of a decision. The novelty of the method consists in the use of flexible, nonlinear predictive functions, and in the suggested post-processing procedure. Our experiments show that the introduced approach is at least competitive in the class of SVD-based prediction methods, and that the precision grows with the increase in the order of the polynomial basis. We suggest that the method may be successfully applied instead of a widely used linear SVD-based methods.\n\n SUMMARY:Polynomial Interpolation for Predicting Decisions and Recovering Missing Data UID:561 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20090223T103000 DTEND;TZID=America/Chicago:20090223T113000 DTSTAMP:20130525T020110 LOCATION:Building 221, Conference Rm. A216, Argonne National Laboratory DESCRIPTION:Funded by the Japanese government, the University of Tokyo, University of Tsukuba, and Kyoto University are developing a seamless and highly productive parallel programming \nenvironment for high performance computing. Started in 2008, this four-year research and development initiative aims to provide a new portable, efficient, and convenient parallel programming environment for a variety of machines, such as small-scale to large-scale PC clusters and next-generation petascale supercomputers. Researchers are developing a new parallel programming language, numerical libraries, and runtime systems. After presenting an overview of the project, researchers will discuss the portable, single runtime environment \ndeveloped at the University of Tokyo, which consists of a portable user-level file system and an implementation-independent MPI runtime system. Using this runtime environment, the binary code developed in the PC cluster may run in a supercomputer center, although the supercomputer runtime environment is different. The presentation will conclude with a preview of future research efforts. SUMMARY:Towards Seamless and High-Productive Parallel Programming Environment UID:562 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20090319T150000 DTEND;TZID=America/Chicago:20090319T160000 DTSTAMP:20130525T020110 LOCATION:Building 221, Room A216, Argonne National Laboratory DESCRIPTION:Optimizing large-scale scientific applications for a variety of modern high-performance computing systems is a non-trivial process given the state-of-the-art in performance tuning software. Current performance systems do not facilitate large-scale performance experiments that require multiple application runs nor do they present performance data in a form that would enable the application scientists to efficiently optimize their codes. We will present a component environment that automates performance data collection, storage, analysis, and visualization for parallel scientific applications. This system aims to enable application scientists to efficiently collect meaningful performance data that will facilitate their efforts in code optimizations.\n SUMMARY:Automatic Generation of Performance/Memory Models for Parallel Scientific Applications UID:563 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20090224T100000 DTEND;TZID=America/Chicago:20090224T110000 DTSTAMP:20130525T020110 LOCATION:Bldg: 360, Conference Room L119, Argonne National Laboratory DESCRIPTION:This talk will have two parts. In the first part, we will survey the current situation with regard to programming models for scalable parallel computers, identifying some important research needed in this area. In the second, we will present the Asynchronous Dynamic Load Balancing library (ADLB), a package that is being used to present a much simpler programming model than message passing while allowing applications to scale to (at least) tens of thousands of processors. It is being used on Intrepid for Argonne\'s GFMC nuclear physics code, an INCITE award winner, but is likely to be useful for other applications as well. SUMMARY:Programming Models, Languages, and Libraries, and a Simple API for Specifying Very Large Computation UID:565 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20090325T150000 DTEND;TZID=America/Chicago:20090325T160000 DTSTAMP:20130525T020110 LOCATION:Building 221, Room A216, Argonne National Laboratory DESCRIPTION:With the increasing scaling of manufacturing technology, process variation is a phenomenon that has become more prevalent. As a result, in the context of Chip Multiprocessors (CMPs) for example, it is possible that identically-designed processor cores on the chip have non-identical peak frequencies and power consumptions. To cope with such a design, each processor can be assumed to run at the frequency of the slowest processor, resulting in wasted computational capability. This talk considers an alternate approach and proposes an algorithm that intelligently maps (and remaps) computations onto available processors so that each processor runs at its peak frequency.\n\n SUMMARY:Process Variation Aware Thread Mapping for Chip Multiprocessors UID:566 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20090302T160000 DTEND;TZID=America/Chicago:20090302T170000 DTSTAMP:20130525T020110 LOCATION:Charles M. Harper Center, 5708 S. Woodlawn Room 10, University of Chicago DESCRIPTION:To be announced. SUMMARY:Energy Challenges in the 21st Century UID:568 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20090408T150000 DTEND;TZID=America/Chicago:20090408T160000 DTSTAMP:20130525T020110 LOCATION:Building 221, Room A216, Argonne National Laboratory DESCRIPTION:The Boltzmann Transport Equation (BTE) is a powerful and general conservation equation which is capable of describing in detail the transport of electrons and phonons in metals, semiconductors, and a variety of solid-state devices. It is often called semi-classical because it describes particles as classical point particles, but includes scattering through quantum-mechanical perturbation theory. Therefore the BTE is capable of describing electrical and thermal transport down to the nanoscale. New structures, such as Carbon Nanotubes (CNTs), provide new challenges to the transport modeling community. In this talk we will discuss the BTE and its application to modeling electrical transport in single-walled CNTs. We will present a simple approach to discretizing the BTE using the upwind method, and cover electron-phonon scattering and the Linear Analytic method for computing scattering rates for use in transport simulation. Finally we will reflect on some possible extensions to enable coupled electro-thermal transport and parallel implementations. SUMMARY:Boltzmann Equations for Nanoscience Applications UID:570 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20090316T103000 DTEND;TZID=America/Chicago:20090316T000008 DTSTAMP:20130525T020110 LOCATION:Building 221 Conference Room A216, Argonne National Laboratory DESCRIPTION:In this talk, I will describe my work on optimizing matrix-vector multiplication with combinatorial techniques. My research has focused on two different combinatorial scientific computing topics related to matrix-vector multiplication.\n\nFor the first topic, I address the optimization of serial matrix-vector multiplication for relatively small, dense matrices, which can be used in finite element assembly. Previous work showed that combinatorial optimization of matrix-vector multiplication can lead to faster evaluation of finite element stiffness matrices by removing redundant operations. Based on a graph model characterizing row relationships, a more efficient set of operations can be generated to perform matrix-vector multiplication. I improved this graph model by extending the set of binary row relationships and using hypergraphs to model more complicated row relationships, yielding significantly improved results over previous models.\n\nFor the second topic, I address parallel matrix-vector multiplication for large sparse matrices. Parallel sparse matrix-vector multiplication is a particularly important numerical kernel in computational science. We have focused on optimizing the parallel performance of this operation by reducing the communication volume through smarter two-dimensional matrix partitioning. We have developed and implemented a recursive algorithm based on nested dissection to partition structurally symmetric matrices. In general, this method has proven to be the best available for partitioning structurally symmetric matrices (when considering both volume and partitioning time) and has shown great promise for information retrieval matrices. SUMMARY:Optimizing Matrix-Vector Multiplication with Combinatorial Techniques UID:571 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20090302T130000 DTEND;TZID=America/Chicago:20090302T143000 DTSTAMP:20130525T020110 LOCATION:Building 221 Confernece Room A216, Argonne National Laboratory DESCRIPTION:Chelsio Communications is leading the convergence of networking, storage and clustering interconnects with its robust, high-performance and proven unified wire technology. Featuring a highly scalable and programmable architecture, Chelsio offers 10-Gigabit Ethernet and multi-port Gigabit Ethernet adapter cards, delivering the low latency and superior throughput required for high-performance computing applications.\n\nThe topics that will be covered by Chelsio’s CEO and President will be:\n\n- An introduction to Chelsio Communications\n- Discussion of Chelsio’s product offerings \n- The directions and features that the 10GE market is talking and how Chelsio is addressing it\n- A brief discussion of Chelsio’s future roadmap\n- Discussion of Argonne\'s needs and future directions in Networking, Clustering and Storage\n- How may Chelsio best support Argonne?\n SUMMARY:A Unified Wire Approach to the HPC Data Center UID:572 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20090402T130000 DTEND;TZID=America/Chicago:20090402T140000 DTSTAMP:20130525T020110 LOCATION:Ryerson room 251, University of Chicago DESCRIPTION:Advanced techniques in biomedical computing, imaging,and visualization are already changing the face of biology and medicine in both research and clinical practice. These techniques have the potential to provide comprehensive models and views of the human body in unprecedented depth and detail. As a result, biomedical computing and visualization will help produce exciting new biomedical scientific discoveries and clinical treatments. In this talk, I will discuss the state of the art in biomedical computing, medical imaging, and visualization research and present examples of their vital roles in cardiology, neuroscience, neurosurgery, and radiology. SUMMARY:Computing the Future of Biomedicine UID:573 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20090309T103000 DTEND;TZID=America/Chicago:20090309T113000 DTSTAMP:20130525T020110 LOCATION:Building 221 Conference Room A216, Argonne National Laboratory DESCRIPTION:Visualization has proven its value in scientific advances by helping scientists gain insight from their data and verify and correct scientific computations. As the amount of scientific data collected from sensors or simulations grows up to the order of petabytes, visualization of this large-scale data often requires high-performance distributed computing. However, it is typically challenging to develop or use visualization tools on a scalable distributed environment since they require a deep understanding of visualization algorithms, parallel processing, complex configurations and data handling. I envision the virtualization of a distributed visualization environment, which enables users to create their visualization as easily as they do on their desktop computers, while fully making use of underlying distributed visualization resource.\n\nThe individual components of a data visualization pipeline can be abstracted as: data retrieval, filtering/mining, rendering and display. My PhD work, the Scalable Adaptive Graphics Environment (SAGE) and Visualcasting, virtualizes the last component in the pipeline. With SAGE and Visualcasting, displays are totally virtualized from visualization applications. They just pass image buffers to SAGE. SAGE then scales the images to arbitrarily sized display walls ranging from a single desktop panel to a scalable array of LCD panels that are stitched together. Visualcasting extends this display virtualization by broadcasting SAGE image streams to multiple heterogeneous display clients. An analytical model of SAGE and Visualcasting was built and verified. This model will be extended to address major research questions in virtualizing the entire data visualization pipeline and then be used to investigate candidate approaches for the virtualization.\n\nThis research will improve the performance and usability of the visualization and analysis tools on ALCF\'s new visualization cluster EUREKA and motivate INCITE project investigators to leverage advanced visualization capabilities in their applications. Potentially this will change the paradigm of large-scale data visualization and expedite adoption of distributed visualization techniques to the broader DOE computational science community. SUMMARY:Virtualization of a Distributed Visualization Environment UID:574 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20090317T103000 DTEND;TZID=America/Chicago:20090317T113000 DTSTAMP:20130525T020110 LOCATION:Bldg: 221, Conference Room A216, Argonne National Laboratory DESCRIPTION:The Parallel Multi-block Adaptive Grid generation (PMAG) algorithm has been designed to handle a general multi-block topology. The algorithm is designed to adapt a multi-block grid concurrently with each block solved in an individual process. These processes can be run on a single parallel machine or distribution over a network of workstations. MPI is used as the message passing interface. Grid adaption in PMAG is based on the weight functions. These weight functions have demonstrated the capacity to detect shocks of differing strengths, primary and secondary vortices, and shear layers adequately. A simple tri-diagonal solver is used for generating the elliptic grids. PMAG is designed to allow boundary point movement by defining the boundaries as NURBS surfaces. This guarantees that the geometric definition is preserved accurately. PMAG is effectively demonstrated in application to realistic cases involving chemically reacting species and flow involving hypersonic flows and viscous boundary layers. SUMMARY:Parallel Multi-block Adaptive Grid Generation For Flow Calculations UID:576 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20090326T103000 DTEND;TZID=America/Chicago:20090326T000008 DTSTAMP:20130525T020110 LOCATION:Building 221, Room A216, Argonne National Laboratory DESCRIPTION:The talk will cover our experiences on performance analysis, tuning and implementation of (1) Parallel I/O* and (2) FFT algorithm optimizations** carried out for Blue Gene/L Supercomputer. \n\nIn order to provide sustainable Parallel I/O performance, we designed and implemented highly scalable parallel file I/O architecture for the Blue Gene/L system. Our architecture leveraged the benefit of the hierarchical and functional partitioning design of the system software with separate computational and I/O cores. Exploiting the scalability aspect of GPFS (General Parallel File System) at the backend and using MPI I/O as an application interface, the architecture was able to deliver at least one order of magnitude higher I/O bandwidth for a real application; i.e., for HOMME application we achieved an aggregate bandwidth of 1.8 GB/Sec and 2.3 GB/Sec for write and read accesses, respectively). The implementation also included the support of high-level parallel I/O data interfaces such as parallel HDF5 and parallel NetCDF scaling up to thousands of processors.\n\nTo enhance the 2D/3D FFT algorithm (as a part of HPC Challenge Benchmark Suite), we have exploited (1) single-node FFT performance, (2) all-to-all collective performance, and (3) overlap of computation and communication. Through effective exploitation of Blue Gene/L\'s double-FPU intrinsics, careful placement of all-to-all operations and synchronizations to maximize the interleave of communications and computations, substantial performance enhancement was achieved; i.e., a highly scalable FFT implementation with 20% performance improvement over the FFTW baseline on the LLNL Blue Gene/L system.\n\n__________________\n *Joint work with ANL (R. B. Ross, R. Thakur, R. Latham, W. D. Gropp) and H. Yu, C. Hawson, J. Moreira, T Engelsiepen from IBM Research.\n**Joint work with J. Gunnels, Y. Shabharwal, R. Garg from IBM Research. SUMMARY:Experiences on Performance Enhancement of Parallel I/O and FFT on Blue Gene/L Supercomputer UID:577 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20090331T103000 DTEND;TZID=America/Chicago:20090331T000008 DTSTAMP:20130525T020110 LOCATION:Building 360, Conference Room L119, Argonne National Laboratory DESCRIPTION:Review of several applications of large-scale parallel computing, with focus on parallelism in the calculations and how it was exploited. From computational science, topics include: Particle-in-cell simulation of tokamak plasma microturbulence, groundwater flow contaminant transport, and multimaterial hydrodynamics. From computational finance, the topic is valuation of mortgage-backed securities. SUMMARY:Plasmas, Fluids, and Toxic Assets: Examples of Large-Scale Parallel Computing UID:583 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20090407T103000 DTEND;TZID=America/Chicago:20090407T113000 DTSTAMP:20130525T020110 LOCATION:Building 360 / Conference Room L119, Argonne National Laboratory DESCRIPTION:Seminar SUMMARY:Architecting FLASH - A Complex Multiphysics Application Code that Scales from Laptops to Largest Supercomputers UID:586 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20090407T103000 DTEND;TZID=America/Chicago:20090407T113000 DTSTAMP:20130525T020110 LOCATION:Bldg: 221, Conference Room A216, Argonne National Laboratory DESCRIPTION:Urgent and emergency computing systems provide scientists with tools and capabilities to quickly allocate high-performance computing resources for time-critical applications. The usage and availability of storage and network resources in urgent computing environments can negatively impact the execution of data intensive emergency computing applications. In this talk, we present our work on supporting time-critical applications dependent on storage and network resources. We analyze the usage of current urgent computing systems and evaluate the data requirements for a potential emergency computing application. We present several data management policies and robust resource allocation techniques for urgent computing environments. Using the previously analyzed application and environment data, we evaluate these policies and techniques. SUMMARY:Data Management for Urgent and Emergency Computing Environments UID:587 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20090409T103000 DTEND;TZID=America/Chicago:20090409T120000 DTSTAMP:20130525T020110 LOCATION:Building 221 Conference Room A216, Argonne National Laboratory DESCRIPTION:The impact of commodity multi-core processors on the general computing community has been extensive. One can no longer assume that an application will run faster on successive generations of processors unless the application has been parallelized to take advantage of the increasing core count. The parallel computing community has also been adversely impacted by multi-core processors. While the increase in computation power and density from multi-core processors is encouraging, the majority of scientific parallel computing applications depend as much on memory subsystem performance as on compute performance. To make the problem worse, the MPI model, upon which nearly all scalable parallel applications are based, exacerbates the limited memory bandwidth available to a processor.\nWe have developed an operating system page table mapping strategy called SMARTMAP that allows processes on a multi-core processor to directly access each other\'s memory through simple virtual address bit manipulation. The SMARTMAP capability allows the cooperating parallel processes on a compute node to run independently as separate address spaces, but also provides the ability for the processes to act as threads running in a single address space. When used to implement MPI, SMARTMAP eliminates all extraneous memory-to-memory copies imposed by UNIX-based shared memory strategies, significantly reducing pressure on the memory subsystem for intra-node data transfers. In addition, SMARTMAP can easily support operations that UNIX-based shared memory cannot, such as direct, in-place, threaded MPI reduction operations and one-sided get/put operations.\nThis talk will describe the implementation of SMARTMAP in the Catamount lightweight kernel that runs on the Cray XT-based Red Storm platform at Sandia National Labs. We will show performance results comparing a SMARTMAP-enabled MPI to traditional UNIX-based shared memory approaches for MPI. We will also briefly describe several related ongoing research projects, including our next-generation open-source lightweight kernel and proposed extensions to MPI for multi-core processors being pursued by Sandia and Oak Ridge National Lab under the auspices of the DOE Institute for Advanced Architectures and Algorithms.\nBio:\nRon Brightwell received his BS in mathematics in 1991 and his MS in computer science in 1994 from Mississippi State University. He joined Sandia National Laboratories in 1995 and is currently a Principal Member of Technical Staff. While at Sandia, he has designed and developed software for lightweight compute node operating systems and high-performance networks on several large-scale massively parallel systems, including the Intel Paragon and TeraFLOPS, and the Cray T3 and XT series of machines. His research interests include high-performance, scalable communication interfaces and protocols for system area networks, operating systems for massively parallel processing machines, and parallel program performance analysis libraries and tools.\n SUMMARY:Enhanced Operating System Support for MPI on Multi-Core Processors UID:589 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20090414T110000 DTEND;TZID=America/Chicago:20090414T120000 DTSTAMP:20130525T020110 LOCATION:Building 221, room A-261, Argonne National Laboratory DESCRIPTION:Many vital structures in complex biological systems are formed through the self-organization of initially disordered, discrete elements into coherent bodies. During cell division, for example, microtubules undergo microscopic interactions with molecular motors to form macroscopic structures including the cytoskeleton of daughter cells and the mitotic spindle which properly segregates chromosomes. Understanding how to predict, control and orchestrate self-assembly processes on a precise biomolecular level is essential to the development of new classes of biosensors and bio-mimetic devices.\n\nWe model the motor-mediated self-organization of microtubules using a mean-field theory that captures the inherent stochastic nature of motor-filament interactions. Our model reveals that for a sufficiently large motor density, the filament network experiences an ordering transition toward a polar state. In our work we explicitly take into account the specifics of motor dynamics such as force-velocity relations and force-dependent detachment rates. We show that the transition to the oriented state is both continuous and discontinuous when force-dependent motor detachment becomes important. This result predicts an ordering hysteresis for experiments on alignment dynamics in semi-dilute and dense solutions of biological filaments.\n\n SUMMARY:Motor-Mediated Self-Organization of Microtubules in Active Filament Solutions UID:590 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20090416T150000 DTEND;TZID=America/Chicago:20090416T160000 DTSTAMP:20130525T020110 LOCATION:Building 221 Conference Room A216, Argonne National Laboratory DESCRIPTION:A chemical supply chain is an integrated network of facilities and transportation options for the procurement of raw materials, transformation of raw materials into intermediate and final products, and distribution of the final products to customers. Due to the increasing pressure to remain competitive in the global marketplace, it has become one of the major goals for most chemical companies to ensure optimal design and operation of not only the chemical production processes but also the entire chemical supply chains. This involves optimizing the network and process design, capacity and production planning, distribution and allocation planning, detailed scheduling and inventory control of a chemical supply chain to reduce the overall cost, and to maximize the profits, responsiveness and customer satisfaction. Furthermore, the supply, manufacturing, and distribution activities of a chemical supply chain have to deal with many uncertainties, such as demands, process yields, prices, breakdowns and natural disasters. Explicitly considering these uncertainties in the design and operation will add more complexity to the optimization models and consequently lead to greater algorithmic challenges. We are interested in the development of mathematical and computational tools that address the following areas: 1) Modeling of the design, planning and scheduling problems for chemical supply chains; 2) Multi-scale optimization to coordinate decision-making across geographically distributed locations and across time horizons spanning from days to years; 3) Optimization under uncertainty to account for stochastic variations and to manage the risks; 4) Algorithms and decomposition methods to support the three previous points. \n\nIn this talk, we will focus on three examples about chemical supply chain optimization under uncertainty. The first example is about the design, planning and scheduling of chemical supply chains under responsive criterion and economic criterion with the presence of demand uncertainty. A bi-criterion mixed-integer nonlinear programming (MINLP) model is developed to take into account multiple tradeoffs and to simultaneously predict the optimal locations of manufacturing sites and distribution centers, process technology, production profiles, detailed schedules, and inventory levels under different specifications of supply chain responsiveness. The second one addresses the design of chemical supply chains with multi-echelon inventory under uncertainty while taking into account risk-pooling effect. The steady-state network design and transportation decisions are integrated well with the stochastic inventory decisions by using an MINLP model, of which the large-scale instances are solved effectively by a tailored global optimization algorithm. The last example is a general computational framework for global chemical supply chain planning under uncertainty. The model formulation, computational strategies and simulation method will be discussed. Real-world industrial applications with up to 12,000 uncertain parameters are investigated to illustrate the economic benefits of considering uncertainties. A few other related projects will also be discussed along with these three examples. We will conclude this talk with some future extensions of these works to address the problems in energy & sustainability areas. SUMMARY:Optimization Models and Algorithms for Chemical Supply Chain Design and Operation under Uncertainty UID:591 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20090415T103000 DTEND;TZID=America/Chicago:20090415T113000 DTSTAMP:20130525T020110 LOCATION:Building 221 Conference Room A216, Argonne National Laboratory DESCRIPTION:Threading Building Blocks (TBB) is an open source (GPL) C++ template library which provides portable support for shared memory task parallelism. It works on Linux, Windows, and Solaris as well as more exotic systems (e.g. Xbox). Its aim is to support scalable parallelism by letting you focus on the tasks which make up your problem rather than the threads which happen to execute it. This presentation will provide a short introduction to TBB and give a flavor of how it can be used. SUMMARY:An Introduction to Threading Building Blocks UID:594 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20090427T164500 DTEND;TZID=America/Chicago:20090427T174500 DTSTAMP:20130525T020110 LOCATION:Charles M. Harper Center, 5807 S. Woodlawn, Room 1, University of Chicago DESCRIPTION:Please join us for University of Chicago\'s Chicago colloquium series: Energy in the 21st Century. Sponsored by the James Franck Institute, the U. Chicago Energy Initiative, and the Computation Institute SUMMARY:Issues and Priorities for Energy UID:600 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20090424T103000 DTEND;TZID=America/Chicago:20090424T113000 DTSTAMP:20130525T020110 LOCATION:Building 221 / Conference Room A216, Argonne National Laboratory DESCRIPTION:Seminar SUMMARY:Scheduling and Synchroniztion for Multiprocessors UID:598 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20090423T103000 DTEND;TZID=America/Chicago:20090423T113000 DTSTAMP:20130525T020110 LOCATION:Building 221 Conference Room A216, Argonne National Laboratory DESCRIPTION:Behemoth petaflop computing systems will propel atmospheric simulation science to unprecedented scales; yet individual node speed remains key for time-critical applications such as real-time forecasting or climate prediction. Even very large simulations able to exploit weak scaling will be limited by cost-performance in terms of energy and dollars consumed. For these reasons, new generations of multi- and many-core processors being mass produced for commercial IT and \"graphical computing\" (video games) are being scrutinized for their ability to exploit fine-grain parallelism abundant in atmospheric models. This presentation will describe work to identify and characterize expensive computational kernels from WRF, and community weather model, in terms of computational intensity, data parallelism, bandwidth requirements and memory footprint with the aim of accelerating weather and climate model using these new processors. SUMMARY:GPU Acceleration for Weather and Climate UID:599 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20090520T150000 DTEND;TZID=America/Chicago:20090520T160000 DTSTAMP:20130525T020110 LOCATION:Building 221, room A-261, Argonne National Laboratory DESCRIPTION:We discuss some graphs coloring problems that are related to the efficient evaluation of sparse derivative matrices. In particular, we consider the problems of finding optimal acyclic and star colorings, which model two different methods for the evaluation of Hessians. Both of these problems are known to be intractable even in severely restricted cases. We present a formula that describes the acyclic and star chromatic numbers of graphs that are decomposable with respect to the join operation, which builds a new graph from a collection of two or more disjoint graphs by adding all possible edges between them. We also show that our results lead to linear time algorithms for finding optimal acyclic and star colorings of cographs, which have the unique property that they are recursively decomposable with respect to the join and disjoint union operations.\n SUMMARY:Acyclic and Star Colorings of Joins of Graphs and an Algorithm for Cographs UID:601 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20090504T130000 DTEND;TZID=America/Chicago:20090504T140000 DTSTAMP:20130525T020110 LOCATION:Building 221 Confernece Room A261, Argonne National Laboratory DESCRIPTION:Bio: James is pursuing a PhD in Computer Science at The Ohio State University\nunder the advisement of Prof. Sadayappan. His work focuses on programming\nmodels and runtime systems for large-scale parallel computing.\n\nAbstract: Unified Parallel C (UPC) is a language extension to the popular C programming language that adds support for parallel programming with distributed, shared data. UPC targets both shared and distributed memory systems and presents the programmer with a logical partitioned global address space that is physically distributed across the processors and is potentially distributed across compute nodes in distributed memory systems or different memory banks in systems with nonuniform memory hierarchies. Under UPC, every data element has affinity to a distinct processor and UPC exposes this data locality information to the programmer so that it can be leveraged to enhance\nperformance. UPC facilitates access to the global address space through built-in language support as well as through explicit one-sided communication operations that allow processors to access any data in the global address space\nregardless of where it is stored.\n\nIn this tutorial we will cover introductory through advanced topics in the UPC programming language as well as runtime and performance considerations when targeting UPC\napplications at large-scale systems. SUMMARY:An Introduction to the Unified Parallel C Parallel Programming Language UID:602 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20090527T150000 DTEND;TZID=America/Chicago:20090527T160000 DTSTAMP:20130525T020110 LOCATION:Building 221, room A-261, Argonne National Laboratory DESCRIPTION:Empirical performance tuning has been emerging as an attractive means of tuning performance for increasingly complex computer architectures and application programs. We have been developing technologies to automate empirical tuning process of leadership class scientific applications. In this talk, I will describe our recent experience of tuning a scientific application called Nek5000 based on empirical performance tuning.\n\n SUMMARY:Empirical Performance Tuning UID:604 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20090515T133000 DTEND;TZID=America/Chicago:20090515T143000 DTSTAMP:20130525T020110 LOCATION:Building 221 Conference Room A216, Argonne National Laboratory DESCRIPTION:Stochastic mixed integer program (SMIP) is considered as one of the most important and challenging problems in operations research and computer science. Such models require strategic discrete decisions to be made without full knowledge of the future, followed by tactical actions after information regarding the future is revealed. This is a natural setting for decision-making under uncertainty with applications arising from telecommunication network design, finance, homeland security, energy systems and many more.\n\nIn this research, we establish and enhance the scalability of stochastic mixed integer programming by first discussing several enhanced cut-generation methods to accelerate the computational performance of the decomposition-based branch-and-cut (D2-BAC) algorithm to tackle very large-scale SMIPs. We also explore the advantages of parallelism over serial processing in solving SMIPs by constructing portable parallel implementations. Moreover, we develop a coupled branch-and-bound algorithm to accommodate a broader class of SMIPs with continuous first-stage variables, and prove its finite convergence. Finally, in collaboration with AT&T, we propose a stochastic programming model to deliver a robust network design for the next-generation IP-over-Optical networks, as part of the DARPA CORONET project. Customized L-Shaped methods are developed for solving some large-scale practical network instances. SUMMARY:Decomposition Methods for Two-Stage Stochastic Mixed-Integer Programming: Algorithms, Applications, and Computations UID:605 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20090520T133000 DTEND;TZID=America/Chicago:20090520T143000 DTSTAMP:20130525T020110 LOCATION:Building 221 Conference Room A216, Argonne National Laboratory DESCRIPTION:In many stochastic optimization applications, there is some uncertainty about the probability distribution P of random parameters. In this talk, we study the minimax decision model for two-stage stochastic linear optimization problems\nwith the assumption that P belongs to a class of probability distributions specified by the first and second moments. We also incorporate risk considerations into the model with piecewise linear disutility functions. We show that the model is tractable for problems with random objective and some special instances of problems with random right-hand side, which are in general NP-hard. We are able to provide explicit constructions of the worst-case\nextremal distributions for the minimax problems in these cases. We then demonstrate and compare the performance of minimax solutions with that of data-driven solutions under contaminated distributions using numerical examples. Applications include a production-transportation problem and a single facility minimax distance problem. \n\nComputational results show that the minimax solutions clearly hedge against the worst-case distributions and provide lower variability in objective value than data-driven solutions under most of the contaminated distributions. Finally, we present an additional application of the proposed minimax model in providing moment bounds for an option pricing problem. SUMMARY:Models for Minimax Stochastic Linear Optimization Problems with Risk Aversion UID:606 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20090521T140000 DTEND;TZID=America/Chicago:20090521T150000 DTSTAMP:20130525T020110 LOCATION:Building 221, Room A216, Argonne National Laboratory DESCRIPTION:A multiscale theoretical and computational methodology will be presented for studying biomolecular systems across multiple length and time scales. The approach provides a systematic connection between all-atom molecular dynamics, coarse-grained modeling, and mesoscopic phenomena. At the heart of the approach is the multiscale coarse-graining method for rigorously deriving coarse-grained models from the underlying molecular-scale interactions. Applications of the multiscale approach will be given for membranes and proteins, although the overall methodology is applicable to many other complex condensed matter systems. Recent applications to large protein complexes will also be described. The computational challenges and opportunities for this area of molecular modeling will be especially emphasized. SUMMARY:Systematic Multiscale Modeling of Biomolecular System UID:607 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20090610T150000 DTEND;TZID=America/Chicago:20090610T160000 DTSTAMP:20130525T020110 LOCATION:Building 221, room A-261, Argonne National Laboratory DESCRIPTION:This seminar is to discuss wind power forecasting and its use in power system operations. The presentation is structured into two parts. The first part concentrates on surveying existing wind power forecasting methodologies and identifying strengths and limitations of different approaches. The second part of the presentation addresses how power system operators can incorporate advanced wind forecasting technologies into their operations. Improved unit commitment and dispatch algorithms to address variability and uncertainty of wind power will be discussed. A numerical example will be shown to illustrate one feasible unit commitment and dispatch solution to integrate variable wind energy into power system operations.\n\n \n SUMMARY:Wind forecasting and integration into power system operations UID:608 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20090528T103000 DTEND;TZID=America/Chicago:20090528T113000 DTSTAMP:20130525T020110 LOCATION:Building 221 / Conference Room A216, Argonne National Laboratory DESCRIPTION:Vehicles reentering the atmosphere travel at very high velocities, causing extremely high temperature gas flows. The protection of the vehicle from this extreme high-energy environment is critical. A commonly used thermal protection system is an ablative heat shield, as used (for example) in the Apollo program and as planned for NASA\'s new Orion vehicle. The physical phenomena involved include strong shocks, aerothermochemistry, thermal non-equilibrium, thermal radiation, turbulence and the response of complex materials. Reliable computational models of this system would be of great value in design and operation of reentry vehicles. However, the models of some of these phenomena are known to be unreliable (e.g. turbulence) and others are difficult to parametrize (e.g. aerochemistry). The complexity of the phenomena, the nature of the models and the difficulty of experiments in this high-energy system make verification and validation of computational models challenging.\n\nThe Center for Predictive Engineering and Computational Sciences (PECOS) at the University Texas is taking on the challenge of verification and validation in the context of reentry vehicle simulations. In this talk, these challenges are discussed, along with some of our strategies being pursued to address them.\n SUMMARY:Validation and Verification Challenges in the Modeling of Atmospheric Reentry Vehicles UID:609 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20090601T140000 DTEND;TZID=America/Chicago:20090601T150000 DTSTAMP:20130525T020110 LOCATION:Room A134, Bdg. 221, Argonne National Laboratory, RI405, 5640 S. Ellis Ave., University of Chicago DESCRIPTION:Abstract: Many computer models, including climate prediction models such as C-goldstein and economic models such as E3MG, take many hours, or even weeks, to execute. This type of model can have tens to hundreds of free (adjustable) parameters, each of which is only approximately known. Under the Bayesian view, the true value of the code output is a random variable, drawn from a distribution that is conditioned by our prior knowledge, and in this case by the previous code runs; the computer code is thus viewed as a random function.\n\nIn this informal talk, I introduce the BACCO suite of software and show how it can be used to generate statistical inferences about such random functions. The software may be used to furnish computationally cheap—yet statistically rigorous—estimates of the computer code output.\n\nThe talk concludes with some examples of a cutting-edge type of analysis (the \'calibrator\') that can be used to assimilate field data into computer models. In particular, the posterior PDF for the model\'s parameter space can be calculated using very flexible and realistic statistical assumptions.\n\n SUMMARY:Bayesian Analysis of Computer Code Output: Something for Nothing UID:614 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20090701T150000 DTEND;TZID=America/Chicago:20090701T160000 DTSTAMP:20130525T020110 LOCATION:Building 221 Conference Room A216, Argonne National Laboratory DESCRIPTION:Energy systems modeling: Guenter Conzelmann and Tom Veselka will provide an overview of the different energy systems modeling activities at Argonne\'s Decision and Information Sciences (DIS) division. The presentation will particularly highlight new modeling opportunities and challenges for DOE in the areas of renewables and commercial buildings\nsimulation.\n SUMMARY:Hydopower and Energy Market Modeling UID:615 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20090715T150000 DTEND;TZID=America/Chicago:20090715T160000 DTSTAMP:20130525T020110 LOCATION:Building 221 Conference Room A261, Argonne National Laboratory DESCRIPTION:We establish results for the problem of tracking a time-moving manifold arising in on-line nonlinear programming by casting this as a generalized equation. We demonstrate that if points along a solution manifold are consistently strongly regular, it is possible to track the manifold approximately by solving a linear complementarity problem (LCP) at each time step. We derive sufficient conditions that guarantee that the tracking error remains bounded to second order with the size of the time step, even if the LCP is solved only to first order accuracy. We make use of these results to derive a fast augmented Lagrangean tracking algorithm and demonstrate the developments through a numerical case study. SUMMARY:Generalized Equation Concepts for On-Line Nonlinear Programming UID:616 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20090602T093000 DTEND;TZID=America/Chicago:20090602T000008 DTSTAMP:20130525T020110 LOCATION:Building 221 Conference Room A216, Argonne National Laboratory DESCRIPTION:Andrew will introduce the new Argonne Leadership Computing Facility (ALCF), its mission, its current systems and the systems expected to arrive at Argonne this fall. The ALCF ( http://www.alcf.anl.gov/) is a national leadership computing facility designed to provide resources that make computationally intensive projects of the largest scales possible. \n\nCurrently the ALCF supports 9 INCITE computational science projects covering disciplines as diverse as protein folding to modeling jet engines. SUMMARY:Overview of the Argonne Leadership Computing Facility UID:617 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20090602T103000 DTEND;TZID=America/Chicago:20090602T113000 DTSTAMP:20130525T020110 LOCATION:Building 221 Conference Room A216, Argonne National Laboratory DESCRIPTION:Scanned volumetric scalar datasets (as from structural MR or CT) have been the focus of much work in visualization and image analysis. Research in volume visualization has successfully blurred the distinction between methods that merely present pictures of volume data (pure \"visualization\"), versus methods that generate spatial information about material regions and boundaries (\"classification\" or \"segmentation\"). The same underlying information about local differential structure, as captured by the image gradient and Hessian, as well as quantities derived from these, play a role in setting the opacity functions that determine visibility in traditional volume rendering, as well as in image analysis methods for extracting geometric models of mathematically defined image features. I will describe work I\'ve done in this topic, ranging from semi-automatic generation of transfer functions, to recent work on using particle systems to sample image features. The recent particle system work may be especially useful for applications of microCT imaging. SUMMARY:Visualizing and Analyzing Local Differential Structure in Volume Data UID:618 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20080511T080000 DTEND;TZID=America/Chicago:20080511T090000 DTSTAMP:20130525T020110 LOCATION:Building 221 Conference Room A216, test DESCRIPTION:test SUMMARY:test UID:619 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20090611T133000 DTEND;TZID=America/Chicago:20090611T143000 DTSTAMP:20130525T020110 LOCATION:Building 221 / Conference Room A216, Argonne National Labortory DESCRIPTION:The advent of the Petascale era, provides a great opportunity as well as a great challenge for computational science and engineering. Large scale scientific applications need to scale to unprecedented numbers of processing cores and adapt to multi-core architectures with complex memory and network hierarchies in order to fully leverage the computational resources available. In addition to possible human errors, the limit for code writing and the ever-growing complexity of many scientific codes make the development of parallel scientific applications an intimidating task. In addressing these issues, we are developing a generic collaborative problem-solving environment from mathematical abstractions for automating the development of highly scalable and efficient codes that can solve a wide range of scientific problems. In such an environment, application developers, either software engineers or domain experts, can contribute to a code with their expertise at their maximum scale, thus enhance the overall programming productivity and speed up scientific discoveries. SUMMARY:Automating the Development of Parallel Multidisciplinary Scientific Applications UID:620 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20090708T150000 DTEND;TZID=America/Chicago:20090708T160000 DTSTAMP:20130525T020110 LOCATION:Building 221 Conference Room A261, Argonne National Laboratory DESCRIPTION:Mixed Integer Nonlinear Programming (MINLP) is a broad class optimization problems that are encountered in many different applications. The presence of both nonconvex functions and variables that are constrained to be integers makes the problem difficult at many levels. In this talk, we describe MINOTAUR, a framework that is under development for solving such problems. SUMMARY:Solving MINLPs using MINOTAUR: Description, Goals and the story so UID:622 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20090730T150000 DTEND;TZID=America/Chicago:20090730T160000 DTSTAMP:20130525T020110 LOCATION:Building 221, room A-261, Argonne National Laboratory DESCRIPTION:Many recent phenomena of high interest in condensed matter physics are linked to the existence of short-ranged correlations.  The most prominent examples are provided by transition metal oxides, where complex disorder results in a balance of the spin, orbital, charge and strain degrees of freedom and give rise to competing ground states with incompatible order and exotic phenomena such as colossal magnetoresistance, negative thermal expansion, quantum spin liquids, and high temperature superconductivity.  Traditional crystallography provides well developed and sophisticated tools for obtaining the long-range order of crystalline solids, but only indirect and insufficient evidence of disorder.  While there are several techniques that are very sensitive to the existence of local disorder, none provide detailed information on the correlations between defects or the length scales of short-range ordering processes.  Single crystal diffuse scattering probes both the local distortions around point defects as well as the defect to defect correlations on length scales of 1 to 10nm and provides the most powerful tool for studying complex short range correlations, such as the formation of stripes, checkerboards, ladders, or phase separation.  However, there remain formidable difficulties in obtaining and analyzing large enough volumes of data with sufficient momentum and energy resolution required for accurate modeling.  I will discuss several systems that we are currently studying at MSD as well as some new techniques being developed to collect and analyze diffuse scattering. SUMMARY:Challenges of using Diffuse Scattering to Measure Short Range Correlations in Crystalline Materials UID:623 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20090618T133000 DTEND;TZID=America/Chicago:20090618T143000 DTSTAMP:20130525T020110 LOCATION:Building 221 Conference Room A216, Argonne National Laboratory DESCRIPTION:HMesher leverages the strengths of its parent, the p-mesher from PolyFEM: relatively coarse mesh while preserving geometric integrity, conforming meshes in complex CAD assemblies, mesh-geometry associativity, and generation of a hex-dominant mesh from the tetrahedral mesh. The HMesher (Hybrid Mesher) offers a meshing solution for an h-type FE, and it is integrated currently in CATIA V5, as part of LMS VirtualLab Meshing SUMMARY:HMesher: Mesh Generation for CAD Assemblies UID:624 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20090916T150000 DTEND;TZID=America/Chicago:20090916T160000 DTSTAMP:20130525T020110 LOCATION:Building 203 Auditorium, Argonne National Laboratory DESCRIPTION:Trajectories generated by classical molecular dynamics (MD) are best\ninterpreted in a statistical sense due to the weak definition of\ninitial conditions and chaotic nature of the underlying equations of\nmotion. For this reason, accuracy and efficiency of a time integration\nscheme should be measured with respect to statistical averages, rather\nthan deviations from an ``exact trajectory\'\'. A practical MD time\nintegrator must be both stable and statistically accurate, allowing for\nits use over long time intervals with large stepsizes. In this talk, I\nwill survey some results from backward error analysis for geometric\nintegrators and show how (under certain assumptions) these results can\nbe applied to understanding the statistical properties of integrators\nfor MD. SUMMARY:Geometric Integrators for Classical Molecular Dynamics: Theory and Application UID:625 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20090930T080000 DTEND;TZID=America/Chicago:20090930T170000 DTSTAMP:20130525T020110 LOCATION:JW Marriott Starr Pass Resort, Tucson, Arizona DESCRIPTION:Participation is encouraged in the Grace Hopper Celebration of Women in Computing Conference. The CELS directorate, along with the CIS and HR divisions, is a Bronze Sponsor at this event. The conference is designed to bring the research and career interests of women in computing to the forefront. It is the largest technical conference for women in computing and results in collaborative proposals, networking and mentoring for junior women and increased visibility for the contributions of women in computing. Conference presenters are leaders in their respective fields, representing industry, academia and government. Top researchers present their work while special sessions focus on the role of women in today’s technology fields. SUMMARY:Grace Hopper Celebration of Women in Computing UID:626 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20090623T150000 DTEND;TZID=America/Chicago:20090623T160000 DTSTAMP:20130525T020110 LOCATION:Building 221 / Conference Room A261, Argonne National Labortory DESCRIPTION:The runtime environment and parallel execution of the ab initio electronic structure package GAMESS is handled by the distributed data interface (DDI). In addition, DDI provides application developers with a virtual shared memory environment built around the model of a parallel global address space that is portable to all major hardware classes. DDI has proved to be a robust implementation and currently supports more than twelve different types of distributed data quantum chemistry application in GAMESS. A decade on and we are planning for the forthcoming era of petascale computing resources that pose new challenges for DDI. In this seminar, I will describe the implementation and use of DDI in GAMESS and work underway to enable DDI to support petascale quantum chemistry applications. This work is support by NSF grant \"Enabling Petascale Applications in the Chemical Sciences\" (Award 0749156). SUMMARY:Enabling Petascale Quantum Chemistry Using the Distributed Data Interface in GAMESS UID:627 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20090602T100000 DTEND;TZID=America/Chicago:20090602T170000 DTSTAMP:20130525T020110 LOCATION:Building 360, rm. L119 or online, Argonne National Laboratory DESCRIPTION:Over one billion processing hours are available through DOE’s INCITE program for 2010, which is jointly managed by the Argonne and Oak Ridge Leadership Computing Facilities. Our Proposal Writing webinar can help you stake your claim. Sign up today – proposals are due July 1, 2009!\n\nJoin Katherine Riley, ALCF scientific applications engineer, and Bronson Messer of Oak Ridge’s Scientific Computing group, as they provide tips and suggestions to improve the quality of your INCITE proposal submission. SUMMARY:INCITE Proposal Writing Webinar UID:628 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20090722T140000 DTEND;TZID=America/Chicago:20090722T164500 DTSTAMP:20130525T020110 LOCATION:Building 221 Conference Room A216, Argonne National Laboratory DESCRIPTION:Wes Kendall\nTitle: Terascale Analysis of Variable Interactions in MODIS\nAbstract:\nThe temporal and spatial dynamics of climatic variables are highly dependent on each other, and it is important to be able to identify multivariate interactions when assessing problems such as drought and fire damage. As climatic datasets grow beyond the terascale, previous serial methods for performing these types of analyses are infeasible to application scientists. To enhance scientific discovery at this magnitude, we developed a parallel framework that closely integrates parallel I/O techniques with an intuitive interface for issuing Boolean range queries on scientific data. Our techniques allowed us to visualize various interactions between two variables in over a terabyte of satellite data from NASA\'s Moderate Resolution Imaging Spectroradiometer (MODIS) project. We applied our methods on the entire dataset with scalability up to 16,000 cores and end-to-end execution times consistently around one minute.\n\nAlexe Mihai\nTitle: Automatic differentiation of the nuclear reactor simulation code MATWS\nAbstract: \nA tangent linear model is built for the FORTRAN 77 nuclear reactor simulation code MATWS using the automatic differentiation tool ADIFOR. The sensitivities of a few variables of interest are computed, and then validated against finite differences. The resulting derivative information will be used for uncertainty quantification and sensitivity analysis. We also describe the a priori code modifications required by ADIFOR, and the post-processing of the differentiated code.\n\nLuis de la Torre\nTitle: Carbon Emissions Trading with a Computable General Equilibrium Model\nAbstract: \nEmissions trading markets seek to reduce pollution levels by setting a cap on total emissions and allowing producers to trade pollution permits. The American Clean Energy and Security Act, passed in the House on June 26, 2009, proposes emissions trading markets for greenhouse gases including carbon dioxide. We use a complementarity problem to model a multi-factor, multi-region economy in which a foreign producer has market power to affect prices, in order to compare policy alternatives between carbon emissions trading markets and production and use taxes. We also discuss a safety valve system, a combination of tax and emissions trading in which emissions permit prices are bounded.\n\n\nJulio Goez\nTitle: Algorithm for the estimation of Computational Noise \nAbstract: \nEstimating the computational noise present in the evaluation of a function f could play an important role in the development of convergence criteria for optimization algorithms. There are different reasons for the presence of noise in the evaluation of a function, for example the use of iterative methods in the computation, or single precision operations. In this presentation we will describe an algorithm for estimating stochastic computational noise. Based on our tests, this algorithm can determine the noise level of a nonlinear function with fewer than 10 evaluations (independent of the dimension), with more evaluations providing more precise estimates. \n\nKathy King\nTitle: Optimizing Public Health Emergency Logistics with Dynamic Programming\nAbstract: \nIn the event of a public health emergency like pandemic flu or a large-scale bioterrorist attack, it is important that vaccines, antibiotics, or other medical countermeasures be distributed quickly to minimize mortality and morbidity. Federal, state, and local authorities plan to work together to rapidly distribute the necessary supplies from stockpiles to the affected population. We use a stochastic dynamic programming model of the state and local level supply chain to determine an optimal strategy for allocating limited staff, resources, and medical supplies. We also present an approximate algorithm for solving the dynamic program with realistically large data sets.\n\nChungen Shen\nTitle: A Non-monotone Filter Method for Nonlinear Optimization\nAbstract: \nWe propose a non-monotone filter sequential quadratical programming (SQP) algorithm in which the l and g-filters are introduced. The g-filter being a traditional monotone filter guarantees the global convergence. The l-filter is a non-monotone filter that allows more flexibility for accepting a trial point. Under standard conditions,we prove the super-linear local convergence without second order correction (SOC) steps.\n\nAndy Terrel\nTitle: FEM automation of Oldroyd-B fluids\nAbstract:\nOver the past several years the FEniCS projects have developed many advances in the automation of finite element codes. These techniques allow the researcher to study numerous models and numerical discretizations quite rapidly. With this technology, we study the numerous viscoelastic models and discretizations presented by Baaijens. This allows us to systematically address many stability questions with quick comparisons between numerous test problems. We present both the abstractions for the FEM automation as well as the comparisons for the different Oldroyd-B type models.\n\nKrystal Richards and Seneca Gibson\nTitle: Experiences On Empirical Performance Tuning of Scientific Applications - S3D and MADNESS\nAbstract:\nEmpirical performance tuning has been emerging as a popular and attractive means of dealing with very complex computer architectures and application programs. Two important scientific application packages, S3D (numerical simulations of turbulent combustion) and MADNESS (simulation framework for solving quantum chemical problems) are installed and tested on target platform (AMD Quad-Core Phenom) for empirical performance tuning. Tasks involving preparation, processing, and methods will be discussed. Some experimental results in profiling, identifying kernels, and tuning the codes will be presented. \n SUMMARY:SASSy : Student Argonne Summer Symposium UID:630 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20090527T080000 DTEND;TZID=America/Chicago:20090527T120000 DTSTAMP:20130525T020110 LOCATION:Argonne Guest House, Argonne National Laboratory DESCRIPTION:Ready to run your project on 40 racks of the Blue Gene/P? Then come Leap to Petascale at the Argonne Leadership Computing Facility (ALCF). We\'ll tell you more about the ALCF and the petascale resources here. Then, ALCF performance engineers will help you scale and tune your applications on 40 racks of Blue Gene/P. This is an especially great opportunity for anyone considering applying for a 2010 INCITE award (Proposals for INCITE are due July 1, 2009). SUMMARY:Leap to Petascale Workshop UID:631 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20090701T103000 DTEND;TZID=America/Chicago:20090701T120000 DTSTAMP:20130525T020110 LOCATION:Building 221 Conference Room A216, Argonne National Laboratory DESCRIPTION:MPI is the industry standard for distributed memory parallel programming and Unified Parallel C (UPC) is a relative newcomer that promises high productivity with comparable performance. In comparison with MPI\'s\ntwo-sided messaging model, UPC provides a partitioned global address space programming model that allows for one-sided access to distributed, shared data.\n\nHybrid programming models that mix MPI with shared memory parallelism through threads or OpenMP have been successful at improving performance through data locality and reducing data replication. In this presentation we will explore a new hybrid programming model that mixes MPI with UPC, drawing on the strengths of each model. We will discuss\nruntime system issues such as ensuring mutual progress on communication and MPMD-style launching of hybrid jobs as well as algorithmic and performance implications of the hybrid model.\n\nBio:\nJames is pursuing a PhD in Computer Science at The Ohio State University under the advisement of Prof. Sadayappan. His work at OSU focuses on dynamic load balancing and scalable runtime systems to support task parallelism. His research interests include parallel programming models, fault tolerance, high performance computing applications, and runtime systems.\n SUMMARY:Hybrid Parallel Programming with MPI and UPC UID:632 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20090812T150000 DTEND;TZID=America/Chicago:20090812T160000 DTSTAMP:20130525T020110 LOCATION:Building 221 Conference Room A261, Argonne National Laboratory DESCRIPTION:Given a graph, a simple strategy for measuring the closeness of vertices is as follows. We initially associate to each vertex a random number, then update the value of each vertex by the average of the values of its neighbors, and perform this update a few times. The closeness between a pair of vertices is indicated by the difference of the two associated values. This closeness is inspired by the concepts used in Bootstrap Algebraic Multigrid in which it allows a better interpolation of low-residual errors.\nIn this talk, I will unravel the mystery of this simple process, explain the intuition behind it, and analyze its convergence properties. This measurement has been successfully used for solving (or improving the existing algorithms for) a number of combinatorial problems, including graph partitioning, maximum matching, minimum p-sum, etc. I will present the results in these scenarios, which show the superior usefulness of this closeness concept. \n\n SUMMARY:Measuring the Connection Strengths between Graph Vertices UID:633 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20090707T103000 DTEND;TZID=America/Chicago:20090707T113000 DTSTAMP:20130525T020110 LOCATION:Building 221, Room A216, Argonne National Laboratory DESCRIPTION:In this talk, I will describe our recent work towards building a lightweight Software Persistent Memory (SoftPM) infrastructure and the new capabilities that it enables. Fundamentally, SoftPM eliminates the duality of data management in applications, whereby memory-resident data is readily accessible but volatile, and storage-resident data is persistent, yet not directly accessible to the process. SoftPM allows applications to allocate persistent memory in much the same way volatile memory is allocated, and easily restore, browse, and interact with past versions of persistent memory state. This simplifies the implementation of three broad capabilities required in a variety of applications -- recoverability (e.g., checkpoint-restart), record-replay (e.g., scientific data visualization), and execution branching (e.g., simulation model-space exploration). I will discuss research challenges, our approach, and some promising preliminary results.\n\nIn the latter part of the talk, I will give a brief overview of other ongoing systems work in our lab. SUMMARY:Software Persistent Memory and Its Applications UID:634 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20090701T150000 DTEND;TZID=America/Chicago:20090701T160000 DTSTAMP:20130525T020110 LOCATION:Building 360 / Conference Room L119, Argonne National Labortory DESCRIPTION:We have developed a program package GPAW [1] for density functional calculations (DFT) using real-space grids together with the projector augmented wave method. In addition to the basic ground state properties, excited state properties such as optical absorption spectra can be calculated with the time-dependent density functional theory (TD-DFT) where we have implemented both linear response and real-time propagation formulations of the theory.\n\nGPAW is well suited for massively parallel calculations, and we will present current parallelization strategies as well as examples about the parallel scalability. We will present also some examples about the applicability of GPAW \nboth in DFT and TD-DFT calculations.\n\n[1] wiki.fysik.dtu.dk/gpaw\n\n SUMMARY:GPAW: Efficient Tool for Real-Space, Real-Time Electronic Structure Calculations UID:635 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20090630T133000 DTEND;TZID=America/Chicago:20090630T143000 DTSTAMP:20130525T020110 LOCATION:Building 221 Conference Room A216, Argonne National Laboratory DESCRIPTION:The work on an arithmetic analogue (i.e., in mixed characteristic) of a certain geometric object, called an affine Grassmannian, that are common over equal-characteristic local fields. A finite-dimensional Grassmannian variety is the parameter space of subspaces of a given vector space; the points of this space (corresponding to subspaces) form a continuous family. Affine Grassmannian is a generalization of this when the given vector space is defined over a local field (think of the field of quotients of formal power series), and that the subspaces are replaced by certain lattices, that is, maximal-rank submodules over the ring of integers of the local filed.\n\nWhen the local field is in mixed characteristic, such parameter space was constructed and studied by W. Haboush. It is a union of finite dimensional projective varieties over the residue field, each of which has singularities. The results that are obtained is that this singular variety is still normal (i.e. singularities are \"not bad\") and locally complete intersection, making it the next best thing to have, after a nonsingular variety.\n\nTerminologies appearing above, in terms of examples will be mentioned. Some ideas will be going into constructing this parameter space, and if time permits, relevance to other branches of mathematics such as number theory, representation theory or physics will be mentioned. SUMMARY:Results on the Parameter Space of Certain Lattices in Mixed Characteristic UID:636 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20090629T103000 DTEND;TZID=America/Chicago:20090629T113000 DTSTAMP:20130525T020110 LOCATION:Building 221 Conference Room A216, Argonne National Laboratory DESCRIPTION:In fluid mechanics two ways can be used to describe the behavior of a flow field. One is the Lagrangian method which describes the time rate of change in the flow properties along the traces of identifiable flow parcels. The other is the Eulerian method which observes the properties of flow at a fixed position over time. \n\nIn this talk, I will discuss the analogies of Lagrangian and Eulerian methods in visualization of time-varying scalar data. I will briefly overview some traditional visualization methods for time-varying data, which I will argue are mostly Lagrangian. Then I will describe our recent research in analyzing time-varying scientific data with Eulerian approaches. The new approaches allow us to obtain more accurate spatio-temporal information about the features in time-varying data sets. I will discuss the essential components of our analysis environment including multiscale data representations, distance metrics and classification schemes, and a method for multivariate data visualization. Our goal is to complement the existing visualization techniques, and provide the scientists with better quantitative data analysis tools. SUMMARY:Visualizing Time-Varying Data using Lagrangian and Eulerian Approaches UID:637 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20090702T140000 DTEND;TZID=America/Chicago:20090702T150000 DTSTAMP:20130525T020110 LOCATION:Building 202, Conference Room B169, Argonne National Laboratory DESCRIPTION:For the last five years a network of research institutions in Europe and the US (including Argonne) have collaborated on the assembly of a minimal protocell - a self-replication system with tightly coupled catalytic cooperation among \"genes\", metabolism, and container. In this talk I will report scientific progress and challenges for this work as well as discuss a vision for a \"living technology\" in part based on our ability to assemble artificial \"living\" materials. Experimentally, we have recently demonstrated this coupling by having an informational molecule (8-oxoguanine) catalytically control the light driven metabolic (Ru-bpy based) production of container materials (fatty acids). This is a significant milestone towards assembling a minimal self-replicating molecular machine. Work in progress for our container associated gene-replication system will also be presented. \n\nCoupling is needed between the gene-metabolism-container system and the container associated gene-replication system to form a functional protocell. I present a variety of physics based simulation (molecular dynamics, dissipative particle dynamics and reaction kinetics) of the coupled protocell components and its life-cycle, and they expose a number of anticipated systemic challenges associated with the remaining experimental implementation of the protocell. Finally I will outline how simple self-replicating materials could be useful as part of \"living technology\", a central component of the nano-bio-info-cogno (NBIC) knowledge convergence, and sketch how the protocell work might provide new clues to the origin of life.\n\n SUMMARY:Assembly of a Minimal Protocell UID:638 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20090701T133000 DTEND;TZID=America/Chicago:20090701T143000 DTSTAMP:20130525T020110 LOCATION:Building 221 Conference Room A216, Argonne National Laboratory DESCRIPTION:The first component of this work is a parallel algorithm for constructing non-uniform octree meshes for finite element computations. Prior to octree meshing, the linear octree data structure must be constructed and a constraint known as 2:1 balancing\" must be enforced; parallel algo-\nrithms for these two subproblems will also be presented. \n\nThe second component of this work is a parallel matrix-free geometric multigrid algorithm for solving elliptic partial dierential equations (PDEs) using these octree meshes. The last component of this work is a parallel multiscale Gauss\nNewton optimization algorithm for solving the elastic image registration problem. The registration problem is discretized using nite elements on octree meshes and the parallel geometric multigrid algorithm is used as a preconditioner in the Conjugate Gradient (CG) algorithm to solve the linear system of equations formed in each Gauss Newton iteration.\n\nThe parallel octree meshing and multigrid algorithms have several physical and computer science applications such as in solid/fluid mechanics, heat/mass transfer, electromagnetism, image processing and unstructured mesh generation. Potential applications for the image registration algorithm include automatic identication of abnormalities in medical images, motion reconstruction from temporal sequences of images and planning of surgeries.\nSeveral ideas were used to reduce the overhead for constructing the octree meshes. These include (a) a way to lower communication costs by reducing the number of synchronizations and reducing the communication message size, (b) a way to reduce the number of searches required to\nbuild element-to-vertex mappings, and (c) a compression scheme to reduce the memory footprint of the entire data structure. To our knowledge, the multigrid algorithm presented in this work is the only matrix-free multiplicative geometric multigrid implementation for solving nite element equations on octree meshes using thousands of processors. The proposed registration algorithm is also unique; it is a combination of many different ideas: adaptivity, parallelism, fast optimization\nalgorithms, and fast linear solvers.\n\nAll the algorithms were implemented in C++ using the Message Passing Interface (MPI) standard and were built on top of the PETSc library from Argonne National Laboratory. The multigrid implementation has been released as an open source software: Dendro. Dendro has been tested on several NSF TeraGrid platforms. Our largest run was a highly nonuniform, 8- billion-unknown, elasticity calculation on 32,000 processors. SUMMARY:Parallel Octrees, Multigrid and Elastic Image Registration UID:639 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20091130T103000 DTEND;TZID=America/Chicago:20091130T000008 DTSTAMP:20130525T020110 LOCATION:Building 240, Room 1406 and 1407, Argonne National Laboratory DESCRIPTION:Modern Museums and Planetariums house state of the art theater spaces that also serve as visualization facilities. Optical star projectors have been replaced with full-dome digital projection systems, in some cases exceeding 50 megapixels. Stereoscopic, high-definition theaters are also commonplace. Museum exhibits increasingly incorporate state of the art displays and human interaction devices. This talk will give an overview of some the visualization activities taking place at the Adler Planetarium. These include the Space Visualization Laboratory, a working visualization laboratory on the floor of the museum and the MoonWall – a tiled-display serving images from the Lunar Reconnaissance Orbiter. We will also describe programs designed to reach beyond our traditional audiences - collaborations with symphonies across the world, and an exhibit at O’Hare international airport. Finally upcoming projects, the renovation of the historic Sky Theater and a new permanent modern cosmology gallery, will be discussed. SUMMARY:AstroViz at the Adler Planetarium UID:747 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20090713T103000 DTEND;TZID=America/Chicago:20090713T113000 DTSTAMP:20130525T020110 LOCATION:Building 221 Conference Room A216, Argonne National Laboratory DESCRIPTION:Major progress in magnetic fusion research has led to ITER - a multi-billion dollar burning plasma experiment supported by seven governments (EU, Japan, US, China, Korea, Russia, and India) representing over half of the world\'s population. Currently under construction in Cadarache, France, it is designed to produce 500 million Watts of heat from fusion reactions for over 400 seconds with gain exceeding 10 - thereby demonstrating the scientific and technical feasibility of magnetic fusion energy. Strong research and development programs are needed to harvest the scientific information from ITER to help design a future demonstration power plant with a gain of 25. Advanced computations at the petascale and beyond in tandem with experiment and theory are essential for acquiring the scientific understanding needed to develop whole device integrated predictive models with high physics fidelity. This is the primary motivation for the Fusion Simulation Program (FSP) - a new DOE-SC initiative supported by the Offices of Fusion Energy Science and Advanced Scientific Computing Research -- that is entering the project definition phase. Since ITER and leadership class computing are prominent missions of the DOE today, producing such a world-leading predictive capability for fusion represents a key exascale-relevant strategic project for the future. SUMMARY:Fusion Simulation Program (FSP): Simulations at the Petascale and Beyond for Fusion Energy Sciences UID:641 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20090805T150000 DTEND;TZID=America/Chicago:20090805T160000 DTSTAMP:20130525T020110 LOCATION:Building 221 Conference Room A261, Argonne National Laboratory DESCRIPTION:Performance experiments can involve multiple execution runs where parameters such as execution platform, measurement tools, methods of measurement, application parameters, and analysis techniques can vary. In order to manage the layers of complexity involved in experimental setup, execution, and post-analysis, a degree of abstraction and automation is necessary at each stage. A layer of abstraction is needed to hide the intricacies involved in experimental set-up and runs for varying sets of experimental parameters. We present an integrated component-based environment that automates the process of running multiple performance experiments and parameter selection of parallel scientific applications. Our toolkit will enable application scientists to easily modify the experimental parameters over multiple execution runs and to selectively retrieve the data for analysis and generation of performance models. SUMMARY:An Automated Component-Based Performance Experimentation Environment UID:642 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20090721T110000 DTEND;TZID=America/Chicago:20090721T000008 DTSTAMP:20130525T020110 LOCATION:Research Institute, Room 480, University of Chicago DESCRIPTION:The profusion of high-throughput instruments and the explosion of new results in the scientific literature, particularly in molecular biomedicine, is both a blessing and a curse to the bench researcher. Even knowledgeable and experienced scientists can benefit from computational tools that help navigate this vast and rapidly evolving terrain. However, effective design and implementation of computational tools that genuinely facilitate the generation of novel and significant scientific insights remains poorly understood. In this talk, I will describe a set of efforts that combines natural language processing for information extraction, graphical network models for semantic data integration, and some novel user interface approaches into a system that has recently played a pivotal role in making a significant biomedical discovery. SUMMARY:3R Systems for Biomedical Discovery Acceleration UID:643 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20091028T150000 DTEND;TZID=America/Chicago:20091028T160000 DTSTAMP:20130525T020110 LOCATION:1416, Conference Center, Bldg 240, Argonne National Laboratory DESCRIPTION:This talk is set at a basic level. Differentiation and Integration are \noften the earliest and possibly the best understood topics in elementary calculus. So it may come as a surprise to some that serious problems may be encountered in programming Numerical Quadrature and Numerical Differentiation.\n\nAmong the topics treated are the performance profile of the automatic\nquadrature routine; a software interface problem; round-off error amplification in numerical differentiation; and, at the very end, possibilities for using complex variable analysis.\n\nThese lead to generally conceptual problems. The questions which arise can be understood with minimal mathematical sophistication. But a particular answer, if one exists, might be complicated. In this talk we concentrate on the questions. SUMMARY:The Area and the Slope; many questions; fewer answers. UID:644 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20090930T150000 DTEND;TZID=America/Chicago:20090930T160000 DTSTAMP:20130525T020110 LOCATION:TCS Building, Argonne National Laboratory DESCRIPTION:TBA SUMMARY:Asynchronous Dynamic Load Balancing UID:647 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20090923T150000 DTEND;TZID=America/Chicago:20090923T160000 DTSTAMP:20130525T020110 LOCATION:TBA, Argonne National Laboratory DESCRIPTION:Ion channels are irresistible objects for biological study because they are the [nano] ‘valves of life’ controlling an enormous range of biological function, much as transistors control computers. Ion channels are much easier to simulate than many proteins because conformation changes are not involved in channel function, once the channel is open. Open channels are interesting objects for chemical study because they effectively select among chemically similar ions, under unfavorable circumstances. Channels are interesting objects for physical study because they contain an enormous density of charge, fixed, mobile, and induced. Direct simulation of channel behavior in atomic detail is difficult if not impossible. Macroscopic electric fields and concentration gradients produce substantial flows which are the natural biological function of the channel, making equilibrium analysis unhelpful. Multiscale issues are nontrivial: simulations must deal with concentrations of 10^(−7) to 55 M of different chemical species. Ion transit takes ~ 10^(−8) sec compared to a calculation time step of 10^(−16) sec and a biological time scale of 10^(−4) — 1 sec.\n\nComputations in full three dimensions are not yet feasible. In reduced dimensionality, however, a simple physical model does surprisingly well. One model with the same two parameters accounts for qualitatively different selectivity of both calcium and sodium channels in a wide range conditions. The model does not involve any traditional chemical binding energies at all. The binding free energy is an output of the calculation, produced by the crowding of charged spheres in a very small space.\nHow can such a simple model give such specific results when crystallographic wisdom and chemical intuition says that selectivity depends on the precise structural relation of ions and side chains? The answer is that structure is the computed consequence of the forces in this model and is very important indeed, but as an output of the model, not as an input. The relationship of ions and side chains vary with ionic solution and are different in different channels and solutions. Selectivity is a consequence of the ‘induced fit’ of side chains to ions and vice versa.\n\nThe simplified model (probably) works because the structures in both the model and the real channel are self-organized and at their free energy minimum, forming different structures in different conditions. A variational approach is immediately suggested by these results and one is well under way, applying the methods of Chun Liu (Associate Director, IMA Minnesota) and colleagues, perfected in electro-rheology.\n\nPractical exploitation to design selective biomimetic membranes for desalination and detoxification depends on computation. Trial and error methods are not likely to be efficient enough to allow practical design. Computational success depends on the accurate (± 3%) estimation of both energy and entropy of the self-organized structure in three dimensions of space and one of time.\n\nComputational design of these biomimetric systems is an unmet challenge that can soon be met because of the memory bandwidth and size of next generation computers.\n\n\n SUMMARY:Self-organized selectivity in Calcium and Sodium Channels: Biomimetic Designs now ready for Serious Computation UID:648 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20090909T150000 DTEND;TZID=America/Chicago:20090909T160000 DTSTAMP:20130525T020110 LOCATION:Building 221 Conference Room A261, Argonne National Laboratory DESCRIPTION:We study the refractive index sensitive transmission of a 3D\nplasmonic crystal, which can exhibit strong optical responses near certain wavelengths and are used in chemical sensing applications. Our goal is to design a a square array of subwavelength cylindrical nanowells in a polymer conformally coated with a gold film that maximizes the optical response.\n\nEvery choice of design parameters requires the solution of a 3D finite-difference time-domain simulations that runs for 12 hours of wall-clock time on a 125-nopde linux cluster. We reveal \"How the Optimization Was Won\" using low-tech tools such as email, matlab, and AMPL.\n\nGovernment health warning: \n-------------------------\nThis talk is content-free! It does not contain any enlightening or educational material. Attending this talk may seriously harm your ability to carry out world-class research!\n\nJoint work with Stephen Gray (CNM), Joanna Maria (UIUC), and\nBoyana Norris (MCS). SUMMARY:Optimization By Email of 3D Plasmonic Crystal Structures UID:650 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20090819T150000 DTEND;TZID=America/Chicago:20090819T160000 DTSTAMP:20130525T020110 LOCATION:Building 221 Conference Room A261, Argonne National Laboratory DESCRIPTION: SUMMARY:The XM Travel System and How it Applies to LANS UID:651 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20090826T103000 DTEND;TZID=America/Chicago:20090826T113000 DTSTAMP:20130525T020110 LOCATION:Building 360 / Conference Room L119, Argonne National Labortory DESCRIPTION:Programmers and application developers using HPC systems require simple-to-use tools to determine how efficiently their codes are executing on a target machine, and for analysis and eventual improvement of their application’s performance. \n\nMany factors contribute to an application’s performance on HPC systems. Extracting meaningful runtime performance data, managing very large performance data volumes, extracting useful information from them, interpreting that runtime information to identify and locate those factors, which most affect performance, are all tasks difficult for a programmer to manage. Tools to extract and record system events indicative of execution performance have been available for some time. What has been lacking is a means to manage, reduce complexity, and extract hidden execution performance problems. \n\nA performance tool set, from a recent system vendor, based on existing performance packages like PAPI cross-platform hardware performance counter interface, developed at the University of Tennessee, TAU (Tuning Analysis and Utilities) performance tool suite, developed at the University of Oregon, and others, is used to illustrate runtime performance inefficiencies in example codes.\n SUMMARY:HPC Tools for Performance Profiling and Analysis on Some Code Examples UID:654 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20091007T150000 DTEND;TZID=America/Chicago:20091007T160000 DTSTAMP:20130525T020110 LOCATION:TCS Building, Argonne National Laboratory DESCRIPTION:Parallel File Systems today are comprised of multiple software components combined to provide concurrent and efficient access across hundreds or possibly thousands of storage devices. Using traditional programming models to develop these systems leads to large code bases, full of code that is difficult to follow, debug, and build on. In this presentation, I will describe a programming model based on concurrent state machines that has been useful in the implementation of the Parallel Virtual File System (PVFS), enabling concise specification of logical control flow. I will also describe a new approach that attempts to address a number of the lessons learned in PVFS. SUMMARY:Event Driven Programming Models for Parallel Storage Systems UID:655 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20091020T080000 DTEND;TZID=America/Chicago:20091020T050000 DTSTAMP:20130525T020110 LOCATION:UIC Forum, Chicago DESCRIPTION: Dramatic growth in data and equally rapid decline in the cost of highly integrated clusters has spurred the emergence of the data center as the platform of choice for a growing class of data-intensive applications. To encourage conversations between those developing applications, algorithms, software, and hardware for such \"cloud\" platforms, we are convening the second workshop on Cloud Computing and Its Applications (CCA\'09).\n\nThis workshop will provide two days of invited talks on cloud computing, data intensive scalable computing, and related topics.\n\nTopics of interest include:\n\n\nOrganizing Committee\n\nIan Foster\nArgonne National Laboratory\nUniversity of Chicago\n\nRobert Grossman\nUniversity of Illinois at Chicago\nOpen Data Group\n\nDouglas Thain\nUniversity of Notre Dame\n\nPhilip Yu\nUniversity of Illinois at Chicago\n\nKate Keahey\nArgonne National Laboratory\nUniversity of Chicago\n\nYunhong Gu\nUniversity of Illinois at Chicago SUMMARY:CCA09: Cloud Computing and Its Applications UID:656 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20090903T150000 DTEND;TZID=America/Chicago:20090903T160000 DTSTAMP:20130525T020110 LOCATION:Building 221 Conference Room A216, Argonne National Laboratory DESCRIPTION:In our work, we consider a class of complex interaction-transport systems of atmospheric chemistry in the context of SVD-based model reduction. Many tasks of simulation, optimization and control can be performed more efficiently, if the intermediate complexity of the chemical model is reduced. We use an SVD-based approach (\"method of snapshots\") to extract information from a set of full model observations and project the model equations onto a reduced order space so that the full dynamics are preserved with only a moderate error. \n\nWe examine and improve many features of the method. In particular, we show how to measure sensitivities of the model reduction process, and use the results to select the placement and weighting of observations to best reproduce specific events in the full model behavior. We develop techniques for reduced space basis selection that allow us to take into account multiple events. We show how to construct reduced models to replace the full model in iterative parameter optimization procedures so that fewer steps and lower computational cost is needed for the search to converge. \n\nThe overall result of our study is a more complete understanding of how to perform factor analysis, simulation and optimization of nonlinear models using model reduction tools. SUMMARY:Model Reduction for Simulation and Optimization in Chemical Kinetics UID:657 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20090910T120000 DTEND;TZID=America/Chicago:20090910T130000 DTSTAMP:20130525T020110 LOCATION:Searle Lab, room 240, University of Chicago DESCRIPTION:The effective use of data is key to advances in almost all \ndisciplines. Data-intensive research is emerging as a new paradigm. Researchers in many disciplines have opportunities for significant advances as a result of the pervasive growth in digital data, communication and devices. There are many challenges in enabling researchers to become adept in this new and fast-changing context.\n\nThe talk will report progress towards understanding how to economically enable a large community of researchers to become fluent in whatever uses of data will benefit their research. We recognize that there are many different data-related activities and that there are many variations in the nature, maturity, structure and scale of the data. We propose a characterisation of clusters of similar requirements in the search for commonalities that may help provision and engagement. The talk will be illustrated with examples from a range of disciplines and is part of \na fact-finding tour of US research centers.\n\n[pdf] SUMMARY:Exploiting and Providing Research Data: Finding Strategies to Help Researchers UID:658 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20090922T110000 DTEND;TZID=America/Chicago:20090922T120000 DTSTAMP:20130525T020110 LOCATION:Research Institue, Room 480, University of Chicago DESCRIPTION:The CI\'s \"Disciplinary Deep Dive\" (3-D) program aims to identify opportunities for collaborative research across the University and Argonne.\n\nA 3-D features an appropriate mix of workshops, lectures, and informal discussions. We welcome ideas for topics and volunteers to organize the meetings. SUMMARY:Explanatory Combinatorial Dictionary in Natural Language Processing UID:661 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20091202T150000 DTEND;TZID=America/Chicago:20091202T160000 DTSTAMP:20130525T020110 LOCATION:Building 240 (TCS), Argonne National Laboratory DESCRIPTION:We develop a technique for a joint inversion of two different kinds\nof tomographic data. The \"objects\" to be recovered are connected\njust through a topological constraint based on the following weak\nassumption linking the two considered data sets: if any\ninhomogeneity is present in the investigated volume, it modifies the\nvalues of both the two recorded physical properties. In comparison\nwith the classical case, where the inversions are performed\nindependently on each data set and then assembled a\nposteriori, the additional a priori information that we use\nis the spatial agreement (shape and position) of the anomalies in\nthe two reconstructed models. The constraint is imposed by\nintroducing, within the Tikhonov regularization framework, a\nstabilizing \"joint functional\" which couples the two types of data\nto be reconstructed. \n SUMMARY:A novel approach to the joint inversion of loosely connected data UID:666 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20091124T150000 DTEND;TZID=America/Chicago:20091124T160000 DTSTAMP:20130525T020110 LOCATION:1404&1405, Conference Center, Bldg 240, Argonne National Laboratory DESCRIPTION:Quantum chemistry is the field of computational science\nwhich attempts to elucidate molecular processes by solving the\nmany-body electronic Schroedinger equation. The computational\nchallenges associated with this task are quite different from\nmainstream supercomputing. Codes like NWChem make very little use of\nMPI and are not frequently used on BlueGene platforms. I will explain\nwhy this is presently true by describing the subset of quantum\nchemistry algorithms resulting from coupled-cluster theory. Unique\nscientific opportunities in quantum chemistry which follow from MPI-3\nand BlueGene/Q conclude this talk. SUMMARY:Quantum chemistry for computer scientists UID:667 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20091104T150000 DTEND;TZID=America/Chicago:20091104T160000 DTSTAMP:20130525T020110 LOCATION:TCS Building, 1404&1405, Conference Center, Bldg 240 DESCRIPTION:Scientists and engineers must understand results from experiment and simulation generated data to gain insights and perform knowledge discovery. However, the potential of large-scale systems for analytic capabilities may be difficult to achieve because of limited I/O, file system performance, and a lack of appropriate interfaces for data analysis. In this talk, I will present an active storage for scalable data analysis that enables end-to-end optimizations needed for performance and productivity gains when utilizing petascale systems. Specifically, I will present our design of an active storage node that will allow data analysis, mining, and statistical operations to be executed from within the parallel I/O runtime systems. I will then present the design of a parallel I/O runtime interface that will utilize customized active storage nodes to perform I/O operations. Lastly, I will present our experimental results using a set of data analysis kernels running on our active storage prototype. SUMMARY:Scalable Data Analysis and Active Storage UID:668 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20091014T150000 DTEND;TZID=America/Chicago:20091014T160000 DTSTAMP:20130525T020110 LOCATION:TBA, Argonne National Laboratory DESCRIPTION:PerfSuite is an open source package for application software performance\nanalysis on Linux-based systems developed at the National Center for\nSupercomputing Applications (NCSA). Since its initial public release in 2003,\nPerfSuite has gained a worldwide user base due to its emphasis on usability and flexibility. In addition to routine use for application measurement, tuning, and benchmarking in production environments, PerfSuite has been extended and deployed in novel ways both at NCSA and through collaborations with other research groups. Through the ongoing NSF SDCI (Software Development for Cyberinfrastructure) program, PerfSuite is currently involved in the POINT (Productivity from Open, Integrated, Tools) project that brings together the PerfSuite, TAU, PAPI, and Scalasca software packages with the overall goal of improving the integration and interoperability of these components.\n\nThis talk will provide an overview of PerfSuite, beginning with the motivation\nfor and history of its development. Past and current collaborative activities\nin which PerfSuite has been involved will be discussed, and finally an overview\nof current directions and development plans for PerfSuite will be outlined,\nincluding recently-released support for the Java programming language. SUMMARY:The PerfSuite Toolset for Application Performance Analysis: Evolution, Status, and Future Directions UID:674 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20090914T143000 DTEND;TZID=America/Chicago:20090914T153000 DTSTAMP:20130525T020110 LOCATION:Building 240, Conference Room 2.C.1, Argonne National Laboratory DESCRIPTION:Researchers have long relied on shared resources and distributed computing,while industry and home users typically have utilized dedicated hardware and single-threaded programs. Recently though, advanced in networking, virtualization and middleware have have enabled many businesses and individual users to move much of their environment to network-based resources. Google, Amazon and others have made vast computing, storage and network available and lowered the cost of entry a few cents. We will examine the evolution of computing toward \"the cloud\" and explore some of the critical technologies. The cloud presents both opportunities and challenges for users of high performance computing (HPC) infrastructure. Rather needing to develop for a highly specialized supercomputing or grid environment, the cloud promises users a highly flexible environment with any resource dynamically available. However, the performance requirements of HPC software puts unique demands on the infrastructure that current cloud implementations cannot satisfy. We will talk about cloud infrastructures and how they might be customized for HPC, with special emphasis on networking. This seminar is intended to provide an opportunity for open discussion. SUMMARY:Clouds for HPC: Opportunities and Challenges in Compute, Storage and Networking UID:663 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20090921T103000 DTEND;TZID=America/Chicago:20090921T113000 DTSTAMP:20130525T020110 LOCATION:Auditorium in building 203, Argonne National Laboratory DESCRIPTION:In the first part of this talk, we will discuss challenges in dealing\nwith large graphs in general, including the task of visualizing all of\nthe sparse matrices in the University Florida Sparse Matrix\nCollection. While traditional graph visualization methods can be\ninvaluable in getting an overall sense out of large data sets, they\nare not as helpful in conveying the underlying structural information,\nclusters, and neighborhoods. In the second part of this talk, we\ndescribe an algorithm, GMap, for visualizing graphs as maps. GMap\novercomes some of the shortcomings with the help of the geographic map\nmetaphor. The effectiveness this algorithm is illustrated with\nvisualization examples from several domains, namely Netflix movies, TV\nshows, Amazon books, and last.fm music. Some results of this talk can\nbe found at http://www.research.att.com/~yifanhu/gallery.html SUMMARY:Visualization of large relational data sets and clusters as graphs and maps UID:664 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20090915T130000 DTEND;TZID=America/Chicago:20090915T170000 DTSTAMP:20130525T020110 LOCATION:Building 401, Room A1100, Argonne National Laboratory DESCRIPTION:One of the foundational issues in high-performance computing is the ability to move large (multi-gigabyte, and even terabyte) data sets between sites. Simple file transfer mechanisms such as FTP and SCP are not sufficient either from the reliability or the performance perspective. Globus implementation of GridFTP is the most widely used open source production quality data mover available today.\n\nIn the first half of this tutorial, the presenters will cover GridFTP server administration. They will walk through the steps required for setting up and configuring GridFTP server on Linux/Unix machines. In the second half, they will explain the steps required for installing the commonly used GridFTP client \'globus-url- copy.\' They will demonstrate the use of globus-url-copy and describe a set of best practices for obtaining maximal file transfer performance with GridFTP.\n\nParticipants are urged to bring a laptop with access to Linux machines to get hands-on experience. SUMMARY:Tutorial on GridFTP UID:665 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20091114T080000 DTEND;TZID=America/Chicago:20091114T170000 DTSTAMP:20130525T020110 LOCATION:Portland Convention Center, Portland, Oregon DESCRIPTION:SC09 will be the 22nd consecutive year of the SC Conference series – once again featuring an exceptional Technical Program, industry and research exhibits, Education Program and many other activities. For 2009, three new Technology Thrust areas have been added: Bio-Computing, Sustainability, and the 3D Internet.\n\nEstablished 21 years ago, the conference has built a diverse community of participants including researchers, scientists, computing center staff members, IT and data center management, application developers, computer manufacturing personnel, program managers, journalists and congressional staffers. This diversity is one of the conference\'s main strengths, making it a yearly \"must attend\" forum for stakeholders throughout the technical computing community. SUMMARY:SC09: Bio-Computing, Sustainability, 3D Internet UID:672 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20090928T143000 DTEND;TZID=America/Chicago:20090928T153000 DTSTAMP:20130525T020110 LOCATION:Building 240, Conference Room 3C1, Argonne National Laboratory DESCRIPTION:Complex Networks are pervasive in our society. Realistic biological, information, social and technical networks share a number of unique features that distinguish them from physical networks. Examples of such features include: irregularity, time-varying structure, heterogeneity among individual components and selfish/cooperative game-like behavior by individual components. Furthermore, the network structure, the dynamical process on the network and the behavior of constituent agents co-evolve over time. The size and heterogeneity of these networks, their co-evolving nature and the technical difficulties in applying dimension reduction techniques commonly used to analyze physical systems makes reasoning, prediction and controlling of these networks even more challenging. Recent quantitative changes in high performance and wireless computing capability have created new opportunities for collecting, integrating, analyzing and accessing information related to such large complex networks. The advances in network and information science that build on this new capability provide entirely new ways for reasoning and controlling these networks. Together, they enhance our ability to formulate, analyze and realize novel public policies pertaining to these complex networks. Over the last 15 years, our group has established a theory based program for modeling, simulation and associated decision support tools for understanding such large socio-technical systems. Complementing this modeling environment is a scalable service delivery framework that provides policy analysts and scientists seamless access to the modeling environment. After a brief overview, I will describe our approach within the context of a specific application: development of modeling and decision support environments to study epidemics in co-evolving social and wireless networks. SUMMARY:Building Virtual Cities: Policy Informatics for Large Co-evolving Socio-Technical Networks UID:670 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20091109T103000 DTEND;TZID=America/Chicago:20091109T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, TCS (1404, 1405, 1406), Argonne National Laboratory DESCRIPTION:Remy Evard, former Deputy Division Director of MCS and CIO of Argonne, has been working for Novartis Pharmaceuticals since early 2007. Remy is the CIO of the Novartis Institutes for BioMedical Research (NIBR), the global research organization of Novartis. NIBR\'s mission is to discover new therapies for patients in need, which it accomplishes by carrying out biomedical scientific research. Remy will compare and contrast the working environments, culture, and use of computing of Argonne and Novartis, considering different aspects such as for-profit vs. non-profit, global vs. local, corporate vs. DOE laboratory. He will also describe the basic pharmaceutical scientific pipeline, his work and focus at NIBR, the challenges in the industry, and some of the potential for collaboration and interaction. This will be an informal talk with plenty of opportunity for questions, answers, and discussion. SUMMARY:Working in the \'Real World\' UID:737 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100119T103000 DTEND;TZID=America/Chicago:20100119T113000 DTSTAMP:20130525T020110 LOCATION:Building 362 Auditorium, Argonne National Laboratory DESCRIPTION:Phillip Colella of Lawrence Berkeley National Laboratory will discuss \"Models, Algorithms, and Software: Tradeoffs in the Design of High-Performance Computational Simulations in Science and Engineering\" at a Director\'s Special Colloquium Tuesday, Jan. 19th.\n\nThe colloquium will begin at 10:30 a.m. in the Building 362 Auditorium. All employees whose schedules permit are encouraged to attend.\n\nColella is a senior staff scientist and leads the Applied Numerical Algorithms Group in Lawrence Berkeley National Laboratory\'s Computational Research Division.\n\nHis talk will describe the tradeoffs between the models, the discretizations, and the software in the development of high-performance computational simulations in science and engineering involving partial differential equations, including some motivating applications, and the combination of analysis and computational experiments that are used to explore the design space.\n\nMany important problems such as combustion, fusion, systems biology, and climate change, involve multiple physical processes operating on multiple space and time scales. In spite of the physical diversity of these problems, there is a great deal of coherence in the underlying mathematical representations. They are all described in terms of various versions of the elliptic, parabolic and hyperbolic partial differential equations of classical mathematical physics.\n\nColella\'s research interests have included numerical methods for partial differential equations. He has made contributions in high-resolution finite-difference methods, adaptive mesh refinement, volume-of-fluid methods for irregular boundaries, and programming language and library design for parallel scientific computing.\n\nHis honors and awards include the Institute of Electrical and Electronics Engineers Sidney Fernbach Award for high-performance computing in 1998, the Society for Industrial and Applied Mathematics/Association for Computing Machinery prize (with John Bell) for computational science and engineering in 2003, and election to the U.S. National Academy of Sciences in 2004. SUMMARY:Models, Algorithms, and Software: Tradeoffs in the Design of High-Performance Computational Simulations in Science and Engineering UID:751 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20090930T130000 DTEND;TZID=America/Chicago:20090930T140000 DTSTAMP:20130525T020110 LOCATION:Cafeteria Conf. Rm. B, Argonne National Labortory DESCRIPTION:Parallab is the unit for High Performance Computing at the University of Bergen and operates the universities supercomputer facilities and other e-infrastructures. The group conducts research and development in scientific computing, engineering, Grid- and other e-infrastructures.\n\nIn this presentation we introduce Parallab and its activities in scientific computing. Specialfocus will be on computational results in miscible displacement porous media flow. These processes can be described by nonlinear partial differential equations. We will address challenges in their numerical treatment and discuss software development as well as numerical solutions strategies. Secondly, we show how results from numerical bifurcation analysis can help to understand the structure of the nonlinear solutions. Finally, we apply statistical methods in order to derive practically relevant results from the nonlinear, instable dynamics. It turns out that large\ncompute- and storage infrastructures are required to efficiently obtain these types of results.\n\n SUMMARY:Porous Media Flow: Learning from Numerical Simulations UID:678 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20091002T094500 DTEND;TZID=America/Chicago:20091002T104500 DTSTAMP:20130525T020110 LOCATION:Bldg. 240, Conference Rom 6C2, Argonne National Laboratory DESCRIPTION:The recent advent of agent-based modeling (ABM) and the availability of software platforms for its implementation offer a powerful alternative to model the spatiotemporal behaviors of a fishery with the consideration of heterogeneity and interactivity. This paper describes a prototype agent-based fishery management model of Hawaii’s longline fishery. The model simulates the daily fishing activities of 120 Hawaii longline vessels of diverse characteristics. Following the strategy of pattern oriented modeling (POM), we use the spatiotemporal distribution pattern of fishing efforts to calibrate the model. While POM has a record of success in ecology, the present application to socioeconomic systems such as fishing and fishery management is almost unprecedented.\n\nWe also use the calibrated model to evaluate three alternative fishery regulatory policies in Hawaii’s longline fishery: 1) no regulation; 2) annual cap of 17 turtle interactions; and 3) close the north central area year round, with respect to their impacts on fishing productivity and by-catch of protected sea turtle. The prototype model, constructed using 1999 data, appears to be able to capture the responses of the fishery to these alternative regulations reasonably well, suggesting its potential as a management tool for policy evaluation in Hawaii’s longline fishe SUMMARY:A Prototype Agent Based Model for Policy Evaluation in Hawaii\'s Longline Fishery UID:687 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20091007T110000 DTEND;TZID=America/Chicago:20091007T120000 DTSTAMP:20130525T020110 LOCATION:Building 202, Room B169, Argonne National Laboratory DESCRIPTION:TBA SUMMARY:The Evolution of ADP-glucose Pyrophosphorylase Subunits UID:688 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20091002T103000 DTEND;TZID=America/Chicago:20091002T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, Room 2C2, Argonne National Laboratory DESCRIPTION:As large-scale numerical simulations permeate science and industry, the resulting deluge of data brings about an urgent need for effective analysis tools to help researchers turn this information into actual insight. Scientific visualization plays an important role in this process by creating a visual interface to massive dataset sets that affords an intuitive basis for interpretation,\nassessment, and decision making. However, the rapidly growing size and complexity of scientific datasets put an increasing emphasis on the ability of visualization methods to clearly convey a high-level picture of the data by characterizing its inherent structure across spatial and temporal scales. In this talk I will describe a general strategy built upon a principled mathematical framework to identify salient structures in vector and tensor fields, which are ubiquitous in practical scenarios. Our methodology combines concepts from topology and dynamical systems, differential geometry, and computer vision to extract important features from large-scale multivariate\ndatasets, thus producing a concise geometric signature that lends itself to automatic processing and insightful visual representations. I will illustrate this basic approach in the context of problems ranging from computational fluid dynamics and solid mechanics to fusion research and medical image analysis. I will also present our ongoing work on the application of a novel Lagrangian method to the structural analysis and interactive multi-scale exploration of transient flows. SUMMARY:Structural Analysis of Vector and Tensor Fields for Effective Visualization UID:689 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20091118T150000 DTEND;TZID=America/Chicago:20091118T160000 DTSTAMP:20130525T020110 LOCATION:1404&1405, Conference Center, Bldg 240, Argonne National Laboratory DESCRIPTION:This talk deals with two quadratic inverse eigenvalue problems that arise in mechanical vibration and structural dynamics. The first one, Quadratic Partial Eigenvalue Assignment Problem(QPEVAP), arises in controlling dangerous vibrations in mechanical structures, such as buildings, bridges, highways, automobiles, air and space crafts, and others. QPEVAP concerns with finding two feedback matrices such that a small amount of the eigenvalues of the associated quadratic eigenvalue problem are reassigned to suitably chosen ones while keeping the remaining large number of eigenvalues and eigenvectors unchanged. For robust and economic control design, these feed-back matrices must be found in such a way that they have the norms as small as possible and the condition number of the modified quadratic inverse problem is minimized. These considerations give rise to two nonlinear unconstrained optimization problems, known respectively, as the Robust Quadratic Partial Eigenvalue Assignment Problem (RQPEVAP) and Minimum Norm Quadratic Partial Eigenvalue Assignment Problem (MNQPEVAP). The other one, Finite Element Model Updating Problem (FEMUP) arising in the design and analysis of structural dynamics, refers to updating an analytical finite element model so that a set of measured eigenvalues and eigenvectors from a real-life structure are reproduced and the physical and structural properties of the original model are preserved. A properly updated model can be used in confidence for future designs and constructions. Another major application of FEMUP is the damage detections in structures. Solutions \nof FEMUP also give rise to several constrained nonlinear constrained optimization problems.\n\nWe will give an overview of the recent developments on computational methods for these difficult nonlinear optimization problems and discuss directions of future research. The talk is interdisciplinary in nature and will be of interests to mathematicians, computational and applied mathematicians, specialists in optimizations, and engineering researchers and practicing engineers. SUMMARY:Computational and optimization methods for quadratic inverse eigenvalue problems in finite element model updating UID:693 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20091005T133000 DTEND;TZID=America/Chicago:20091005T143000 DTSTAMP:20130525T020110 LOCATION:Building 240, Room 1406 and 1407, Argonne National Laboratory DESCRIPTION:Ever increasing research and development in science and engineering is based on I/O and data intensive simulations and/or analysis of observational data, requiring the use of high-performance computing (HPC) systems. Traditional\ninterfaces in file systems and storage systems are designed to handle the worst-case scenarios for conflicts, synchronization, locking, coherence checks, and other issues, which adversely affect the I/O performance. In\nmany cases, the problem is not that of having insufficient I/O capacity or bandwidth, but it is the excessive synchronization of I/O accesses at the I/O layer, generated by massively parallel applications.\n\nWe have designed and implemented an I/O delegate system that uses a subset of processes to carry out the I/O tasks for an application. By placing the I/O system close to the applications and allowing the applications to pass the high-level data access information, the I/O system has more opportunity to provide better performance. One of the most important features of the I/O delegate system is that it allows communication among delegates and enables their collaboration for further optimizations, such as collaborative caching, I/O aggregation, load balancing, and request alignment. We achieved I/O Bandwidth improvement percentages ranging from 25 % to 260 % by allocating 2-3 % additional compute nodes to be used as I/O delegate nodes.\n\nUsing I/O delegate system as the basic infrastructure, we have developed a method that assigns disjoint file regions to the I/O delegates such that, lock conflicts and overlapping accesses are resolved at the delegate system.\nFor example, Lustre file system has a server-based locking protocol, we configure I/O delegates to have one-to-one or one-to-many mapping to the I/O servers. This strategy of having persistent pairing between servers and clients reduces the number of lock acquisitions to only one, eliminates lock contention altogether, produces more effective data prefetching, and less cache coherence control overhead. By allocating only 1/8th of additional\ncompute nodes as I/O delegates, we achieved I/O Bandwidth improvement percentages ranging from 100 % to 10000% for applications using MPI Independent I/O operations. SUMMARY:Optimizing I/O for Large-Scale Scientific Applications using I/O Delegation UID:694 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20091009T110000 DTEND;TZID=America/Chicago:20091009T120000 DTSTAMP:20130525T020110 LOCATION:Building 203 Auditorium, Argonne National Laboratory DESCRIPTION:During the last several years there has been an increasing awareness that exascale (1018 operations/sec) computer systems may be feasible by the end of the next decade. This community awareness began through a series of town hall meetings held at Argonne, Berkeley and Oak Ridge in the spring of 2007, a series of DARPA sponsored workshops and study groups looking at challenges for exascale hardware and software and most recently a series of eight DOE workshops covering applications ranging from climate, to high-energy and nuclear physics to biology. DOE has launched a joint planning effort between the Office of Science and the National Nuclear Security laboratories to develop the science case and a roadmap that includes technology research and development, hardware platform development, programming models and software development and science code development. While it is generally assumed in the science community that there are many problems whose solution can be advanced by access to three orders of magnitude more computing capability than exists today, it is unclear at this point how many problems will actually be able to effectively utilize the systems that will be possible to build. In this talk I will describe the technical barriers (e.g. device power, optics, packaging, concurrency, etc.) to developing exascale systems, the current approaches to attack those barriers and the science that we hope will drive the venture. SUMMARY:Building the Science Case for Exascale Computing UID:695 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20091008T133000 DTEND;TZID=America/Chicago:20091008T143000 DTSTAMP:20130525T020110 LOCATION:Searle Lab, room 240, University of Chicago DESCRIPTION:The Petascale Active Data Store (PADS) project is a National Science Foundation (NSF) Major Research Infrastructure grant. Using the funds from NSF and the University of Chicago the project will acquire and operate a substantial data storage (~500 terabytes) and analysis system (9 teraflop/s) for the Computation Institute community. \n\nThis talk will provide an overview of the PADS project, including an update on the state of hardware and software. It will introduce the upcoming seminar series being held this fall and the class on data intensive computing to be held winter quarter. Finally it will open up for discussion the research issues associated with the project. SUMMARY:PADS Overview and Update UID:697 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20091015T133000 DTEND;TZID=America/Chicago:20091015T143000 DTSTAMP:20130525T020110 LOCATION:Searle Lab, room 240, University of Chicago DESCRIPTION:As HPC systems have approached petascale, the demands on storage and I/O throughput of HPC workloads have brought about highly developed storage systems with unprecedented performance. Likewise, data intensive computing has driven new approaches to storage architectures that meet the needs of those programming models. This seminar presents the Parallel Virtual File System, its role as a software component in the petascale storage architecture, and as a platform that supports development of a variety of storage models and architectures. SUMMARY:Maximizing throughput of Petascale Storage with PVFS UID:706 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20091209T150000 DTEND;TZID=America/Chicago:20091209T160000 DTSTAMP:20130525T020110 LOCATION:Building 240 (TCS) Conference Center, Argonne National Laboratory DESCRIPTION:The Swift parallel scripting language lets users apply parallel composition constructs to existing sequential or parallel programs to express highly parallel scripts.\n\nSwift scripts are flexible and portable, and can run efficiently on platforms ranging from multicore workstations to petascale supercomputers. For performing parameter sweeps and data analysis with exiting application programs, parallel scripting is typically easier and more productive than tightly-coupled parallel programming.\n\nThis talk will give an overview of Swift and how its used to run scientific applications in parallel on clusters, grids, clouds, and petascale systems. The architectural challenges of scripting on large-scale systems will be covered, and case studies will be presented.\n SUMMARY:Parallel Scripting with Swift UID:707 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100106T150000 DTEND;TZID=America/Chicago:20100106T160000 DTSTAMP:20130525T020110 LOCATION:Building 240 (TCS) Conference Center, Argonne National Laboratory DESCRIPTION:The cancer Biomedical Informatics Grid, or caBIG™, is a voluntary virtual informatics infrastructure that connects data, research tools, scientists, and organizations to leverage their combined strengths and expertise in an open federated environment with widely accepted standards and shared tools. The underlying service oriented infrastructure that supports caBIG™ is referred to as caGrid. Driven primarily by scientific use cases from the cancer research community caGrid provides the core enabling infrastructure necessary to compose the Grid of caBIG™. It provides the technology that enables collaborating institutions to share information and analytical resources efficiently and securely, and allows investigators to easily contribute to and leverage the resources of a national-scale, multi-institutional environment. The caBIG initiative is a four-year project, funded by the National Cancer Institute, with the mission of linking the more than 60 cancer centers across the U.S. into an integrated distributed-computing system. In this talk I will go over description of the caBIG program, and how the concepts of service oriented science are used in achieving the objective the caBIG program (finding a cure for cancer !) SUMMARY:caBIG: A Case Study of Service Oriented Science UID:708 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20091023T153000 DTEND;TZID=America/Chicago:20091023T163000 DTSTAMP:20130525T020110 LOCATION:TBA, Argonne National Laboratory DESCRIPTION:The optimal Jacobian accumulation problem is one of the original\ncombinatorial problems arising in automatic differentiation.\nThe vast majority of approaches to this problem have attempted to\nexploit its similarity to the problem of minimizing fill during LU\nfactorization of sparse, unsymmetric matrices.\nIn this talk, we take a complexity-theoretic approach to Jacobian accumulation.\nSpecifically, we explore relationships between this problem and others\nthat occur in the context of algebraic and Boolean complexity,\nas well as discuss the results that are implied by these relationships. SUMMARY:New Complexity Results for Jacobian Accumulation UID:709 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100217T150000 DTEND;TZID=America/Chicago:20100217T160000 DTSTAMP:20130525T020110 LOCATION:Building 240, Rm 1404-1405, Argonne National Laboratory DESCRIPTION:Multicore architecture has become the trend of high performance processors. While it is generally accepted that we have entered the multicore era, concerns exist on when or will moving into the manycore stage. Recently, Hill and Marty presented a pessimistic view of multicore scalability, citing Amdahl\'s law and the memory-wall problem. Technology is available, but major vendors are hesitant in making processors that have a large number of cores. This is a very interesting phenomenon, where history seems to repeat itself on the scalability debate of parallel processing that occurred 20 years ago. In this introductory talk we first review the history and concepts of scalable computing, and review the current technologies and the memory-wall problem. We then use the same hardware cost model of multicore chips used by Hill and Marty to introduce two performance models from the scalable computing point of view. These models show that there is no inherent, immovable upper bound on the scalability of multicore architectures. Finally, we conclude with proposed solutions to the memory-wall problem to make the potential scalability of multicore reachable in practice. SUMMARY:Reevaluating Amdahl\'s Law in the Multicore Era UID:790 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20091111T150000 DTEND;TZID=America/Chicago:20091111T160000 DTSTAMP:20130525T020110 LOCATION:1404&1405, Conference Center, Bldg 240, Argonne National Laboratory DESCRIPTION:Message passing via MPI is widely used in single-program, multiple-data (SPMD) parallel programs. Existing data-flow frameworks do not model the semantics of message-passing SPMD programs, which can result in less\nprecise and even incorrect analysis results. We present a data-flow analysis framework for performing interprocedural analysis of message-passing SPMD programs. The framework is based on the MPI-ICFG representation, which is an interprocedural control-flow graph augmented with communication edges between possible send and receive pairs and context-sensitivity. We demonstrate our techniques on the nonseparable analysis, activity analysis. Activity analysis is a domain-specific analysis used to reduce the computation and storage requirements for automatically differentiated MPI programs. Our experimental results have shown that using the MPI-ICFG data-flow analysis framework improves the precision of activity analysis and as a result significantly reduces memory requirements for the automatically differentiated versions of a set of parallel benchmarks. SUMMARY:Improving the Precision of Activity Analysis on MPI Programs UID:711 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20091023T103000 DTEND;TZID=America/Chicago:20091023T113000 DTSTAMP:20130525T020110 LOCATION:Building 240 / Room 1C2, Argonne National Labortory DESCRIPTION:We describe the development of graphics processor (GPU) accelerated iterative methods and preconditioners for solving linear systems arising in science and engineering applications. Krylov subspace methods including GMRES and TFQMR have been implemented on the GPU using Nvidia Cuda. The performance of these algorithms is intimately related to the performance of matrix-vector multiplies and sparse-matrix storage formats. The convergence of an iterative solver is most often determined by the preconditioner employed. The data-parallel architecture of the GPU computing platform must be considered when selecting a preconditioner. We also report on the development of a set of preconditioners, working together with GPU accelerated iterative solvers, to provide significant acceleration. These include, but are not limited to, block Jacobi, algebraic multigrid (AMG) with novel smoothing techniques, sparse approximate inverses and CPU based ILU(k) preconditioners that work with GPU solvers. SUMMARY:GPU Accelerated Krylov Subspace Methods and Preconditioners UID:712 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20091029T133000 DTEND;TZID=America/Chicago:20091029T143000 DTSTAMP:20130525T020110 LOCATION:Searle Lab, room 240, University of Chicago DESCRIPTION:Increasingly massive datasets produced by tomorrow\'s simulations beg the question: How will we connect this data to the computational and display resources that support visualization and analysis? This question is driving research into new approaches to allocating computational, storage, and network resources. Some of these approaches rest on creating and exploiting ways to optimally couple these resources in real time.\n\nExamples of what we mean by resource-coupled computations abound. For example, remote visualization is an activity which may couple data and large computation resources at the shared facility to client software and display hardware at the remote site. In situ analysis and visualization contemporaneously merges simulation and analysis onto the shared resource of the supercomputing platform. Co-analysis approaches will directly couple simulations running on a primary supercomputer to live analysis running on an optimized visualization and analysis platform over a high performance network.\n\nConsequently, we are working on a systems approach to modeling the end-to-end activity of extracting understanding from computational models. In this paper we will present our models and results from experiments designed to test them. SUMMARY:Modeling Resource-Coupled Computations UID:713 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20091022T133000 DTEND;TZID=America/Chicago:20091022T143000 DTSTAMP:20130525T020110 LOCATION:Searle Lab, room 240, University of Chicago DESCRIPTION:Scripting languages let users rapidly compose assemble existing programs into more powerful applications. In parallel scripting, users apply parallel composition constructs to existing sequential or parallel programs to develop highly parallel applications.\n\nParallel scripts are quite flexible and portable, and can run efficiently on platforms ranging from multicore workstations to petascale supercomputers. For many applications, like parameter sweeps and data analysis, parallel scripting is easier, more accessible, and more productive than tightly-coupled parallel programming.\n\nIn this talk, we will describe how the Swift parallel scripting system (www.ci.uchicago.edu/swift) is used to run scientific applications on petascale systems like IBM Blue Gene/P and Sun Constellation. We’ll present case studies, discuss architectural challenges of large-scale systems in the areas of scheduling and data management, and examine solutions that are also relevant to effectively using the PADS system. SUMMARY:Parallel Scripting for Science Applications at the Petascale and Beyond UID:714 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20091105T133000 DTEND;TZID=America/Chicago:20091105T143000 DTSTAMP:20130525T020110 LOCATION:Searle Lab, room 240, University of Chicago DESCRIPTION:The Petascale Active Data Store (PADS) Fall Seminar Series is a forum for discussions of data intensive computing. Our first series held Fall 2008 introduced the National Science Foundation funded PADS system, its hardware and software infrastructure, and highlight some of the scientific domain partners and how they plan to use the environment.\n\nThis week\'s lecture is with CI Faculty and Fellow, Gordon Kindlmann. SUMMARY:Particle Systems for Robust Feature Sampling and Visualization of Three-Dimensional Imaging UID:715 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20091112T133000 DTEND;TZID=America/Chicago:20091112T143000 DTSTAMP:20130525T020110 LOCATION:Searle Lab, room 240, University of Chicago DESCRIPTION:Montage is an astronomical image mosiacking tool that has been used in a variety of infrastructures. This talk will discuss the history of the application and how it has been used on single processor systems, clusters, and grids, including discussing performance on these systems. More recently, we have looked at some general questions about distributed applications, using Montage as a sample application. We have some initial ideas on objectives for developing, deploying, and executing distributed applications, and we have attempted to begin using these ideas for Montage, and will discuss this work, which includes running Montage on a mix of grids an SUMMARY:Using Montage, an Astronomical Mosaicking Application, to Explore Data-Oriented Distributed Computing UID:716 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20091124T133000 DTEND;TZID=America/Chicago:20091124T143000 DTSTAMP:20130525T020110 LOCATION:Building 240, Room 1404-1405, Argonne National Laboratory DESCRIPTION:Interactions among individuals are often modeled as social networks\nwhere nodes represent individuals and an edge exists if the\ncorresponding individuals have interacted during\nthe observation period. The model is essentially static in that the\ninteractions are aggregated over time and all information about the time\nand ordering of social interactions is discarded. We show that such\ntraditional social network analysis methods may result in incorrect\nconclusions on dynamic data about the structure of interactions and the\nprocesses that spread over those interactions.\n\nWe have extended computational methods for social network analysis to\nexplicitly address the dynamic nature of interactions among individuals.\nWe have developed techniques for identifying persistent communities,\ninfluential individuals, and extracting patterns of interactions in\ndynamic social networks. We will discuss computational properties of the\nanalysis problems and algorithms for solving them. Time permitting, we\nwill demonstrate the applicability of the techniques by analyzing zebra\nsocial networks.\n SUMMARY:Computational Analysis of Dynamic Networks UID:741 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20091207T100000 DTEND;TZID=America/Chicago:20091207T110000 DTSTAMP:20130525T020110 LOCATION:Knapp Center for Biomedical Discovery, Room 1103, University of Chicago DESCRIPTION:This is a special seminar sponsored by the Department of Medicine, Department of Radiology, Computation Institute and Institute for Genomics & Systems Biology. SUMMARY:Is Systems Biology Becoming a Data Intensive Science? Assuming So, Are You Ready? UID:749 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20091023T110000 DTEND;TZID=America/Chicago:20091023T120000 DTSTAMP:20130525T020110 LOCATION:Building 240, Conference Center Rooms 1406 & 1407, Argonne National Laboratory DESCRIPTION:The talk will cover the development of cluster ion beam technology, including historical background, fundamental characteristics of cluster ion to solid surface interactions, emerging industrial applications, and identification of some of the significant events which occurred as the technology has evolved into what it is today. The new ion beam processes, using cluster ions which consist of substantial numbers of atoms or molecules, are very different from those produced by impact of conventional ions comprised of single atoms or molecule. Cluster-surface collisions produce important non-linear effects which are being applied to shallow junction formation, to etching and smoothing of semiconductors, metals, and dielectrics, to assisted formation of thin films with nano-scale accuracy, and to other surface modification applications. Some of the unique industrial applications which have been done by a Japanese large-scale government project (METI/NEDO) will also be presented. SUMMARY:Cluster ION Beam Technology for Nano-scale Surface Processing UID:719 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20091021T133000 DTEND;TZID=America/Chicago:20091021T143000 DTSTAMP:20130525T020110 LOCATION:Building 240, Room 3C1, Argonne National Laboratory DESCRIPTION:Storage systems on leadership-class machines can deliver tens of gigabytes of performance at peak, yet applications frequently see much less performance in practice. High Level I/O libraries such as HDF5 can bridge the gap between application I/O and the storage system. This seminar will give an introduction to using the HDF5 library, with a focus on parallel I/O and performance tuning options.\n\nQuincey has been with The HDF Group since its founding and started with the HDF team in 1991, when it was still part of the National Center for Supercomputing Applications. He serves as the Director of Software Development, overseeing the design and architecture of the HDF5 software, as well as providing software engineering leadership. Quincey received his Bachelor’s degree in Electrical Engineering \nfrom the University of Illinois and is pursuing his Master\'s degree in Computer Science from the U of I. SUMMARY:Parallel I/O with HDF5: Overview and Tuning UID:720 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20091029T103000 DTEND;TZID=America/Chicago:20091029T113000 DTSTAMP:20130525T020110 LOCATION:Bldg: 240, Conference Room 1404, 1405, and 1406, Argonne National Laboratory DESCRIPTION:Autotuning technology has emerged recently as a systematic process for evaluating alternative implementations of a computation to select the best-performing solution for a particular architecture. At a LANS seminar in May, I introduced compiler-based empirical performance tuning and presented my success of applying it to a dense matrix-multiply kernel for small, rectangular matrices. \n\nIn this talk, I will begin with a summary of the talk, and then present my recent progress since then. A major result is that I could use the same technique to tune a higher-level kernel which is a loop with a call to a dense matrix multiply routine for small matrices. The kernel performance is up to 82% of peak on an AMD Phenom processor. With the tuned higher-level kernel and the library of tuned matrix multiply routines produced earlier, the whole Nek5000 program achieves 21% speedup on 256 nodes of the Cray XT5 at Oak Ridge National Laboratory. Also, I will show the overheads and fluctuations in measurements and how I overcame them for this experiment. SUMMARY:Speeding up Nek5000 with Autotuning and Specialization UID:723 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20091028T110000 DTEND;TZID=America/Chicago:20091028T120000 DTSTAMP:20130525T020110 LOCATION:Building 240, TCS Conference Center (1416), Argonne National Laboratory DESCRIPTION:Doug Kothe is the director of science for the NCCS at ORNL, responsible for guiding the multidisciplinary research teams using the center’s leadership computing systems. Doug has more than 20 years of experience in computational science research. His research interests and expertise have centered on developing physical models and numerical algorithms for simulating physical processes in the presence of incompressible and compressible fluid flow. A leader in modeling interfacial flows, he has been the principal developer of broadly disseminated scientific simulation tools. His most notable contribution has been the development of methods for flows possessing interfaces, especially free surfaces.Before joining the NCCS, Doug was deputy program director for Theoretical and Computational Programs in the Advanced Simulation and Computing (ASC) Program at Los Alamos National Laboratory (LANL). He served for several years as the leader of ASC’s Telluride Project, which developed the advanced manufacturing simulation tool “Truchas” for the Department of Energy complex. He joined the technical staff at LANL in 1988 as a member of the Fluid Dynamics Group, in which he helped develop the Ripple, Pagosa, and CFDLIB computational fluid dynamics codes. He later worked in the Structure/Relations Group and was group leader of the Continuum Dynamics Group.\n\nDoug received his B.S. in chemical engineering from the University of Missouri–Columbia and his M.S. and Ph.D. in nuclear engineering from Purdue University. He is the author of more than 60 refereed publications and has written more than a half-million lines of source code.\n SUMMARY:Progress & Challenges in the High-Fidelity Modeling of Interfacial Flows in Scalable Multi-Physics Applications UID:724 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20091028T140000 DTEND;TZID=America/Chicago:20091028T150000 DTSTAMP:20130525T020110 LOCATION:Building 240, Room 5178, Argonne National Laboratory DESCRIPTION:Radar-based Observations of Convection SUMMARY:Radar-based Observations of Convection - A Tale of Two Cities UID:725 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20091029T080000 DTEND;TZID=America/Chicago:20091029T000008 DTSTAMP:20130525T020110 LOCATION:Building 240, Rooms 1404, 1405, Argonne National Laboratory DESCRIPTION:Ray Bair, LCRC director will cover how employees may obtain Fusion accounts, computer time and support. Fusion is intended for laboratory-wide use and is located in the Theory and Computing Sciences Building SUMMARY:LCRC Supercomputer Briefing UID:726 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20091105T103000 DTEND;TZID=America/Chicago:20091105T113000 DTSTAMP:20130525T020110 LOCATION:1406&1407, Conference Center, Bldg 240, Argonne National Laboratory DESCRIPTION:A particle-based nonlinear filtering scheme will be presented. This algorithm is based on implicit sampling, a new sampling technique related to chainless Monte Carlo method. Posterior densities are represented by pseudo-Gaussians and the filter is designed to focus particle paths sharply so as to reduce the number of particles needed in the nonlinear data assimilation. Examples will be given. SUMMARY:Implicit sampling for nonlinear filters UID:727 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20091103T103000 DTEND;TZID=America/Chicago:20091103T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, Conference Room 1406 and 1407, Argonne National Laboratory DESCRIPTION:This talk presents the design and implementation of an asynchronous data-staging strategy for \nfile accesses based on ROMIO, the most popular MPI-IO distribution, and ZOID, the I/O forwarding component \nof ZeptoOS, an open source operating system solution for Blue Gene systems. We describe and evaluate \na two-level file write-back and prefetching solution. The experimental results demonstrate that our\napproach achieves high performance through a high degree of overlap between computation, communication, \nand file I/O. SUMMARY:Design and evaluation of multiple level file write-back and prefetching for Blue Gene/P UID:728 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20091104T103000 DTEND;TZID=America/Chicago:20091104T113000 DTSTAMP:20130525T020110 LOCATION:Building 240 / Room 1172 (1C2), Argonne National Labortory DESCRIPTION:Recent advances in computer technology combined with new theoretical methodology and algorithms have allowed for dramatic improvement in the computation of molecular properties. This has allowed computational chemistry to become a rapidly growing field of research, which plays a key role in many different areas of chemistry ranging from the interpretation of experiments, understanding of species that can be difficult or impossible to study experimentally (e.g. arsenic containing compounds and interstellar molecules), and chemical reactivity, to name just a few. Unfortunately, even with these advances, the extensive computational cost (i.e. computer time, memory, and disk space) of the sophisticated methods required to achieve a high level of accuracy effectively limits the size of molecules that can be studied to fewer than 10-15 atoms. Several approaches have been developed to help reduce the computational cost of expensive ab initio methods, and I will discuss several such methods including the correlation consistent Composite Approach (ccCA), hydrogen basis set truncation, and local correlation. SUMMARY:Reducing the Computational Cost of Ab Initio Methods UID:732 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20091106T103000 DTEND;TZID=America/Chicago:20091106T113000 DTSTAMP:20130525T020110 LOCATION:Building 240 / Room 1172 (1C2), Argonne National Labortory DESCRIPTION:Molecular dynamics (MD) simulations offer a tantalizing window into the workings of biological systems in full atomistic detail. The promise of this methodology for obtaining detailed understanding of biomolecular mechanisms depends in part on the ability of these simulations to reach biologically-relevant timescales. Due to the tremendous gap between the timescales of atomistic motion and the timescales of biological processes, simulation of these processes requires extremely efficient parallel implementations of MD algorithms on high performance computing systems. Recent implementations of MD algorithms on specialized hardware are poised to significantly increase our ability to study biological processes, as well as vigorously test the accuracy of these methods. In this talk I will discuss, in the context of these recent developments, some of my work to facilitate the study of biological processes through molecular dynamics simulations. SUMMARY:The Challenge of Long-Timescale Molecular Dynamics Simulations of Biological Systems UID:733 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100127T150000 DTEND;TZID=America/Chicago:20100127T160000 DTSTAMP:20130525T020110 LOCATION:TCS Building, Argonne National Laboratory DESCRIPTION:Parallel computers with more than a million processing cores will be available in the next three years. Although MPI is the dominant programming interface today for large-scale systems that already exceed 100,000 cores, many people wonder whether MPI will scale to systems with millions of cores. This talk will examine the issue of scaling MPI to such large core counts, including what needs to be fixed in the MPI standard and what needs to be fixed by MPI implementations. We will present some results of improving MPI\'s memory consumption at scale on Argonne\'s Blue Gene/P. SUMMARY:MPI on Millions of Cores UID:734 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20091106T110000 DTEND;TZID=America/Chicago:20091106T120000 DTSTAMP:20130525T020110 LOCATION:203 Auditorium, Argonne National Laboratory DESCRIPTION:Experiments with cold Fermi atoms provide an exciting new testing ground for our understanding of strongly-correlated matter. Even though cold atoms can be described by extremely simple interactions the resulting physics is extremely rich. A prototypical example is the BCS-BEC transition, where fermion pairing evolves from weak pairing of typical superfluids to very strong pairing into bosons. May of the phenomenon observed experimentally have clear analogues in nuclear physics and related fields. Examples include the equation of state and superfluid pairing, vortices and lattices, sound velocities and the ratio of shear viscosity to entropy. Searches and predictions for new \'exotic\' states of polarized superfluids will also be described, as well as directions for future research. SUMMARY:Rich Physics from Simple Interactions UID:739 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20091124T103000 DTEND;TZID=America/Chicago:20091124T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, TCS Conference Center (1406 & 1407), Argonne National Laboratory DESCRIPTION:Computer simulations are currently used to predict nuclear reactor phenomena such as neutron and heat transport, fluid dynamics, and the effect or radiation on the properties of nuclear fuel elements and structural materials. An important component of the complex simulation methodology is predicting the performance of nuclear fuel elements. After a brief review of world-wide status of fuel performance codes, the presentation focuses on recent Finite Element simulations of coupled heat transfer, chemical species diffusion and thermal expansion of UO2+x fuel elements with metallic clad. The continuum simulations incorporate multi-scale models and simulations of fuel properties, such as atomistic (Molecular Dynamics) models of point defect concentration and meso-scale (Phase Field) simulations of gas bubble formation. The continuum, coupled simulations demonstrate that including the dependence of thermal conductivity and density on local composition (oxygen and fission products content) leads to changes in the predicted centerline temperature that exceed 5%. The final part of the talk is dedicated to a discussion of national and international strategies for developing advanced, innovative models and high performance simulations for nuclear energy applications. SUMMARY:Advanced Models and Simulations for Nuclear Energy UID:743 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20091130T133000 DTEND;TZID=America/Chicago:20091130T000008 DTSTAMP:20130525T020110 LOCATION:Building 240 / Conference Center Room 1404 / 1405, Argonne National Laboratory DESCRIPTION:The presentation is intended to provided a forum for discussion of a collaboration between the GFDL and ANL on development of an ultra high-resolution global climate/weather model for exascale systems. Over the last year, significant progress has been made in development of a 4.5km global non-hydrostatic atmospheric dynamical core on the IBM-BG/P. \n\nThe talk will describe the software issues that were addressed to enable the infrastructure to support this development. Construction has begun on a prototype 4.5km global model with full-physics and we shall describe the software challenges that will need to be addressed to accomplish the projects science and software infrastructure goals. We intend that the collaboration between the GFDL and ANL will expand into a community activity in which the global model will be used by the community as a resource for experimentation into climate change. SUMMARY:How to Build and Support a Prototype Ultra High-Resolution Climate/Weather Model for ExaScale Systems UID:746 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20091210T103000 DTEND;TZID=America/Chicago:20091210T113000 DTSTAMP:20130525T020110 LOCATION:Building 240 / Room 1404, Argonne National Labortory DESCRIPTION:Combustion process is at the root of most energy production systems. This is especially the case for propulsion systems (piston engines/aeronautical turbines or rockets). The giant leaps performed in computer science over the past two decades have allowed us to use simulation to understand combustion in real configurations instead of only in academic laboratory burners. This presentation discusses the application of high performance computing for Computational Fluid Dynamics (CFD) in industrial cases for aeronautical turbines using the Large Eddy Simulation (LES) approach. Special attention is given to work performed under the INCITE program using the ALCF BlueGene/P to study to unsteady dangerous behaviours of flames, which these combustors often exhibit. SUMMARY:Application of High Performance Computing to Study Combustion on Industrial Gas Turbines Using Large Eddy Simulation UID:752 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100127T080000 DTEND;TZID=America/Chicago:20100127T170000 DTSTAMP:20130525T020110 LOCATION:Building 240 (TCS) Conference Center, Argonne National Laboratory DESCRIPTION:The workshop will provide both new and renewed INCITE projects with valuable information on ALCF services and resources, technical details on the IBM Blue Gene/P architecture. SUMMARY:INCITE \'Getting Started\' Workshop UID:757 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100325T150000 DTEND;TZID=America/Chicago:20100325T160000 DTSTAMP:20130525T020110 LOCATION:Building 240, Room 1404-1405, Argonne National Laboratory DESCRIPTION:Stochastic programming (SP) is concerned with optimization problems incorporating uncertain parameters in the objective and/or constraints. The sample average approximation (SAA) method is used to solve SP problems as very large deterministic optimization problems.\n\nInterior-point methods (IPM) proved in the last twenty years an unequaled efficiency in solving very large scale optimization problems. In a primal-dual IPM framework, the most time expensive task is solving for Newton linearization steps from a linear system involving the Hessian matrix. The constraints of the deterministic SAA SP problems have a block half-arrow shaped Jacobian that allows the Hessian to be permuted in a nested block arrow shape form. We employ the Direct Schur Complement method to make use of the particular structure of the Hessian. This method is also suitable for a scenario-based tree-like parallelization. The only bottleneck in the parallel execution flow is the factorization of the dense Schur complement. The Preconditioned Schur Complement method removes this bottleneck by using preconditioned Krylov subspace iterative methods. The preconditioner we propose approximates the inverse of the Schur complement \"exponentially\" better as a larger number of scenarios are considered.\n\nWe also present PIPS, a parallel solver for multi-stage programming based on interior-point methods. In this talk we report on the performance of PIPS in the context of two real-life stochastic optimization problems: a building energy system control problem and a unit commitment problem. SUMMARY:Scalable Multi-stage Stochastic Programming via Interior-point Methods UID:758 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100303T150000 DTEND;TZID=America/Chicago:20100303T160000 DTSTAMP:20130525T020110 LOCATION:Building 240, Room 4301, Argonne National Laboratory DESCRIPTION:In practice, samples from a probability conditioned on a large amount of history observation datum are required. Due to the high dimensionality and complexity of the distribution, sequential important sampling, resampling (SMC) methods is used. Rather than sample from the targeted distribution once, it generates a sample step by step. In this talk, we will introduce the method and use it to do safety assessment of chemical plant. SUMMARY:Sequential Monte Carlo Method in Chemical Safety Assessment UID:759 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100317T150000 DTEND;TZID=America/Chicago:20100317T160000 DTSTAMP:20130525T020110 LOCATION:Building 240, Room 1404-1405, Argonne National Laboratory DESCRIPTION:In this talk, we will discuss the solution methods for two classes of mixed-integer nonlinear programming (MINLP) problems that usually occur in the design and operations of process systems, namely the mixed-integer linear fractional programs and the separable concave minimization problems. To solve these problems effectively, we proposed global optimization algorithms that rely on solving only a sequence of mixed-integer linear programming (MILP) subproblems, without the need of handling any nonlinear program. We will prove the optimality of these algorithms and discuss their convergence properties. Computational results will be presented to show that orders of magnitude reduction in CPU times can be achieved when using these algorithms compared to solving the large-scale problems with commercial MINLP solvers. Real world applications of these methods in the chemical, pharmaceutical, industrial gas, paper and pulp industries will also be addressed. SUMMARY:MINLP in the Process Industry: Theory, Computation and Real World Applications UID:760 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100113T160000 DTEND;TZID=America/Chicago:20100113T170000 DTSTAMP:20130525T020110 LOCATION:Biological Sciences Learning Center, Room 205, 924, University of Chicago DESCRIPTION:Gordon Mills, University of Texas SUMMARY:Systems Approach to Personalized Medicine UID:764 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100113T150000 DTEND;TZID=America/Chicago:20100113T160000 DTSTAMP:20130525T020110 LOCATION:Bldg 240 Conference Center (1404-1405), Argonne National Laboratory DESCRIPTION:Given a set of jobs need to be done, and also the multiple inter cyclic dependencies between those jobs, how to find a optimal strategy to schedule those jobs and make them all done in the most efficient manner? How to represent the relations between all the jobs using some kind of mathematical structure? These questions will be addressed in the talk with the introducing of high-dimension graph and an algorithm based on it. Some application using the algorithm will also be shown in the talk. SUMMARY:Scheduling Algorithm for Cyclic Dependent Jobs UID:765 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100115T103000 DTEND;TZID=America/Chicago:20100115T113000 DTSTAMP:20130525T020110 LOCATION:Building 240 / Room 1172 (1C2), Argonne National Labortory DESCRIPTION:The objective of this seminar is to present the author\'s work in the conception, development and validation of new functionalities into the parallel and unstructured LES (large-eddy simulation) solver AVBP, and is based on her PhD thesis work and on recent progresses made at CERFACS on the subject. All this work is motivated by the rapid increase in computing power, which opens a new way for simulations that were prohibitive one decade ago. SUMMARY:Development of a Lagrangian Particle Tracking Scheme on a Parallel and Unstructured Solver -- Influence of Developments on Other Applications. UID:766 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100113T103000 DTEND;TZID=America/Chicago:20100113T000008 DTSTAMP:20130525T020110 LOCATION:Bldg: 240, Conference Rooms 1404 and 1405, Argonne National Laboratory DESCRIPTION:A large number of matrix techniques and theories were developed in the past few decades to address the problems of partial differential equations, which arguably shaped the current era of numerical linear algebra. Its main use lies in solving linear systems and eigenvalue problems, which constitute the major challenges in a large number of scientific and engineering applications. On the other hand, the well developed matrix techniques can be exploited in many scenarios, where the link between the application and its linear algebra content may sometimes be subtle. We discuss in this talk five such instances, all of which can be translated into matrix problems, and existing matrix techniques can be adapted or novel methods can be developed to solve the original applications. They include performing dimension reductions to a high dimensional data set, constructing nearest neighbors graphs for high dimensional data, identifying communities in a network system, measuring the connectivity of a pair of nodes in a graph, and sampling points from a multivariate normal distribution. These examples indicate that numerical linear algebra plays an important role in emerging areas such as data mining, and that it provides indispensable tools for computational and scientific applications. SUMMARY:Numerical Linear Algebra for Data Mining and Scientific Computing UID:767 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100331T150000 DTEND;TZID=America/Chicago:20100331T160000 DTSTAMP:20130525T020110 LOCATION:Bldg 240, Room 4301, Argonne National Laboratory DESCRIPTION:Higher order derivative tensors are efficiently computed via univariate Taylor polynomials in multiple directions. This permits parallelization across the directions. We compare various parallelization approaches in the context of generated overloading libraries. In particular we compare the use of OpenMP vs another approach that uses pthreads with atomic operations provided by MCS\'s own OPA library. We discuss the bottlenecks and the overhead in the current implementation in the Rapsodia tool. SUMMARY:Asynchronous parallel computation of higher order derivative tensors UID:769 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100115T103000 DTEND;TZID=America/Chicago:20100115T000008 DTSTAMP:20130525T020110 LOCATION:Bldg: 240, Conference Rooms 1404 and 1405, Argonne National Laboratory DESCRIPTION:Dan will present S2QP: a second-derivative sequential quadratic programming (SQP) method. The method is based on the S$\\ell_1$QP method by Fletcher and is thus based on minimizing the $\\ell_1$ exact penalty function. The key contribution (improvement), however, is that every subproblem is either convex and may be solved efficiently, or need not be solved globally. Dan will present limited numerical results from the CUTEr test set, which indicate that S2QP is at least capable of solving (efficiently) some large problems consisting of 10,000-100,000 variables/constraints.\n\nIn the last part of the talk, Dan will briefly consider future research.\n\nThis includes the future of S2QP, the formulation of a new trust-funnel algorithm, development of enriched multi-level recursive optimization, and the development of a regularized SQP method based on a new primal-dual augmented Lagrangian merit function. SUMMARY:Recent Work in Second-derivative Sequential Quadratic Programming Methods UID:770 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100122T150000 DTEND;TZID=America/Chicago:20100122T160000 DTSTAMP:20130525T020110 LOCATION:Building 240 / Conference Center Room 1404, Argonne National Labortory DESCRIPTION:The advent of petascale computing brings with it the promise of substantial increases in physical fidelity for a host of scientific problems. However, the realities of computing on these resources are daunting, and the architectural features of petascale machines often require considerable innovation for effective use. Nevertheless, there exists a class of scientific problems whose ultimate answer requires the application of petascale (and beyond) computing. One example is ascertaining the core-collapse supernova mechanism and explaining the rich phenomenology associated with these events. These stellar explosions produce and disseminate a dominant fraction of the elements in the Universe; are prodigious sources of neutrinos, gravitational waves, and photons across the electromagnetic spectrum; and lead to the formation of neutron stars and black holes. I will describe our recent multidimensional supernova simulations performed on petascale platforms fielded by the DOE Office of Science and the National Science Foundation. SUMMARY:Petascale Core-Collapse Supernova Simulation UID:773 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100126T150000 DTEND;TZID=America/Chicago:20100126T160000 DTSTAMP:20130525T020110 LOCATION:Building 240 TCS Conference rooms 1404 & 1405, Argonne National Laboratory DESCRIPTION:Chapel is a new programming language being developed by Cray Inc. as part of the DARPA-led High Productivity Computing Systems program (HPCS). Chapel strives to increase productivity for supercomputer users by supporting higher levels of abstraction compared to current parallel programming models while also supporting the ability to optimize for performance that meets or surpasses current technologies. Chapel is designed for portability -- from desktop multicore workstations to commodity clusters to the high-end machines developed by Cray and our competitors. In this talk, I will provide an overview of the Chapel language, including motivating philosophies and recent work on user-defined data distributions. I\'ll also mention several opportunities for collaboration and future work.\n\n SUMMARY:Chapel, the Cascade High-Productvity Language UID:775 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100127T133000 DTEND;TZID=America/Chicago:20100127T143000 DTSTAMP:20130525T020110 LOCATION:240 Conference Room 1404-1405, Argonne National Laboratory DESCRIPTION:Several factors are driving the growth of scientific simulations. Computational power of computer clusters is growing while the price of individual computers is decreasing. Distributed computing techniques allow thousands of computer nodes to participate in a single simulation. The benefit of this computational power is that simulations are getting more accurate and useful for predicting complex phenomena. The downside to this growth in computational power is that enormous amounts of data need to be saved and analyzed to determine the results of the simulation. The ability to generate data has outpaced our ability to save and analyze the data. This bottleneck is throttling our ability to benefit from our improved computing resources. \n\nIn this talk, I will discuss a few of Kitware\'s projects that aim to close the gap between simulation and analysis. The main focus will be on in-situ processing (aka co-processing). In-situ processing involves tying the visualization/analysis code with the simulation code. We have been developing tools to enable this type of processing by extending the ParaView visualization framework. I will also briefly talk about collaborative visualization using desktop and web applications as well as the analysis of dataset ensembles.\n SUMMARY:Ultra-scale Visualization with Open-Source Software UID:782 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100129T103000 DTEND;TZID=America/Chicago:20100129T000008 DTSTAMP:20130525T020110 LOCATION:Bldg: 240, Conference Rooms 1404 and 1405, Argonne National Laboratory DESCRIPTION:Creating peak-performance and scalable compute codes on graphics processors is a challenge that is aggravated by complicated and constantly changing hardware. In this talk, I will describe techniques and tools to tap the enormous performance potential of GPUs for discontinuous Galerkin finite element solvers. Particular emphasis will\nbe on the advantages that high-order discretizations offer on modern SIMD-like architectures. I will explain a few of the design considerations and tricks that enabled sustained single-chip floating point performance of above 200 GFlops/s across a wide range of discretization parameters and equation types. I will introduce tools for run-time code generation and empirical optimization from a high-level language that were crucial to the present effort. With the infrastructure in place, further discussion will concern some potential applications and perspectives on how this technology might change requirements for algorithms that work alongside PDE solvers, such as time steppers and linear solvers. SUMMARY:High-Order Discontinuous Galerkin Methods by GPU Metaprogramming UID:783 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100129T083000 DTEND;TZID=America/Chicago:20100129T170000 DTSTAMP:20130525T020110 LOCATION:Building 240 TCS Conference Center, Argonne National Laboratory DESCRIPTION:The International Workshop on Unipolar Arc will explore high-electric field gradient arcing in many environments: tokamaks, accelerators, laser ablation, spark gaps, cathodic arcs, e beam welding, explosive electron mission, etc., to try to understand if there are mechanisms in common and if data from one field is relevant and useful in others. Unipolar arcs may or may not be a general phenomenon, however the idea may be potentially very useful in understanding plasma material interactions. SUMMARY:The International Workshop on Unipolar Arc UID:784 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100201T103000 DTEND;TZID=America/Chicago:20100201T113000 DTSTAMP:20130525T020110 LOCATION:Building 240 / Conference Center Room 1407, Argonne National Labortory DESCRIPTION:Scientific software has always enjoyed its own special niche in the world of programming. From specialized languages, to the deep domain knowledge of its authors, lessons learned in the broader software community have often felt irrelevant or quaint. Yet this community is building increasingly complex software that is expected to survive for decades rather than years, and is also spanning multiple scientific disciplines. As computational science rightly takes its place along theory and experiment as a full-fledged scientific discipline, we must recognize and acknowledge that business as usual may not always be enough. And while science domain experts are clearly required to ensure success, these same leaders who are usually called upon to manage such efforts may often overlook, or simply not understand the importance of a number of other factors key to making a large project run smoothly. In this talk, I’ll use my 15 years of experience working on a large multi-disciplinary physics application to reflect on what I feel are important aspects for project leaders, managers, and developers to recognize and adopt. From technology, to software engineering, to the “soft sciences” of making a large team work together, I’ll touch on a range of issues that are common to all large scale scientific software effo SUMMARY:The Challenges of Multi-disciplinary Scientific Software Development UID:786 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100503T080000 DTEND;TZID=America/Chicago:20100503T170000 DTSTAMP:20130525T020110 LOCATION:Theory and Computer Science Conference Center (240, Argonne National Laboratory DESCRIPTION:The goal of this Symposium is to advance our understanding of how computational science and engineering can accelerate progress in scalable electrochemical energy storage science and technology.\n\nMotivated by society\'s great need for advances in energy-storage technology, and by the demonstrated achievements and tremendous potential for computational science and engineering, a consortium of IBM Research and three U.S. National Laboratories (Argonne, Oak Ridge, and Pacific Northwest) will hold a symposium on 3-4 May 2010 at Argonne National Laboratory to discuss how computational science will advance electrochemical energy storage science and technologies to accelerate innovation, decrease costs, and improve the safety of advanced, scalable electrochemical energy storage concepts and systems \"beyond lithium ion\".\n\nElectric vehicles can help us to end our dependence on petroleum imports, and to integrate renewable electricity generation into our transmission grid. Both of these outcomes have the potential for strong, positive impact on society and each is dependent upon breakthroughs and advances in electrical energy storage systems. SUMMARY:Beyond Lithium-Ion: Computational Perspectives UID:857 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100215T150000 DTEND;TZID=America/Chicago:20100215T160000 DTSTAMP:20130525T020110 LOCATION:Building 240, Rm 1404-1405, Argonne National Laboratory DESCRIPTION: The Lasso---the maximum likelihood estimator for a linear regression model with a penalty on the absolute values of the regression coefficients---is an important statistical method for a wide range of fields. The Lasso plays a key role in genetic association studies, where the goal is to discover genetic variants that make us more predisposed to certain diseases. One major appeal of the Lasso is that there are efficient optimization algorithms for computing the solution. The drawback is that the Lasso is unable to report its confidence in a solution, and this is vital for genetic association studies. I explain how to pose the full inference problem as an optimization problem with semidefinite constraints, and how formulating it in this way leads to a natural convex relaxation. This is work in progress. SUMMARY:Inference as optimization in a multivariate linear regression with Laplace priors UID:792 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100311T180000 DTEND;TZID=America/Chicago:20100311T203000 DTSTAMP:20130525T020110 LOCATION:The Oriental Institute, 1155 East 58th Street, University of Chicago DESCRIPTION:\"The Art of Science,\" is the first in a series\nof joint speaker events for university faculty and Argonne and Fermilab scientists, researchers and engineers.\n\nThe reception and panel discussion will be held Thursday, March 11, 2010, from 6-8:30 p.m. at The Oriental Institute, 1155 East 58th Street, Chicago, on the University of Chicago campus.\n\nThe moderator will be Rocky Kolb, Arthur Holly Compton Distinguished Service Professor in the university\'s Department of Astronomy and Astrophysics.\n\nPanelists will include:\n SUMMARY:The Art of Science UID:800 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100222T103000 DTEND;TZID=America/Chicago:20100222T113000 DTSTAMP:20130525T020110 LOCATION:Bldg: 240, Conference Room 4301, Argonne National Laboratory DESCRIPTION:We have developed several computational tools for metagenomic sequence annotation. I will mainly talk about FragGeneScan and SWIFT. FragGeneScan is a tool for (fragmental) gene prediction from short reads. It is based on a Hidden Markov Model (HMM) that incorporates the frame-shift-causing sequencing errors. We have tested FragGeneScan on simulated short reads as well as reads from metagenomic projects, and the results show that FragGeneScan outperforms other gene predictors on very short reads, and reads with sequencing errors. SWIFT is a tool for fast protein similarity search, which achieves speedup by utilizing reduced amino acid alphabet and flexible seeds. SWIFT achieves up to ~40 times speedup as compared to BLAST at the cost of a small loss of similarity detection sensitivity for short reads. SUMMARY:Computational tools for annotating metagenomic sequences: FragGeneScan and SWIFT UID:794 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100219T103000 DTEND;TZID=America/Chicago:20100219T113000 DTSTAMP:20130525T020110 LOCATION:Building 240 / Conference Room 4301, Argonne National Labortory DESCRIPTION:Scientific computing is facing another leap with the looming era of exascale computing. For the last 10 years or so, I have been developing techniques that can be used in tools for improving performance of scientific applications. In this talk, I will talk about two examples of such tools: TUNE and SIMDex. In December last year, I presented about TUNE, a compiler-based tool for empirical performance tuning, and its application on Nek5000. In this talk, I will follow up with new results that extend the earlier results in more detail. Also, I will summarize the unique features of SIMDex, a source-to-source SIMD code generator. While these tools are developed for Cray XT5 and AltiVec (as in G5 and Power6), they can be as useful on IBM BlueGenes. I conclude this talk by stressing on the importance of tool development and use. SUMMARY:Tool-Based Approach for High-Performance Computing UID:795 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100222T103000 DTEND;TZID=America/Chicago:20100222T113000 DTSTAMP:20130525T020110 LOCATION:Building 240 Conference Room 5178, Argonne National Laboratory DESCRIPTION:n preparation for the Vancouver 2010 Olympic and Paralympic Winter Games, Environment Canada expanded its observational network along the Sea-to-Sky corridor (Highway 99 in British Columbia), to support weather forecasting operations and research in this mountain region. High winds, reduced visibility and mixed-phase precipitation are some of the forecasting challenges for the outdoor Games venues, all enhanced by the complex terrain conditions.\n \nTropospheric profiling is enhanced here by using Microwave Profiling Radiometry, UHF Wind Profiler, K-band Micro Rain Radar, and C-band Doppler weather radar. A dedicated weather radar in the Sea-to-Sky provides local analysis in the main valleys around Whistler. Operational weather radars at Mount Sicker and Aldergrove provide a mesoscale context over the venues. Radiometry observations generate profiles of vapor density, of cloud liquid water content, and of temperature. These facilitate monitoring the growth and depletion of ice particles and super-cooled droplets in winter environments. This analysis leads to a technique for nowcasting precipitation phase. A 915 MHz wind profiler is located upstream of the Olympic venues, at the junction of three valleys. Analysis of these wind profiler observations provide insight on the precipitation phase and topographic influence over the local winds.\n \nThe objective here is to demonstrate how remote-sensing retrievals are enhancing the diagnosis and nowcasting of winter weather over complex terrain. For that, we present multi-sensor analyzes on the cloud and precipitation processes occurred at the Sea-to-Sky one winter prior to the Games and during the actual 2010 Winter Games. SUMMARY:Cloud Processes over the 2010 Winter Games Venues as Analyzed by an Enhanced Remote-Sensing Network UID:796 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100225T103000 DTEND;TZID=America/Chicago:20100225T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, Conference Room 4301, Argonne National Laboratory DESCRIPTION:As projected in 2018, an exascale system will have millions of processors with some processors potentially having hundreds of threads of execution. Such large scale, heterogeneous systems challenge system configurations and application adaptations. A potential solution is to establish performance models to quickly explore the space of system configurations and identify application-specific adaptations. Such a framework involves data collection, analysis, modeling, system reconfigurations, and choices of optimizations. It also requires a programming model that allows the same code to adapt to different workload decomposition and distribution schemes. Similar issues already arise in many-core systems, although at a much smaller scale. I will present my dissertation work with a focus on performance modeling and automatic optimization on GPUs, as well as hierarchical task decomposition for many-core architectures. I will then conclude the lessons learned, and talk about the possibilities to extend the work to exascale systems. \n\nBio:\n\nJiayuan Meng is a Ph.D. candidate at University of Virginia. His research interests mainly include parallel computation for data-intensive applications and infrastructural design for scalable computation platforms. His dissertation topic is data management for multi-threaded cores. He has built MV5, an event-driven, cycle-accurate many-core simulator based on M5. He has proposed techniques to improve performance scalability on many-core architectures from the aspects of caching protocols, run-time scheduling, dynamically adaptive SIMD architectures, and analytical performance models for GPU applications. He is also collaborating with NEC Laboratories America on domain-specific programming models for emerging recognition and mining applications. He has experiences with various applications including fluid dynamics, image synthesis, 3D rendering, and semantic analysis. He has received the 2009-2010 NVIDIA research fellowship and the 2010 U.Va. Award for Excellence in Scholarship in the Sciences & Engineering. SUMMARY:Towards a Scalable Framework for Performance Modeling and Automatic System Optimization --- from many-core to heterogeneous exascale systems UID:802 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100426T080000 DTEND;TZID=America/Chicago:20100426T170000 DTSTAMP:20130525T020110 LOCATION:TBA, Argonne National Laboratory DESCRIPTION:Molecular biology has revolutionized the study of microorganisms in the environment and improved our understanding of the composition, phylogeny, and physiology of microbial communities. The current molecular toolbox encompasses a range of DNA-based technologies and new methods for the study of RNA, proteins, and lipids extracted from environmental samples. Currently there is a major emphasis on the application of \"omics\" approaches to determine the identities and functions of microbes inhabiting different environments, which when combined with traditional molecular approaches and microbial physiology, biochemistry, and physical/chemical characterization of the environment, provides new opportunities for fundamental discoveries in microbial ecology.\n \nThis workshop will bring together researchers in the central Great Lakes region to discuss the latest advances in environmental molecular microbiology and microbial ecology and their potential for providing new insight into fundamental processes involved the biogeochemical cycling of major and minor elements, contaminant fate and transport, and climate change. Keynote addresses will be given by Dionysios Antonopoulos of Argonne National Laboratory/University of Chicago, and by Terence Marsh of Michigan State University. A combination of oral presentations and poster sessions will afford maximum opportunities for discussion of the latest advances in environmental molecular microbiology and identifying opportunities for collaboration. Tours will be offered for those interested in visiting the High-Throughput Genome Analysis Core, High-throughput Protein Production, and Protein Mapping facilities, as well as macromolecular crystallography (GM/CA CAT and the Structural Biology Center), X-ray absorption spectroscopy (MRCAT/EnviroCAT), and X-ray imaging capabilities (XOR) at the Advanced Photon Source. SUMMARY:Challenges in Environmental Molecular Microbiology Workshop UID:803 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100305T100000 DTEND;TZID=America/Chicago:20100305T110000 DTSTAMP:20130525T020110 LOCATION:Building 240 Conference Room 1404-1405, Argonne National Laboratory DESCRIPTION:Spiral (www.spiral.net) is a program and hardware design generation system for linear transforms such as the discrete Fourier transform, discrete cosine transforms, filters, and others. We are currently extending Spiral beyond its original problem domain, using coding algorithms (Viterbi decoding and JPEG 2000 encoding) and image formation synthetic aperture radar, SAR) as examples. For a user-selected problem specification, Spiral autonomously generates different algorithms, represented in a declarative form as mathematical formulas, and their implementations to find the best match to the given target platform. Besides the search, Spiral performs deterministic optimizations on the formula level, effectively restructuring the code in ways unpractical at the code or design level. Spiral generates specialized single-size implementations or adaptive general-size autotuning libraries, and utilizes special instructions and multiple processor cores. The implementation generated by Spiral rival the performance of expertly hand-tuned libraries.\n\nIn this talk, we give a short overview on Spiral. We explain then how Spiral generates efficient programs for parallel platforms including vector architectures, shared and distributed memory platforms, and GPUs; as well as hardware designs (Verilog) and automatically partitioned software/hardware implementations. We overview how Spiral targets the Cell BE and PowerXCell 8i, the Blue Gene/P PPC450d processors, as well as Intel\'s upcoming Larrabee GPU and AVX vector instruction set. As all optimizations in Spiral, parallelization and partitioning are performed on a high abstraction level of algorithm representation, using rewriting systems. SUMMARY:Spiral: Program Generation for Linear Transforms and Beyond UID:809 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100309T103000 DTEND;TZID=America/Chicago:20100309T113000 DTSTAMP:20130525T020110 LOCATION:Building 240 Seminar Room 4301, Argonne National Laboratory DESCRIPTION:Abstract\n\nConsidering the emerging large-scale scientific applications, there is a need for the system to provide a timely response to an important event. Computations need to be performed to handle the detected event. Often, this response can require significant computation and possibly communication, and can be very challenging to complete within the time-frame. The resources available for the processing may only be detected when the event occurs, and may not be known in advance. At the same time, there could be application-specific edibility in the computation that may be desired. There could be an application-specific benet function, which captures what is most desirable to compute.\n\nThe applications are usually composed of multiple components that are inherently dynamic and heterogeneous. There could be complex interactions between these components. In particular, our goal is to complete the task within the pre-specified time frame, while attempting to maximize the pre-specified benet function. Therefore, numerous performance-related parameters must be continuously tuned. Usually, there is a trade between parameter adaptation in favor of benet optimization and application execution time. This further depends on the capacity of the resources that the task was allocated. Thus, allocating appropriate resource collections to applications and setting optimal values for adjustable parameters are critical to application performance. Furthermore, as applications and resources are interacting in a complex and dynamic way, it is highly desirable that the procedure for resource allocation and parameter adaptation could be self-managing and self-optimization, requiring only a high-level objective, such as the benet function, as input to the system. \n\nWe have developed a middleware to support such functionality and to enable development and deployment of large-scale scientific applications. The main functionality of our middleware is to enable time-critical event handling to achieve the maximum benet, as per the application specific benet function, while satisfying the time constraint. This requires support for self-adaptation. Performing such optimization further leads us to a resource selection and scheduling problem. We have given a formal formulation based on optimal control theory and developed an autonomic adaptation algorithm. We did an efficiency value to react how effectively a particular service can be executed on a particular node. Based on which we have developed a greedy scheduling algorithm to schedule these service components. Our middleware is also based on the existing Grid infrastructure and Service-Oriented Architecture (SOA) concepts. Furthermore, we considered the research problems on how to successfully complete the time-critical event processing in presence of resource failures and how to maximize the benet given a resource budget in cloud computing environments. Experimental results have demonstrated the effectiveness of our proposed approaches in supporting time-critical events processing in distributed environments.\n SUMMARY:Supporting Time-Critical Event Processing in Distributed Environments UID:811 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100323T090000 DTEND;TZID=America/Chicago:20100323T154500 DTSTAMP:20130525T020110 LOCATION:Building 240 / Conference Center Room 1406/1407, Argonne National Labortory DESCRIPTION:-- Learn what cloud computing is all about\n-- Explore Magellan, Argonne\'s new cloud\n \nPart I: Get Your Head in a Cloud – an overview\n[9-11:30 a.m.]\nClouds have come to Argonne. Are they an appropriate resource for your research? Join us for an overview of our new cloud, Magellan, and discover the types of projects best suited for this mid-range computing system.\n \nPart II: Get Your Hands on a Cloud – hands-on session [1:30-3:45 p.m.]\nIf you’re ready to road test the system, email us for an account at magellan-support@alcf.anl.gov by March 17th. Then, bring your laptop and your code for hands-on assistance from our experts. SUMMARY:Welcome to Magellan Day UID:814 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100311T103000 DTEND;TZID=America/Chicago:20100311T113000 DTSTAMP:20130525T020110 LOCATION:Building 240 / Conference Room 4301, Argonne National Labortory DESCRIPTION:We are at the dawn of a new era in computational biology. DNA sequencing projects that required years of effort and hundreds of millions of dollars of equipment just a few years ago, can now be performed quickly and cheaply by individual labs. This dramatic shift is expanding the scale and scope of sequencing to previously unimaginable limits, and will ultimately lead to new discoveries of our basic biology, the diversity of life, and personalized medicine. However, these ambitious goals can only be realized if we can develop new computational methods that can effectively analyze the overwhelming volumes of data generated. \nIn my presentation, I’ll describe my research developing efficient methods for analyzing large biological datasets, including by using the parallel computing framework MapReduce developed by Google. My programs CloudBurst, Crossbow, and Contrail demonstrate how this technology can be applied to the critical tasks of large-scale alignment and genome assembly, enabling genotyping and de novo assembly of whole human genomes from billions of short reads in an afternoon. Coupled with inexpensive cloud computing, these programs can quickly, cheaply, and accurately analyze tremendous biological datasets and have the potential to make otherwise infeasible studies practical.\n SUMMARY:Scalable Solutions for DNA Sequence Analysis UID:815 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100315T103000 DTEND;TZID=America/Chicago:20100315T113000 DTSTAMP:20130525T020110 LOCATION:Building 240 / Conference Room 4301, Argonne National Labortory DESCRIPTION:Multi-core processors are becoming omnipresent in all kinds of computing platforms. They are moving quickly from few cores to hundreds or even thousands on the same chip. Along this increase in number of cores, each core is becoming simpler in design and faster in performance. These architectural changes mandate explicit parallelism at a fine-grained level in most of the HPC algorithms’ design and implementation. Moreover, such granularity adds more pressure on the system’s memory, which makes the memory latency again one of the most challenging barriers to fully utilize multi/many-core architectures. Traditional techniques, such as data prefetching, instruction-level parallelism, and coarse-grained multi-threaded models, are having diminishing returns of performance improvement or inapplicable for many algorithms. \n\nWe are proposing the micro-threading framework as a promising solution to hide memory latency inside multi/many-core architectures. In this research, we are adding another level of parallelism utilizing some of the multi/many-cores architectural aspects, such as explicit cache management, cores interconnection network, and cores heterogeneity. We are utilizing the Cell Broadband Engine as one of the leading heterogeneous multi-core processors to implement and experiment our micro-threading framework. Our implementation and measures using micro-threads show good performance improvements in scientific algorithms, such as the FFT. \n SUMMARY:Micro-Threading for Multi/Many-Cores Architectures UID:816 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100412T103000 DTEND;TZID=America/Chicago:20100412T113000 DTSTAMP:20130525T020110 LOCATION:Building 240 Seminar Room 4301, Argonne National Laboratory DESCRIPTION:In this talk, we will discuss two applications of optimization to real-world problems: graph partitioning and sparse image and signal reconstruction.Given a graph with edge weights, the graph partitioning problem is topartition the vertices into two sets satisfying specified size constraints, while minimizing the sum of the weights of the edges that connect the vertices in the two sets. This NP-complete combinatorial problem arises in many areas including parallel computing, sparse matrix factorizations, VLSI design, and data mining. We present an exact algorithm for solving this problem which is based on a branch and bound method applied to a continuous quadratic programming formulation. Lower bounds are obtained by decomposing the objective function into convex and concave parts and replacing the concave part by an affine underestimate.Many signal and image reconstruction problems amount to finding a sparse approximate solution to an underdetermined system of linear equations. Under certain assumptions, the problem can be approximated by a convex optimization problem. The SpaRSA algorithm has been shown to work well in practice for solving this nonlinear problem, however, had not been analyzed. We prove that if the objective function is convex, then the error in the function value at iteration k, for k sufficiently large, is bounded by a/(b+k) for suitable choices of a and b. Moreover, if the objective function is strongly convex, then the convergence is R-linear. An improved version of the algorithm based on a cycle version of the BB iteration and an adaptive line search is given. Numerical results providing empirical support to the theoretical results will be presented.\n SUMMARY:Optimization and its applications to graph partitioning and sparse recovery UID:841 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100310T130000 DTEND;TZID=America/Chicago:20100310T140000 DTSTAMP:20130525T020110 LOCATION:Bldg. 240, Conference Room 4301, Argonne National Laboratory DESCRIPTION:Direct volume rendering and isosurfacing are ubiquitous rendering techniques in scientific visualization, commonly employed in imaging 3D data from simulation and scan sources. Conventionally, these methods have been treated as separate modalities, necessitating different sampling strategies and rendering algorithms. In reality, an isosurface is a special case of a transfer function, namely a Dirac impulse at a given isovalue. However, artifact-free rendering of discrete isosurfaces in a volume rendering framework is an elusive goal, requiring either infinite sampling or smoothing of the transfer function. While preintegration approaches solve the most obvious deficiencies in handling sharp transfer functions, artifacts can still result, limiting classification.\n\nAaron will introduce a method for rendering such features by explicitly solving for isovalues within the volume rendering integral. In addition, we present a sampling strategy inspired by ray differentials that automatically matches the frequency of the image plane, resulting in fewer artifacts near the eye and better overall performance. SUMMARY:Better Adaptive Volume Ray Casting UID:818 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100311T150000 DTEND;TZID=America/Chicago:20100311T160000 DTSTAMP:20130525T020110 LOCATION:Building 240, Conference Room 4301, Argonne National Laboratory DESCRIPTION:Non-covalent van der Waals (vdW) forces are crucial for the formation,\nstability and function of molecules and materials. At present,\nubiquitous vdW interactions can only be accounted for properly by\nhigh-level quantum-chemical wave function or by the Quantum Monte\nCarlo (QMC) method. In contrast, the correct long-range interaction\ntail, e.g., for separated molecules, is absent from all popular\nlocal-density or gradient corrected exchange-correlation functionals\nof density-functional theory. We have recently developed an efficient\nmethod to obtain accurate vdW dispersion coefficients from first\nprinciples [1]. This method can be coupled to DFT calculations and\neven to quantum-chemical MP2 approach [2]. I will discuss the\ntheoretical underpinnings of the method and applications to different\ncases: intermolecular and intramolecular interactions [1,2],\norganic/organic interfaces [3], organic/inorganic interfaces [4,5] and\ncohesive properties of ionic and semiconductor solids.\n\n[1] A. Tkatchenko and M. Scheffler, Phys. Rev. Lett. 102, 073005 (2009).\n[2] A. Tkatchenko, R. A. DiStasio Jr., M. Head-Gordon and M.\nScheffler, J. Chem. Phys. 131, 094106 (2009).\n[3] N. Marom, A. Tkatchenko, M. Scheffler and L. Kronik, J. Chem.\nTheory Comp. 6, 81 (2010).\n[4] E. McNellis, J. Meyer, and K. Reuter, Phys. Rev. B 80, 205414 (2009).\n[5] G. Mercurio et al., Phys. Rev. Lett. 104, 036102 (2010).\n SUMMARY:Van der Waals Forces in Molecules and Solids UID:819 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100319T103000 DTEND;TZID=America/Chicago:20100319T113000 DTSTAMP:20130525T020110 LOCATION:Building 240 Seminar Room 1404 and 1405, Argonne National Laboratory DESCRIPTION:Abstract:\n\nPartitioned Global Address Space (PGAS) programming models such as Global Arrays (GA) and Unified Parallel C (UPC)provide the programmer with a global view of shared data that maybe partitioned across the physical memories of multiple nodes in a distributed memory cluster. This global address space supports asynchronous and irregular accesses to distributed shared data, however most PGAS models rely on expressing the computation using the regular, process-centric SPMD model. This disconnect between the irregular, asynchronous data model and the regular, process-centric computation model can pose significant challenges to the efficient and scalable implementation of irregular computations.\n\nIn this talk I will present Scioto, a scalable task parallel programming model that compliments existing PGAS memory models with a dynamic, task parallel view of the computation. Under the Scioto model, the programmer expresses their computation as a set of tasks that execute in the context of the global address space. The Scioto run time system manages parallel task execution and provides scalable run time services to enhance performance and program ability, including dynamic load balancing to manage irregularity and imbalance; task pushing and termination detection to support dynamic parallelism; and locality aware execution to allow the programmer to co-locate tasks with data to reduce communication overhead.\n\nIn addition to this work, I will briefly present ongoing and future research work in fault tolerance, including mechanisms for selective restart in the task parallel programming model; hybrid parallel programming models, including the MPI+UPC model; and PGAS work,including work on the Global Trees system for distributed shared linked data structures. SUMMARY:Scalable Task Parallel Programming in the Partitioned Global Address Space UID:821 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100414T150000 DTEND;TZID=America/Chicago:20100414T160000 DTSTAMP:20130525T020110 LOCATION:Bldg 240, Conference Center, Argonne National Laboratory DESCRIPTION:InfiniBand (IB) network architecture was initially proposed for system area (specifically data-center) networking. Given its rise in popularity over the past decade or so, researchers have started studying its relevance in other environments such as distributed computing systems, virtual machines, and other traditionally-Ethernet-based environments. Based on advances in these\nareas, people are now considering IB as a potential networking technology to be used in \"cloud systems\", which would use a combination of these enhancements. In this talk, I\'ll describe the status of IB in a number of environments including: (a) wide-area lambda grids, (b) performance and security in virtual machines, and (c) converged Ethernet/IB technologies that allow IB to be used in environments that have traditionally been dominated by Ethernet-only communication. In short, I\'ll discuss aspects related to the relevance of IB on cloud computing systems. SUMMARY:InfiniBand on the Cloud UID:824 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100519T150000 DTEND;TZID=America/Chicago:20100519T160000 DTSTAMP:20130525T020110 LOCATION:Bldg 240 Conference Center 1404-1405, Argonne National Laboratory DESCRIPTION:In order to improve numerical weather prediction, observations of the state of the lower atmosphere and surface are essential. Orbiting satellites provide a dense set of observations; however, the data is in the form of radiant flux density. Determining the temperature, wind speed, pressure and other meteorological quantities is thus an inverse ill-posed problem. The presence of clouds is a further complication that makes the optimization problem to use this data, called data assimilation, non-smooth in nature. In this talk, I will present the Aeolus framework developed over the last four months for simplifying the use of the Weather Research and Forecasting (WRF) model, present a brief introduction to data assimilation, review the problem of data assimilation of satellite radiances, and discuss non-smooth optimization strategies I will be investigating as part of my Ph. D. thesis at the Florida State University. SUMMARY:Non-Smooth Optimization in the Data Assimilation of Satellite Observations In Numerical Weather Prediction UID:854 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100728T150000 DTEND;TZID=America/Chicago:20100728T160000 DTSTAMP:20130525T020110 LOCATION:Bldg 240, Conference Center (1406-1407), Argonne National Laboratory DESCRIPTION:The Environmental Science Division (EVS) has been assisting the Department of Energy\'s Office of Environmental Management (DOE/EM) over the past few decades in cleaning up its nuclear complex. While tremendous progresses have been made on reducing DOE\'s footprint in contaminated lands over time, progressive technical challenges continue to pose as a threat to its cleanup schedule for the long term. Such challenges include the very complex migration of radioactive contamination in the environment and particularly through the subsurface formations over a widespread area such as DOE\'s Hanford Site. At the center of the issues lies an important question: How Clean Is Clean? The Environmental Science Division has developed the RESRAD Family of Codes to address such an issue. It is developed to derive the allowable residual radioactivity for meeting the regulatory limits and demonstrating compliance. Over time, RESRAD continues to evolve with increasing complexity and an equal demand for computing performance. In a recent effort, DOE/EM launched a multiple-lab, multiple-year initiative on the Advanced Simulation Capability for Environmental Management (ASCEM) aimed to introduce the modern day high-performance computing technology into simulating the environmental system and discovering better solutions to the complex subsurface contamination issues. Being the developer of the RESRAD series of codes and also a collaborator to the ASCEM program, we will highlight the RESRAD program and discuss future needs.\n SUMMARY:Development of RESRAD Series of Models for Environmental Cleanup UID:856 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100402T103000 DTEND;TZID=America/Chicago:20100402T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, Room 4301, Argonne National Laboratory DESCRIPTION:For fifteen years, we\'ve been hearing that IP version 4 was about to run out of addresses and IPv6 was about to replace it. Today, IPv4 address are indeed within a couple of years of running out. Is IPv6 ready and what are ISPs doing about it? The reality is that we face an indefinite period of IPv4/IPv6 interworking, and this will affect everybody, not just networking specialists. The seminar will address the latest developments in the IETF and present some results from\na recent survey of ISP experience and plans for IPv6.\n\nBiography\n\nBrian E. Carpenter joined the University of Auckland in September 2007. He was appointed Professor in January 2009 (with a part-time appointment).\n\nBefore that, he spent ten years with IBM at various locations, working on Internet standards and technology. From 1997 he was at IBM\\\'s Hursley Laboratory in England. From 1999 to 2001 he was at iCAIR, the international Center for Advanced Internet Research, sponsored by IBM at Northwestern University in Evanston, Illinois. He was most recently based in Switzerland as a Distinguished Engineer and a member of the IBM Academy of Technology.\n\nBefore joining IBM, he led the networking group at CERN, the European Laboratory for Particle Physics, in Geneva, Switzerland, from 1985 to 1996. This followed ten years\\\' experience in software for process control systems at CERN, which was interrupted by three years teaching undergraduate computer science at Massey University in New Zealand.\n\nHe holds a first degree in physics and a Ph.D. in computer science, and is a Chartered Engineer (UK). He has been an active participant in the Global Grid Forum, and in the Internet Engineering Task Force (IETF), where he has worked on IPv6 and on Differentiated Services. He has also worked with the CERN Openlab for Datagrid Applications. He served from March 1994 to March 2002 on the Internet Architecture Board, which he chaired for five years. He also served as a Trustee of the Internet Society, and was Chairman of its Board of Trustees for two years until June 2002. He was Chair of the IETF from March 2005 to March 2007. SUMMARY:IPv4 Exhaustion and IPv6 State of the Union UID:825 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100416T110000 DTEND;TZID=America/Chicago:20100416T120000 DTSTAMP:20130525T020110 LOCATION:Building 223 Room S-105, Argonne National Laboratory DESCRIPTION:Magnetization dynamics in nanoscale patterned multilayer systems has been the subject of very active research over the past decade. This is because new and interesting phenomena occur and are observable in nanoscale magnetic systems, as a result of size confinement and proximity effects. Some of the key areas addressed by recent research are vortex dynamics and the nature and dispersion of spin wave eigenmodes, and what controls them, as well as the effect of spin torque. \n\nI will here present results on micromagnetic modeling of vortex dynamics and low-lying spin wave eigenmodes in circularly exchange biased bi- and trilayer discs. In these systems, the equilibrium magnetization is in a vortex configuration, and one layer is in contact with an antiferromagnet with the exchange bias set with the ferromagnet in a vortex configuration. The dynamics can be controlled by the exchange bias strength and by the interlayer exchange coupling. \n\nI will also discuss spin torque effects, in which a spin-polarized current exerts a torque on the magnetic layers. It is known that the spin torque contains two components, one in-plane and one perpendicular component. Both are important in magnetic tunnel junctions, but the perpendicular component has been notoriously difficult to measure. The modeling that I have carried out shows that the strength of the so-called perpendicular spin torque affects the lowest spin wave eigenmode, and thus enables simple experimental determination of the value of the perpendicular spin torque. \n\nI will show some very recent experimental results on magnetic tunnel junctions about 50 nm in lateral size in which the lowest ferromagnetic resonance frequency is shifted by bias voltage and how this shift is directly proportional to the perpendicular spin torque component. SUMMARY:Magnetization dynamics and spin torque effect in nanoscale magnetic bi-and trilayer systems UID:843 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100401T120000 DTEND;TZID=America/Chicago:20100401T130000 DTSTAMP:20130525T020110 LOCATION:Cafeteria, Private Dining Rooms A&B, Argonne National Laboratory DESCRIPTION:Only given the url below. SUMMARY:Cosmic Velocity Flows in the Large Scale with Sloan Data Release 7 Early-Type Galaxies UID:826 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100407T103000 DTEND;TZID=America/Chicago:20100407T113000 DTSTAMP:20130525T020110 LOCATION:Building 240 Room 4301, Argonne National Laboratory DESCRIPTION:Over the last few decades coupled cluster (CC) theory has evolved into a basic formalism and has been used to describe a gamut of many-electron systems. \nThis can be attributed to three major factors: \n(1) the development of new CC methods capable of providing highly-accurate description of correlation effects in \nthree basic domains of CC theory: ground- and excited-state methods and linear response theory \n(2) the development of highly scalable CC codes capable of \nbeing executed across large numbers of CPUs\n(3) the development of multiscale CC-based approaches such as QM/MM methods or Embedded Cluster approaches where the CC method plays an essential role in describing the so-called quantum region (which is described by the first-principle approaches in the multiscale framework). \n\n These combined approaches have led to the emergence of new application domains for CC methods including the \ncalculations for electronic structure of biologically relevant systems and excited-state calculations for surface localized states. Since the highly-accurate CC formalisms will define a next generation of quantum chemistry imulations,in this talk we will discuss two newly developed formalisms - ground-state coupled-cluster methods based on the use of generating functional and non-iterative Equation-of-Motion CC methods, which significantly improve the balance between the ground- and excited-state correlation effects. SUMMARY:New Theoretical Models for High Precision Coupled Cluster Calculations UID:828 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100421T103000 DTEND;TZID=America/Chicago:20100421T113000 DTSTAMP:20130525T020110 LOCATION: Bldg 240 Rm: 4301, Argonne National Laboratory DESCRIPTION:Abstract:\nDuring the past few years we have witnessed dramatic advances in DNA sequencing and mapping technologies. These technologies generate data orders of magnitude faster, and at just a fraction of the costs previously possible. As a result, DNA sequencing is rapidly becoming a critical tool in many areas of biology research. At the same time, however, the wealth of data being generated is rapidly challenging the capacity of the computational infrastructure available to researchers. In my talk I will discuss the ways in which these data are affecting metagenomic studies, both by creating significant computational challenges, and by forcing us to rethink the way in which data should be analyzed.\n SUMMARY:How next generation sequencing data are changing metagenomics UID:829 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100415T103000 DTEND;TZID=America/Chicago:20100415T113000 DTSTAMP:20130525T020110 LOCATION:Bldg. 240, Conference Room 4301, Argonne National Laboratory DESCRIPTION:Technological advances have enabled whole genome sequencing of organisms from a wide range of taxa and surveys of genome-wide genetic diversity for microbial, viral and eukaryotic populations. A major challenge is the development of computational methods for analysis of these novel large-scale data sets. Such methods promise to deliver a more detailed understanding of biological processes with relevance for medical, biotechnological, agricultural and environmental applications. I will present an overview of the group’s research, which focuses on the computational analysis of metagenomic and population-level genomic data sets. In the area of metagenomics we are working on methods for the sequence-composition based ‘binning’ or taxonomic assignment of metagenome sequence fragments. We are furthermore seeking to understand and detect the imprint of selection from population-level genomics data, with a particular focus on understanding (and predicting) the evolution of influenza A virus. \n SUMMARY:Learning how to read genomes UID:830 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100409T103000 DTEND;TZID=America/Chicago:20100409T113000 DTSTAMP:20130525T020110 LOCATION:Building 240 / Conference Room 4301, Argonne National Labortory DESCRIPTION:In this seminar, we report the design and implementation of components in the domain of quantum chemistry. Specifically, components interfacing quantum mechanics/effective fragment potential (QM/EFP) method with any Common Component Architecture (CCA) aware package were developed. EFP is an electronic structure theory based, semi-classical potential to treat inter-molecular interactions. QM/EFP is an efficient and rigorous method suitable for modeling active (QM) and inactive (EFP) regions of large molecular systems (enzymes, proteins, nanomaterials in solution, heterogeneous gaseous molecular clusters) in a balanced manner. Developed components were tested by constructing potential energy surface of the indole-benzene sandwich model, a prototype for modeling protein interactions.\nOur current efforts are to design and implement components to interface efficient molecular dynamics code (LAMMPS, NAMD) within the CCA framework. Together with the developed EFP components, this will enable realistic and predictive modeling of the excited state dynamics of photochemical reactions in the atmosphere or in a biological environment.\n\nAnother ongoing project to extend capabilities of GDDI – a portable, low-weight communication library used by GAMESS quantum chemistry package – will be briefly discussed. In particular, a key-value database (associative array) feature is being implemented. The in-memory database provides a unified way to communicate between sub-programs. An access to a parallel I/O library (e.g. pNetCDF) can be provided through the same interface, as a disk-backed database.\n SUMMARY:High Performance Computing and Molecular Simulations Plug and Play is the Way Forward! UID:832 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100407T140000 DTEND;TZID=America/Chicago:20100407T150000 DTSTAMP:20130525T020110 LOCATION:Building 240 Room 1406-1407, Argonne National Laboratory DESCRIPTION:Computing has been an enormous accelerator to science and industry alike and it has led to an information explosion in many different fields. The unprecedented volume of data acquired by sensors, derived by simulations and analysis processes, and shared on the Web opens up new opportunities, but it also creates many challenges when it comes to managing and analyzing these data. \nIn this talk, I discuss the importance of maintaining detailed provenance also referred to as lineage and pedigree for digital data. Provenance provides important documentation that is key to preserve data, to determine the data\'s quality and authorship, to understand, reproduce, as well as validate results. Besides presenting techniques we have developed to efficiently manage and re-use provenance information, I will give an overview of the provenance infrastructure we have built for the open-source VisTrails system. I will also describe emerging applications and novel uses of provenance for enabling collaborative data analysis, teaching science, and publishing reproducible results.\n SUMMARY:Managing Provenance for Reproducibility and Beyond UID:833 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100407T100000 DTEND;TZID=America/Chicago:20100407T110000 DTSTAMP:20130525T020110 LOCATION:Building 240 Room 1406-1407, Argonne National Laboratory DESCRIPTION:Future advances in science depend on our ability to comprehend the vast amounts of data being produced and acquired, and scientific visualization is a key enabling technology in this endeavor. We posit that visualization should be better integrated with the data exploration process instead of being done after the fact - when all\nthe science is done - simply to generate presentations of the findings. An important barrier to a wider adoption of visualization is complexity: the design of effective visualizations is a complex, multistage process that requires deep understanding of existing techniques, and how they relate to human cognition. We envision visualization software tools evolving into scientific discovery environments that support the creative tasks in the discovery pipeline, from data acquisition and simulation to hypothesis testing and evaluation, and that enable the publication of results that can be reproduced and verified.\n\nIn this talk, we will cover our recent work on the development of visualization tools and techniques for scientific discovery. We will use our recent collaborations with scientists in oceanography, neuroscience, and high-energy physics as a backdrop for motivating our\nwork. SUMMARY:Visualization Tools and Techniques for Scientific Discovery UID:835 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100512T150000 DTEND;TZID=America/Chicago:20100512T160000 DTSTAMP:20130525T020110 LOCATION:Building 240, Rm 1404-1405, Argonne National Laboratory DESCRIPTION:With the ever-increasing numbers of cores per node in high-performance computing systems, a growing number of applications are using threads to exploit shared memory within a node and MPI across nodes. This hybrid programming model needs efficient support for multithreaded MPI communication. We describe the optimization of one aspect of a multithreaded MPI implementation: concurrent accesses from multiple threads to various MPI objects, such as communicators, datatypes, and requests. The semantics of the creation, usage, and destruction of these objects implies, but does not strictly require, the use of reference counting to prevent memory leaks and premature object destruction. We demonstrate how a naive multithreaded implementation of MPI object management via reference counting incurs a significant performance penalty. We then detail two solutions that we have implemented in MPICH2 to mitigate this problem almost entirely, including one based on a novel garbage collection scheme. In our performance experiments, this new scheme improved the multithreaded messaging rate by as much as 31% over the naive reference counting method. SUMMARY:MPI Object Lifetimes: Semantics and Solutions for Multithreading UID:837 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100526T150000 DTEND;TZID=America/Chicago:20100526T160000 DTSTAMP:20130525T020110 LOCATION:Bldg 240, Conference Center (1406-1407), Argonne National Laboratory DESCRIPTION:I will discuss our recent work on calculation of ion density profiles in confined geometries using the classical density functional theory framework. Essentially, the problem here is that of minimization of a highly nonlinear functional that describes the (near) equilibrium profiles of charged hard sphere liquids. This is a computationally intensive problem integro differential problem that is poorly amenable to treatment with Newton\'s method. In particular, the most computationally-expensive step results from an application of a \"pseudo-convolution\" operator within a continuation loop. In order to accelerate this computation we devised a spectral quadrature approach, which is well-suited for GPU acceleration and doesn\'t suffer from the use of single-precision arithmetic. Coupling our original PETSc DFT code to PyCUDA resulted in a substantial acceleration of the overall calculation and enabled substantially more interesting computations.\n\nI will outline the DFT model of ionic channel permeation, outline our computational approach to it and discuss the CUDA-based results and their incorporation into the PETSc framework.\n\nThis is joint work with Matthew Knepley (UofC), Peter Brune (UofC) and Dirk Gillespie (Rush). SUMMARY:Recent computational results on classical DFT of electrolytes: PETSc + PyCUDA UID:838 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100623T150000 DTEND;TZID=America/Chicago:20100623T160000 DTSTAMP:20130525T020110 LOCATION:Bldg 240 Conference Center 1404-1405, Argonne National Laboratory DESCRIPTION:In the late 1990\'s, astrophysics made the shocking discovery that the\nexpansion rate of the Universe is increasing with time. \"Dark energy,\"\nwhich experiences gravity as a repulsive force, is thought to explain this\ncosmic acceleration. Several independent observational methods provide\nevidence pointing to the conclusion that 95% of the current energy density\nof the Universe exists in the form of the \"dark sector\" that is different\nfrom normal, i.e., \"baryonic,\" matter. In particular, the dark sector is\nthought to be comprised of 22% \"dark matter\" that interacts with baryonic\nmatter ostensibly only through gravity, and 73% dark energy. The evidence\nfor the dark sector represents one of the greatest mysteries of modern\nscience in that very little is known about the fundamental nature of the\ndark sector components. It is presently an exciting time to be involved in\ncosmology because planned astronomical surveys will effectively result in\nobservational probes of the dark sector becoming systematics-limited,\nmaking numerical simulations crucial to formulating precision cosmological\nconstraints. Topics to be discussed include 1) an introduction to\ncosmology and the dark sector, 2) a overview of the next generation Dark\nEnergy Survey observational astronomy project, and 3) two high performance\ncomputing simulation efforts aimed at elucidating the dark sector that are\nbeing pursued on the Intrepid supercomputer at Argonne.\n SUMMARY:Intrepid Astrophysics and Cosmology Simulations UID:839 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100721T150000 DTEND;TZID=America/Chicago:20100721T160000 DTSTAMP:20130525T020110 LOCATION:Bldg 240 Rm 4301, Argonne National Laboratory DESCRIPTION:We present a fast multiscale approach for the network minimum logarithmic arrangement problem. This type of arrangement plays an important role in a network compression and fast node/link access operations. The algorithm is of linear complexity and exhibits good scalability which makes it practical and attractive for using on large-scale instances. Its effectiveness is demonstrated on a large set of real-life networks. These networks with corresponding best-known minimization results are suggested as an open benchmark for a research community to evaluate new methods for this problem. SUMMARY:Network compression-friendly ordering UID:840 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100412T103000 DTEND;TZID=America/Chicago:20100412T113000 DTSTAMP:20130525T020110 LOCATION:Building 240 Room 1405, Argonne National Laboratory DESCRIPTION:Substantial changes in population size, age structure, and urbanization are expected in many parts of the world this century. Although such changes can affect energy use and greenhouse gas emissions, emissions scenario analyses have either left them out or treated them in a fragmentary or overly simplified manner. We carry out the first comprehensive assessment of the implications of demographic change for global emissions of carbon dioxide. Using a new energy-economic growth model that accounts for a range of demographic dynamics, we show that slowing population growth could provide 16-29% of the emissions reductions suggested to be necessary by 2050 to avoid dangerous climate change. We\nalso find that aging and urbanization can substantially influence emissions in particular world regions. SUMMARY:Demographic impacts on future carbon emissions: A global assessment UID:842 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100419T103000 DTEND;TZID=America/Chicago:20100419T113000 DTSTAMP:20130525T020110 LOCATION:Bldg. 203 Room E-142, Argonne National Laboratory DESCRIPTION:Explaining species distribution using local environmental features is a long standing ecological problem. Often, available data is collected as a set of presence locations only thus precluding the possibility of a presence-absence analysis. We propose that it is natural to view presence-only data for a region as a point pattern over that region and to use local environmental features to explain the intensity driving this point pattern. This suggests hierarchical modeling, treating the presence data as a realization of a spatial point process whose intensity is governed by environmental covariates. Spatial dependence in the intensity surface is modeled with random effects involving a zero mean Gaussian process. Highly variable and typically sparse sampling effort as well as land transformation degrades the point pattern so we augment the model to capture these effects. The Cape Floristic Region (CFR) in South Africa provides a rich class with such species data. The potential, i.e., nondegraded presence surfaces over the entire area are of interest from a conservation and policy perspective. Our model assumes grid cell homogeneity of the intensity process where the region is divided into _ 37,000 grid cells. To work with a Gaussian process over a very large number of cells we use predictive process approximation. Bias correction by adding a heteroscedastic error component is implemented. The model was run for a number of different species. Comparison is made with the now popular Maxent approach, though the latter is much more limited with regard to inference. Additional inference such as investigation of species richness immediately follows from our modeling framework. SUMMARY:Point Pattern Modeling for Degraded Presence-Only Data over Large Regions UID:849 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100415T130000 DTEND;TZID=America/Chicago:20100415T140000 DTSTAMP:20130525T020110 LOCATION:Building 240 Seminar Room 4301, Argonne National Laboratory DESCRIPTION:Victor will explore two topics of industrial interest. First, he will discuss challenges arising in the implementation of large-scale optimization models in highly dynamic environments. In particular, Victor will demonstrate that there exist complex interactions between the system performance, solution time and accuracy, sampling frequency, and the forecasting capabilities of the model. Victor will use this insight to motivate several strategies to achieve faster solutions. In the second part of the talk, Victor will explore further the impact of forecasts, but this time also from an economic perspective. Victor will explain how these topics can be brought together to handle several operational challenges arising in the implementation of the next-generation power grid. SUMMARY:Real-Time Optimization, Feedback, and Forecasts UID:846 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100416T103000 DTEND;TZID=America/Chicago:20100416T113000 DTSTAMP:20130525T020110 LOCATION:Building 240 / Conference Room 4301, Argonne National Labortory DESCRIPTION:The forward-backward semiclassical dynamics (FBSD) method alleviates the notorious quantum dynamical sign problem through pre-cancellation of the forward and backward classical action integrals; the resulting quasiclassical scheme has been applied successfully to condensed phase systems as quantum mechanical as superfluid helium. Recent developments in FBSD have accelerated convergence with respect to the number of imaginary time steps; applications to model anharmonic systems and Lennard-Jones fluids will be discussed.\n\nBailey\'s generalization of Hamilton\'s principle of stationary action has been used to solve Bohm’s hydrodynamic equations of motion in the forward-backward quantum dynamics (FBQD) formulation of time correlation functions, yielding the first reported stable, synthetic calculation of quantum trajectories for the one-dimensional quartic oscillator at zero and low temperatures. Interestingly, accuracy and stability vary more with asymmetry than anharmonicity.\n\nFBSD references\nN. Makri, A. Nakayama and N. Wright, J. Theor. Comp. Chem. 3, 391-417 (2004).\nA. Nakayama and N. Makri, Proc. Nat. Acad. Sci. U.S.A. 102, 4230-4234 (2005).\nJ. Chen and N. Makri, Mol. Phys. 106, 443-453 (2008).\n\nFBQD references\nN. Makri, J. Phys. Chem. 108, 806-812 (2004).\nJ. Chen and N. Makri, J. Chem. Phys. 131, 124107 (2009).\nJ. Chen and N. Makri, Chem. Phys. (2010), in press.\n SUMMARY:Forward-Backward Semiclassical and Quantum Dynamics UID:847 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100428T150000 DTEND;TZID=America/Chicago:20100428T160000 DTSTAMP:20130525T020110 LOCATION:Bldg 240 Conference Center 1404-1405, Argonne National Laboratory DESCRIPTION:In biology, the ancestral relations among a set of species with a common ancestor are described by a phylogenetic tree (also known as a phylogeny). Given a set of species that are described in terms of their characteristics (in matrix form), the Perfect Phylogeny problem (PP) asks whether there is a phylogeny that is consistent with respect to these characteristics. In graph terms, this question asks whether a given graph can be made chordal (or \"triangulated\") by adding edges under certain constraints.\n\nA problem that is perhaps more familiar to LANS is the Acyclic Coloring problem (AC), which models the efficient computation of sparse Hessian matrices using substitution-based methods. Both of these problems are NP-complete in the general case, and remain so even under severe restrictions.\n\nIn this talk, I will demonstrate positive algorithmic results for both the PP and AC problems by means of a deep connection involving the tree structure of chordal graphs and the powerful notion of treewidth. These results include constructive, polynomial-time algorithms when the input graph belongs to the relatively large class of weakly chordal graphs, which generalizes a number of known results.\n\nThough I will assume some familiarity with basic graph-theoretic concepts, I will not assume any knowledge of evolutionary biology. SUMMARY:What does efficient Hessian computation have to do with inferring evolutionary history? UID:852 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100517T143000 DTEND;TZID=America/Chicago:20100517T163000 DTSTAMP:20130525T020110 LOCATION:TCS Building 240 - Rooms 1406/1407, Argonne National Laboratory DESCRIPTION:Over one billion processing hours are available through DOE’s INCITE program for 2011. Our Proposal Writing Webinar can help you stake your claim. Remember: INCITE 2011 proposals are due June 30, 2010. \n\nKatherine Riley, ALCF scientific applications engineer, and Bronson Messer of Oak Ridge’s Scientific Computing group, will provide tips and suggestions to improve the quality of your INCITE proposal submission.\n\nWhen you register, please let us know if you will be attending in person or via the online webinar function. Webinar logon instructions will be emailed separately. SUMMARY:INCITE Proposal Writing Webinar/Lecture UID:853 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100421T130000 DTEND;TZID=America/Chicago:20100421T140000 DTSTAMP:20130525T020110 LOCATION:Building 240 Seminar Room 4301, Argonne National Laboratory DESCRIPTION:In the first part of the talk, Prasanna will focus on automatic tuning of parameterized algorithms. Many high performing algorithms for computationally hard problems require the setting of a number of parameters to optimize their performance. The task of setting appropriate parameters is essentially a stochastic optimization problem that arises repeatedly in tuning heuristic algorithms and it is relevant for a variety of other contexts. While practitioners frequently tackle this task based on their experience and using rules of thumb, this task can effectively be automatized using automated tuning procedures. Prasanna will discuss our recent research on automated parameter tuning methods for off-line parameter tuning. In particular, Prasanna will present the recently developed Iterative F-race algorithm that tackles the tuning task from a machine learning perspective. \n\nIn the second part of the talk, Prasanna will focus on stochastic routing problems. In previous research on stochastic routing problems, most commonly analytical computation approach was used in iterative improvement algorithms and metaheuristics. This is particularly the case for the prototypical examples of stochastic routing problems, such as the probabilistic traveling salesman problem (PTSP). The alternative empirical estimation approach has never been thoroughly investigated. Prasanna will give an overview of our recent research efforts to engineer effective estimation-based iterative improvement algorithms and metaheuristics for the PTSP. Using experimental studies, Prasanna will show that the empirical estimation approach is actually a viable alternative even for the simple, prototypical stochastic routing problems and the estimation-based algorithms outperform by quite a substantial margin previously proposed state-of-the-art algorithms. SUMMARY:Heuristics for Stochastic Optimization: Case Studies in Parameter Tuning and Stochastic Routing Problems UID:851 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100426T103000 DTEND;TZID=America/Chicago:20100426T113000 DTSTAMP:20130525T020110 LOCATION:Building 240 / Conference Room 4301, Argonne National Labortory DESCRIPTION:The design of new materials with specific physical, chemical, or biological properties is a central goal of much research in pharmaceutical and materials science. However, due to the combinatorial nature of chemical compound space (stoichiometry space), brute-force computational screening of all possible compounds for interesting properties is beyond any current capacity. Consequently, when it comes to properties or systems that require first principles calculations, reliable optimization algorithms must not only trade-off sufficient accuracy and computational speed, but must also aim for rapid convergence in terms of number of compounds \"visited\". I will briefly discuss recent progress on two fronts: Accuracy of calculated properties and efficient sampling of compound space. Specifically, density functional theory (DFT) based estimates of interatomic two- and three-body van der Waals contributions to binding energies in gas and condensed phase systems will be presented. Thereafter, I will show how inclusion of nuclear quantum effects can qualitatively alter the classical free energy landscape of proton transfer reactions using DFT based path-integral ab initio molecular dynamics calculations of DNA base pair models at room temperature. Finally, a DFT approach for constructing high-dimensional yet analytical property gradients in chemical compound space will be discussed. I will conclude the presentation with an outlook on potential future design projects that crucially depend on the availability of high-performance computing capabilities.\n\n SUMMARY:From Density Functional Theory Based Calculations Towards the Design of New Materials UID:858 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100609T150000 DTEND;TZID=America/Chicago:20100609T160000 DTSTAMP:20130525T020110 LOCATION:Bldg 240 Conference Center 1404-1405, Argonne National Laboratory DESCRIPTION:While data analysis and visualization have contributed to scientific discoveries since the 1600s, what we think of today as modern scientific visualization dates back to the late 1980s. Over the past 20 years or so, computer scientists have developed fundamental algorithms that produce meaningful analysis of computed datasets, but as computations grow in size and complexity, the resulting datasets are beginning to overwhelm the way that we are accustomed to processing them. Finding scalable solutions to data analysis is a challenging problem now, one that will only grow as we advance toward exascale computing. In particular, the high cost of moving and accessing data motivates performing analysis in some unconventional places in HPC systems. These include computational nodes, I/O nodes, and the storage system, alongside the conventional data analysis cluster nodes. This talk will examine the feasibility of performing parallel analysis directly on the same compute nodes used to generate the simulation, and will draw upon examples in parallel volume rendering, parallel image compositing, and parallel particle tracing to analyze the scalability of these approaches. SUMMARY:Scalable Data Analysis UID:861 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100505T140000 DTEND;TZID=America/Chicago:20100505T150000 DTSTAMP:20130525T020110 LOCATION:Building 240, Conference Room 4301, Argonne National Laboratory DESCRIPTION:The MPI-2 specification introduced Remote Memory Access (RMA) semantics in the late 1990s. RMA semantics hold the promise effectively leveraging Remote Direct Memory Access (RDMA) feature provided by advanced network adapters and helping scientific applications achieve good communication and computation overlap. However, only a relatively small number of real applications currently take advantage of RMA to overlap communication latency. While some researchers have raised questions about the usability of this interface, others have questioned the performance and scalability offered by RMA operations.\n\nIn this work, we demonstrate how a real NSF TeraGrid application, AWM-Olsen (recently renamed to AWM-ODC), can be modified to expose computation and communication overlap. This application runs on tens of thousands of cores and consumes several million CPU hours on the TeraGrid Clusters every year. Some of the most detailed simulations to date of earthquakes along the San Andreas fault were carried out using this code, including the well-known TeraShake, SCEC ShakeOut simulations. A significant portion of its run-time (37% in a 4K process run), is spent in MPI communication routines. Using our modified AWM-ODC application and leveraging MPI-2 Active Target Synchronization semantics, the performance of the application can be boosted by 12% for 4K processors and 10% at 8K processors.\n\nBIO: Dr. Sayantan Sur is a Research Scientist at the Department of Computer Science at The Ohio State University. His research interests include high speed interconnection networks, high performance computing, fault tolerance and parallel computer architecture. He has published more than 15 papers in major conferences and journals related to these research areas. He is a member of the Network-Based Computing Laboratory lead by Dr. D. K. Panda. He is currently collaborating with National Laboratories, Supercomputer Centers, and leading InfiniBand companies on designing various subsystems of next generation high performance computing platforms. He has contributed significantly to the MVAPICH/MVAPICH2 (High Performance MPI over InfiniBand and 10GigE/iWARP) open-source software packages. The software developed as a part of this effort is currently used by over 1,110 organizations in 56 countries. In the past, he has held the position of Post-doctoral researcher at IBM T. J. Watson Research Center, Hawthorne and Member Technical Staff at Sun Microsystems. Dr. Sur received his Ph.D. degree from The Ohio State University in 2007.\n SUMMARY:Can MPI-2 RMA Operations Benefit Real World Applications? UID:864 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100510T103000 DTEND;TZID=America/Chicago:20100510T113000 DTSTAMP:20130525T020110 LOCATION:Building 240 Seminar Room 4301, Argonne National Laboratory DESCRIPTION:In the first part of the talk, Prasanna will focus on automatic tuning of parameterized algorithms. Many high performing algorithms for computationally hard problems require the setting of a number of parameters to optimize their performance. The task of setting appropriate parameters is essentially a stochastic optimization problem that arises repeatedly in tuning heuristic algorithms and it is relevant for a variety of other contexts. While practitioners frequently tackle this task based on their experience and using rules of thumb, this task can effectively be automatized using automated tuning procedures. Prasanna will discuss his recent research on automated parameter tuning methods for off-line parameter tuning. In particular, he will present the recently developed Iterative F-race algorithm that tackles the tuning task from a machine learning perspective. \n\nIn the second part of the talk, Prasanna will focus on stochastic routing problems. In previous research on stochastic routing problems, most commonly analytical computation approach was used in iterative improvement algorithms and metaheuristics. This is particularly the case for the prototypical examples of stochastic routing problems, such as the probabilistic traveling salesman problem (PTSP). The alternative empirical estimation approach has never been thoroughly investigated. Prasanna will give an overview of his recent research efforts to engineer effective estimation-based iterative improvement algorithms and metaheuristics for the PTSP. Using experimental studies, he will show that the empirical estimation approach is actually a viable alternative even for the simple, prototypical stochastic routing problems and the estimation-based algorithms outperform by quite a substantial margin previously proposed state-of-the-art algorithms. SUMMARY:Heuristics for Stochastic Optimization: Case Studies in Parameter Tuning and Stochastic Routing Problems UID:865 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100511T103000 DTEND;TZID=America/Chicago:20100511T113000 DTSTAMP:20130525T020110 LOCATION:Bldg: 240, Conference Center Rm: 1406 & 1407, Argonne National Laboratory DESCRIPTION:There is much hype about leveraging 3.5 billion years of evolution to solve pressing problems concerning environment and energy. The basic strategy is to design novel (unnatural) biochemical capabilities for environmental cleanup and energy production by recombining and rationally re-engineering biological circuits from diverse organisms. In principle, it is possible to decipher and re-engineer the complex information processing circuits that dynamically reconfigure the physiology of a particular organism in response to environmental change. In practice, this would require extensive mining of systems measurements for conditional relationships among patterns of changes in gene expression (mRNA, protein and ncRNA), interactions (P-P and P-D), modifications (protein and DNA), and metabolism (metabolite levels, enzyme activities). For synthetic biology applications, it is essential to capture relevant conditional relationships both at a systems level and at a sufficiently high resolution to mechanistically describe and predict how environmental change influences the execution of these cellular algorithms at multiple scales. Clearly, the experimental and computational tools necessary for doing this type of multi-scale network inference and modeling are diverse, disjointed, and constantly changing for a purely monolithic knowledge-base solution to be practical. A knowledge-base that is built upon an architecture of loosely coupled resources, on the other hand, is both highly adaptable to change and essential for large collaborative and interdisciplinary systems biology efforts for rapid inference and re-engineering of biological circuits. SUMMARY:A knowledgebase for rapid inference and re-engineering of biological circuits UID:866 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100518T100000 DTEND;TZID=America/Chicago:20100518T110000 DTSTAMP:20130525T020110 LOCATION:Building 240, TCS Conference Center (1404 & 1405), Argonne National Laboratory DESCRIPTION:This talk will discuss black-box problem diagnosis in parallel file systems, in which we focus on automatically diagnosing different performance problems in parallel file systems by identifying, gathering and analyzing OS-level, black-box performance metrics on every node in the cluster. Our peer-comparison diagnosis approach compares the statistical attributes of these metrics across I/O servers, to identify the faulty node. We develop a root-cause analysis procedure that further analyzes the affected metrics to pinpoint the faulty resource (storage or network), and demonstrate that this approach works commonly across stripe-based parallel file systems. We demonstrate our approach for realistic storage and network problems injected into three different file-system benchmarks (dd, IOzone, and PostMark), in both PVFS and Lustre clusters. SUMMARY:Black-Box Problem Diagnosis in Parallel File Systems UID:868 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100513T150000 DTEND;TZID=America/Chicago:20100513T160000 DTSTAMP:20130525T020110 LOCATION:Building 240 / Conference Room 4301, Argonne National Labortory DESCRIPTION:In 2011 IBM will install the Blue Waters computing system with over 300,000 cores and 10 petaflops peak performance in Urbana, Illinois. One of the first programs to run on the machine will be the molecular dynamics code NAMD, which was specified as a target petascale application by the NSF due to its dual status as both a highly scalable parallel code and a practical NIH-funded scientific tool with thousands of users.\n\nNAMD is based on the prioritized message-driven execution capabilities of the Charm++ parallel runtime system. NAMD and Charm++ have evolved through a close collaboration between computer science researchers working to make parallel programming easier and computational scientists seeking flexible, reliable, and high-performance biomolecular simulation tools. In this lecture I will relate lessons learned from my career at the nexus of this unique and highly successful collaboration.\n SUMMARY:Lessons from 15 Years of NAMD Development with Charm++ and Message-Driven Execution UID:869 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100513T150000 DTEND;TZID=America/Chicago:20100513T160000 DTSTAMP:20130525T020110 LOCATION:Building 240 / Conference Room 4301, Argonne National Labortory DESCRIPTION:In 2011 IBM will install the Blue Waters computing system with over 300,000 cores and 10 petaflops peak performance in Urbana, Illinois. One of the first programs to run on the machine will be the molecular dynamics code NAMD, which was specified as a target petascale application by the NSF due to its dual status as both a highly scalable parallel code and a practical NIH-funded scientific tool with thousands of users.\n\nNAMD is based on the prioritized message-driven execution capabilities of the Charm++ parallel runtime system. NAMD and Charm++ have evolved through a close collaboration between computer science researchers working to make parallel programming easier and computational scientists seeking flexible, reliable, and high-performance biomolecular simulation tools. In this lecture I will relate lessons learned from my career at the nexus of this unique and highly successful collaboration.\n SUMMARY:Lessons from 15 Years of NAMD Development with Charm++ and Message-Driven Execution UID:870 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100514T103000 DTEND;TZID=America/Chicago:20100514T113000 DTSTAMP:20130525T020110 LOCATION:Building 240 / Conference Room 4301, Argonne National Labortory DESCRIPTION:Science is increasingly relying on computation. This trend has been steadily gaining momentum with\nadvancement of faster and affordable compute hardware and is further boosted by its wider availability\nand accessibility to researchers. Computational simulations are becoming an important aspect of\nresearch in virtually every discipline. This has led to a rapid growth of scientific data and is expected to\ncontinue expanding. Among several significant challenges involved with scientific data a chief one is\nmaking sense out of it. Unfortunately, there has not been a corresponding level of development to\nanalyze the deluge of data; moreover our ability to understand data is no better than earlier.\nOne of the untapped potential to understand the data is through scientific visualization. Our visual\nsystem is highly developed and could be harnessed to gain insight by encoding data into visual form.\nThis is the key objective of scientific visualization, which aims to provide an intuitive way to investigate\nand analyze data. Often a visual representation has the potential to make the results widely accessible\nto a broad range of people. Based on his work, Amit will broadly discusses the role scientific\nvisualizations have played in several disciplines including astrophysics, biology, climate modeling,\ncomputational fluid dynamics, molecular dynamics, nanotech, structural engineering and seismology.\nThe presentation will briefly describe problems, bottlenecks on visualization front and will also feature\nfew animations.\n\nBiography: Amit Chourasia leads the Visualization Services group at the San Diego Supercomputer\nCenter. His work has been focused on leading the research, development and application of software\ntools for scientific visualization; for data typically generated by massively large computer simulations in\nvarious fields of science and engineering. Key portion of his work is to find ways to represent data in a\nvisual form that is clear, succinct and accurate (a challenging yet very exciting endeavor). He believes\nthat with the help of visualization, domain scientists can gain significant insights about their data, pose\nrelevant questions of their research, share results in addition to visual validation of their data.\nAmit\'s application and research interests are in area of animation, computer graphics, visualization and\nvisual perception. He received a Master\'s degree in Computers Graphics Technology from Purdue\nUniversity, West Lafayette and a Baccalaureate degree in Architecture (Honors) from Indian Institute of\nTechnology, Kharagpur.\n\nClick the link to add the event to your calendar. SUMMARY:Visual Analysis UID:873 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100512T103000 DTEND;TZID=America/Chicago:20100512T113000 DTSTAMP:20130525T020110 LOCATION:Building 240 / Conference Room 4301, Argonne National Labortory DESCRIPTION:Dynamic migration of virtual machines on a cluster of physical machines is designed to maximize resource utilization by balancing loads across the cluster. When the utilization of a physical machine is beyond a fixed threshold, the machine is deemed overloaded. A virtual machine is then selected within the overloaded physical machine for migration to a lightly loaded physical machine. Key to such threshold-based VM migration is to determine when to move which VM to what physical machine, since wrong or inadequate decisions can cause unnecessary migrations that would adversely affect the overall performance. We present a learning framework that autonomously finds and adjusts thresholds at runtime for different computing requirements. Central to our approach is the previous history of migrations and their effects before and after each migration in terms of standard deviation of utilization. We set up an experimental environment that consists of extensive real world benchmarking problems and a cluster of 16 physical machines each of which has on average eight virtual machines. We demonstrate through experimental results that our approach autonomously finds thresholds close to the optimal ones for different computing scenarios and that such varying thresholds yield an optimal number of VM migrations for maximizing resource utilization. SUMMARY:Autonomous Migration of VMs for Maximizing Resource Utilization UID:871 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100521T140000 DTEND;TZID=America/Chicago:20100521T150000 DTSTAMP:20130525T020110 LOCATION:Builing 240 conference Room 4301, Argonne National Laboratory DESCRIPTION:Controlled thermonuclear fusion is a topic of prime interest as a source of sustainable energy, which could possibly compete with classical fission reactors in terms of efficiency and as a key process to produce energy. In the forthcoming years, the main challenge for the fusion community will be to develop experimental scenarios for ITER (« International Thermonuclear Experimental Reactor », the largest fusion process ever built and with a first run planned for 2018). Amongst the key issues, the main control challenges are related to the plasma shape control, advanced equilibrium profiles for Tokamaks and the stabilization of magnetohydrodynamic (MHD) modes. Shape control has been studied extensively but many results still have to be discovered on the other two topics, where the nonhomogeneous transport of waves, energy and particles appear as fundamental.\n\nAfter a general overview on fusion and the key issues from the control point of view, the proposed talk will detail some recent advances on current profiles control carried in Tore Supra (CEA Cadarache, South of France). Indeed, a particular interest is given to the current density and the way to produce plasma current. Due to the intrinsic limitation on magnetic flux availability in fusion processes to maintain a purely inductive current, the use of non-inductive sources to generate most of the current is inevitable. Modelling and real-time control of radiofrequency antennas (current source distributed in the plasma) are of prime importance to optimize the confinement and to ensure the profiles robustness with respect to external perturbations.\n\nAnother advanced problem for control will also be considered: the stabilization of MHD modes. MHD phenomena have several impacts on plasma; one of them is to generate unstable modes, such as those studied in the Reversed Field Pinch EXTRAP-T2 (Alfvčn lab, KTH, Sweden). These modes are of capital importance in fusion reactors, as a lack of their stabilization leads to the loss of confinement. Their time constant is typically a few milliseconds, with a control cycle of 100 μs for the simplest feedback rules. The complexity of the dynamics and the real-time constraints motivate new approaches for modelling and control, with an emphasis on time-delays, as discussed in this talk.\n\nShort bio:\nEmmanuel Witrant is an Associate Professor at University Joseph Fourier in the Automatic Control Department of GIPSA-lab, Grenoble, France. He obtained the B.Sc. in Aerospace Engineering from Georgia Institute of Technology in 2001 and the Ph.D. in Automatic Control from Grenoble Institute of Technology in 2005. His research interests are focussed on the non-homogeneous transport phenomena (information, energy, gases...) and the use of time-delay or optimization approaches to tackle the modeling and control of large-scale systems with complex dynamics. This research is applied to networked control systems (control with communication constraints), large-scale instruments (tokamak and reverse field pinch, Large Hadron Collider) and process ventilation control (mining industry and intelligent buildings). Dr. Witrant is also the head of the international master on Industrial Processes Automation in Grenoble and has a particular interest for curriculum design in systems and information technologies.\n SUMMARY:Profiles Control and Stability in Thermonuclear Fusion: Some Issues for ITER UID:875 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100521T103000 DTEND;TZID=America/Chicago:20100521T113000 DTSTAMP:20130525T020110 LOCATION:Building 240 Conference Room 2172, Argonne National Laboratory DESCRIPTION:State estimation techniques are used in weather and climate prediction, hydrogeology, seismology, as a way to blend model output and real data in order to improve on predictions from the exclusive use of the model or the data alone.\n\nTechniques that are based upon least-squares ideas, such as the family of Kalman Filter/Smoothers, or Variational Data Assimilation, are optimal in linear/Gaussian problems. However, they often fail in problems in which nonlinearities are important and/or when Gaussianity in the statistics cannot be assumed. Even linearization may fail, and so do ensemble techniques that make nonlinear predictions but rely on linear analyses. These comprise the practical state of the art, at least in weather forecasting and in hydrogeology. I will describe these as well as how failures arise in these methods.\n\nWe have created a number of nonlinear/non-Gaussian data assimilation techniques. Our present efforts are to make them computationally practical as well as to use of these to do problems that are otherwise intractable using conventional means.\n\nOne such application is in Lagrangian data assimilation: here we tackle the problem of blending data that has been sampled along paths, which when blended in traditional ways on Eulerian grids will lead to loss of critical features even though the estimates may be variance-minimizing. SUMMARY:Climate: When Data Fail Us, Nonlinear/Non-Gaussian Estimation UID:876 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100521T140000 DTEND;TZID=America/Chicago:20100521T150000 DTSTAMP:20130525T020110 LOCATION:Building 240 Seminar Room 4301, Argonne National Laboratory DESCRIPTION:Profiles Control and Stability in Thermonuclear Fusion: Some Issues for ITER\nEmmanuel Witrant (GIPSA-lab, University Joseph Fourier, Grenoble, France)\n\n\nControlled thermonuclear fusion is a topic of prime interest as a source of sustainable energy, which could possibly compete with classical fission reactors in terms of efficiency and as a key process to produce energy. In the forthcoming years, the main challenge for the fusion community will be to develop experimental scenarios for ITER (« International Thermonuclear Experimental Reactor », the largest fusion process ever built and with a first run planned for 2018). Amongst the key issues, the main control challenges are related to the plasma shape control, advanced equilibrium profiles for Tokamaks and the stabilization of magnetohydrodynamic (MHD) modes. Shape control has been studied extensively but many results still have to be discovered on the other two topics, where the nonhomogeneous transport of waves, energy and particles appear as fundamental.\n\nAfter a general overview on fusion and the key issues from the control point of view, the proposed talk will detail some recent advances on current profiles control carried in Tore Supra (CEA Cadarache, South of France). Indeed, a particular interest is given to the current density and the way to produce plasma current. Due to the intrinsic limitation on magnetic flux availability in fusion processes to maintain a purely inductive current, the use of non-inductive sources to generate most of the current is inevitable. Modelling and real-time control of radiofrequency antennas (current source distributed in the plasma) are of prime importance to optimize the confinement and to ensure the profiles robustness with respect to external perturbations.\n\nAnother advanced problem for control will also be considered: the stabilization of MHD modes. MHD phenomena have several impacts on plasma; one of them is to generate unstable modes, such as those studied in the Reversed Field Pinch EXTRAP-T2 (Alfvčn lab, KTH, Sweden). These modes are of capital importance in fusion reactors, as a lack of their stabilization leads to the loss of confinement. Their time constant is typically a few milliseconds, with a control cycle of 100 μs for the simplest feedback rules. The complexity of the dynamics and the real-time constraints motivate new approaches for modelling and control, with an emphasis on time-delays, as discussed in this talk.\n\n\nShort bio:\nEmmanuel Witrant is an Associate Professor at University Joseph Fourier in the Automatic Control Department of GIPSA-lab, Grenoble, France. He obtained the B.Sc. in Aerospace Engineering from Georgia Institute of Technology in 2001 and the Ph.D. in Automatic Control from Grenoble Institute of Technology in 2005. His research interests are focussed on the non-homogeneous transport phenomena (information, energy, gases...) and the use of time-delay or optimization approaches to tackle the modeling and control of large-scale systems with complex dynamics. This research is applied to networked control systems (control with communication constraints), large-scale instruments (tokamak and reverse field pinch, Large Hadron Collider) and process ventilation control (mining industry and intelligent buildings). Dr. Witrant is also the head of the international master on Industrial Processes Automation in Grenoble and has a particular interest for curriculum design in systems and information technologies.\n SUMMARY:Profiles Control and Stability in Thermonuclear Fusion: Some Issues for ITER UID:877 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100604T100000 DTEND;TZID=America/Chicago:20100604T110000 DTSTAMP:20130525T020110 LOCATION:Building 240 Conference Room 1407, Argonne National Laboratory DESCRIPTION:Abstract: There has been significant interest in the possibility that nonlinear threshold behavior could cause a catastrophic collapse of Arctic sea ice. An innovative dynamical systems model of Arctic sea ice has recently been proposed that suggests that ice self-insulation can prevent tipping point behavior in summer sea ice. Here a framework for analyzing and understanding cloud feedbacks using this dynamical systems model of Arctic sea ice is advanced. It is shown that cloud feedbacks strongly influence nonlinear threshold behavior of Arctic sea ice and that irreversible loss of summer sea ice is possible in some portion of the cloud feedback parameter space. Comparison with a global climate model and reanalysis data suggests that the cloud feedback regime in which nonlinear summer sea ice loss is possible may be relevant for future climate. SUMMARY:Cloud Feedbacks and Catastrophic Arctic Sea Ice Loss UID:878 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:00000000T103000 DTEND;TZID=America/Chicago:00000000T113000 DTSTAMP:20130525T020110 LOCATION:Building 240 / Conference Room 4301, Argonne National Laboratory DESCRIPTION:High-performance discrete Fourier transform (DFT) libraries are an important requirement on many computing platforms. The complexity of manually writing and tuning DFT libraries to target architectures is the motivation behind the SPIRAL project, which can automatically generate platform-adapted libraries.\n\nHowever, current techniques in SPIRAL don\'t work well for all target platforms. In particular, several existing and emerging target platforms incorporate a distributed memory parallel processing paradigm, where the cost of accessing non-local memories is relatively high. The paradigm has been long used in supercomputing environments, and is now also finding its way into desktop computing (e.g., IBM\'s Cell processor), GPGPUs, and embedded processors.\n\nThe goal of this work is to enable computer generation of high-performance DFT libraries for a wide range of distributed memory parallel processing systems, given only a high-level description of a DFT algorithm and some platform parameters. The main challenges include generating code for multiple target programming paradigms, delivering load balanced parallelization across multiple layers of the compute hierarchy, orchestrating explicit memory management to best exploit available bandwidth, and overlapping computation with communication.\n\nWe attack the problem by building algorithm components that describe parallelization, streaming, and data exchange in a domain-specific declarative mathematical language. We then rewrite existing DFT algorithms to expose these components, which are guaranteed to \"fit\" the target architecture and extract maximum performance.\n\nClick on this link to add the event to your calendar.\n SUMMARY:Computer Generation of Fourier Transform Libraries for Distributed Memory Computing UID:879 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:00000000T103000 DTEND;TZID=America/Chicago:00000000T113000 DTSTAMP:20130525T020110 LOCATION:Building 240 / Conference Room 4301, Argonne National Laboratory DESCRIPTION:High-performance discrete Fourier transform (DFT) libraries are an important requirement on many computing platforms. The complexity of manually writing and tuning DFT libraries to target architectures is the motivation behind the SPIRAL project, which can automatically generate platform-adapted libraries.\n\nHowever, current techniques in SPIRAL don\'t work well for all target platforms. In particular, several existing and emerging target platforms incorporate a distributed memory parallel processing paradigm, where the cost of accessing non-local memories is relatively high. The paradigm has been long used in supercomputing environments, and is now also finding its way into desktop computing (e.g., IBM\'s Cell processor), GPGPUs, and embedded processors.\n\nThe goal of this work is to enable computer generation of high-performance DFT libraries for a wide range of distributed memory parallel processing systems, given only a high-level description of a DFT algorithm and some platform parameters. The main challenges include generating code for multiple target programming paradigms, delivering load balanced parallelization across multiple layers of the compute hierarchy, orchestrating explicit memory management to best exploit available bandwidth, and overlapping computation with communication.\n\nWe attack the problem by building algorithm components that describe parallelization, streaming, and data exchange in a domain-specific declarative mathematical language. We then rewrite existing DFT algorithms to expose these components, which are guaranteed to \"fit\" the target architecture and extract maximum performance.\n\nClick on this link to add the event to your calendar.\n SUMMARY:Computer Generation of Fourier Transform Libraries for Distributed Memory Computing UID:880 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:00000000T103000 DTEND;TZID=America/Chicago:00000000T113000 DTSTAMP:20130525T020110 LOCATION:Building 240 / Conference Room 4301, Argonne National Laboratory DESCRIPTION:High-performance discrete Fourier transform (DFT) libraries are an important requirement on many computing platforms. The complexity of manually writing and tuning DFT libraries to target architectures is the motivation behind the SPIRAL project, which can automatically generate platform-adapted libraries.\n\nHowever, current techniques in SPIRAL don\'t work well for all target platforms. In particular, several existing and emerging target platforms incorporate a distributed memory parallel processing paradigm, where the cost of accessing non-local memories is relatively high. The paradigm has been long used in supercomputing environments, and is now also finding its way into desktop computing (e.g., IBM\'s Cell processor), GPGPUs, and embedded processors.\n\nThe goal of this work is to enable computer generation of high-performance DFT libraries for a wide range of distributed memory parallel processing systems, given only a high-level description of a DFT algorithm and some platform parameters. The main challenges include generating code for multiple target programming paradigms, delivering load balanced parallelization across multiple layers of the compute hierarchy, orchestrating explicit memory management to best exploit available bandwidth, and overlapping computation with communication.\n\nWe attack the problem by building algorithm components that describe parallelization, streaming, and data exchange in a domain-specific declarative mathematical language. We then rewrite existing DFT algorithms to expose these components, which are guaranteed to \"fit\" the target architecture and extract maximum performance.\n\nClick on this link to add the event to your calendar.\n SUMMARY:Computer Generation of Fourier Transform Libraries for Distributed Memory Computing UID:881 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100603T140000 DTEND;TZID=America/Chicago:20100603T150000 DTSTAMP:20130525T020110 LOCATION:Building 240 / Room 1172 (1C2), Argonne National Laboratory DESCRIPTION:High-performance discrete Fourier transform (DFT) libraries are an important requirement on many computing platforms. The complexity of manually writing and tuning DFT libraries to target architectures is the motivation behind the SPIRAL project, which can automatically generate platform-adapted libraries.\n\nHowever, current techniques in SPIRAL don\'t work well for all target platforms. In particular, several existing and emerging target platforms incorporate a distributed memory parallel processing paradigm, where the cost of accessing non-local memories is relatively high. The paradigm has been long used in supercomputing environments, and is now also finding its way into desktop computing (e.g., IBM\'s Cell processor), GPGPUs, and embedded processors.\n\nThe goal of this work is to enable computer generation of high-performance DFT libraries for a wide range of distributed memory parallel processing systems, given only a high-level description of a DFT algorithm and some platform parameters. The main challenges include generating code for multiple target programming paradigms, delivering load balanced parallelization across multiple layers of the compute hierarchy, orchestrating explicit memory management to best exploit available bandwidth, and overlapping computation with communication.\n\nWe attack the problem by building algorithm components that describe parallelization, streaming, and data exchange in a domain-specific declarative mathematical language. We then rewrite existing DFT algorithms to expose these components, which are guaranteed to \"fit\" the target architecture and extract maximum performance.\n\nClick on this link to add this event to your calendar.\n SUMMARY:Computer Generation of Fourier Transform Libraries for Distributed Memory Computing UID:882 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100625T103000 DTEND;TZID=America/Chicago:20100625T113000 DTSTAMP:20130525T020110 LOCATION:Building 240 / Conference Room 1404, Argonne National Laboratory DESCRIPTION:Brocade NDA Briefing including IBM DCN networking updates. (Please note this seminar is restricted to CELS/ALCF/MCS full-time employees only.)\n\nClick on this link to add this event to your calendar.\n\n\n SUMMARY:Brocade NDA Briefing UID:884 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100623T143000 DTEND;TZID=America/Chicago:20100623T153000 DTSTAMP:20130525T020110 LOCATION:Argonne National Lab, TCS/Room 5C2 (5172), University of Chicago, Searle 240A, 5735 S. Ellis Ave. DESCRIPTION:The publication of a series of Google technical reports describing the Google File System, MapReduce and BigTable and the development of the Hadoop system and related projects that provided an open source implementation of this technology changed the way that scientists computed with big data. Today, developing systems to manage and analyze big data is an active area of research.\n\nIn this talk, we give an overview of some of this work. The Sector/Sphere system is one of the systems being developed for working with big data. It consists of a parallel programming framework called Sphere that enables arbitrary user defined functions to be executed over the data managed by the Sector distributed storage system. In this talk, we describe Sector and Sphere and some of the lessons that we have learned over the past several years using Hadoop, Sector/Sphere and related systems when working with big data.\n\n\n\n\n\n SUMMARY:My Other Computer is a Data Center: The Sector Perspective on Big Data UID:885 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100607T093000 DTEND;TZID=America/Chicago:20100607T103000 DTSTAMP:20130525T020110 LOCATION:Building 240 Room 1407, Argonne National Laboratory DESCRIPTION:Tisha will introduce the Argonne Leadership Computing Facility (ALCF), its mission, its current systems, and systems expected to arrive at Argonne in the near future. The ALCF (http://www.alcf.anl.gov/) is a national leadership computing facility designed to provide resources that make computationally intensive projects of the largest scales possible. Currently the ALCF supports 36 INCITE computational science projects covering disciplines as diverse as protein folding to modeling aircraft engines.\n SUMMARY:Overview of the Argonne Leadership Computing Facility UID:886 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100608T093000 DTEND;TZID=America/Chicago:20100608T103000 DTSTAMP:20130525T020110 LOCATION:Building 240 Room 1406 & 1407, Argonne National Laboratory DESCRIPTION:A Programming model is the high-level structure within which an algorithm is implemented. The complexity of the programming model required for a given application depends greatly on the type of parallelism to be expressed. In the first part of this talk, Jeff will highlight some of the most common programming models (e.g. mast-worker) and describe their implementation within MPI. The second part will focus on asynchronous programming models using one-sided communication and the utility of this model within application codes in biology and chemistry. SUMMARY:Programming Models for High Performance Scientific Computing UID:887 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100616T153000 DTEND;TZID=America/Chicago:20100616T163000 DTSTAMP:20130525T020110 LOCATION:Bldg 240 Conference Center 1406-1407, Argonne National Laboratory DESCRIPTION:This talk will describe a new tool, ADIC2, for automatic di fferentiation (AD) of C and C++ code through source-to-source transformation. ADIC2 is the successor of the ADIC diff erentiation tool, which supports forward mode AD of C and a small subset of C++. ADIC2 was completely redesigned and re-implemented as part of the OpenAD software framework, resulting in a robust, flexible, and extensible tool for differentiating C and some features of C++, with plans for full support of C++ in the near future. We discuss some of the challenges in creating AD tools for C and C++ in general and describe the component approach employed in the design and implementation of ADIC2. In particular, the talk will describe ROSE - an open source compiler toolkit developed at Lawrence Livermore National Laboratory - and how it can be used to generate source to source transformation tools. SUMMARY:ADIC2: Development of a Component Source Transformation System for Differentiating C and C++ UID:889 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100610T140000 DTEND;TZID=America/Chicago:20100610T150000 DTSTAMP:20130525T020110 LOCATION:Building 240 Conference Room 1404, Argonne National Laboratory DESCRIPTION:Several lines of evidence point to the conclusion that the largest cosmic structures -- galaxies, clusters of galaxies, and superclusters -- have formed through the action of gravitational instability on a mixture of ordinary atomic matter and weakly interacting \"dark matter\" that is dominated by the latter form. This instability empties vast regions of space to collect both kinds of matter into enormous filamentary structures along which galaxies are found. The tasks of cosmological simulation are to understand in detail how this process operates, how the galaxies evolved and became luminous, and how the observed cosmic structures can be used to place constraints on the fundamental parameters underlying cosmological models. Growth in the size and complexity of observational datasets and the advent of petascale computational resources have forced computational cosmologists to confront the scalability and accuracy of their simulation codes as never before. I will discuss some of the factors behind these challenges and describe how my group and our collaborators are attempting to address them within the context of the adaptive mesh refinement code FLASH. SUMMARY:Petascale Challenges for Computational Cosmology UID:890 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100614T103000 DTEND;TZID=America/Chicago:20100614T113000 DTSTAMP:20130525T020110 LOCATION:Building 240 / Room 4301, Argonne National Laboratory DESCRIPTION:Currently, there exists a lack of confidence in the computational simulation of turbulent separated flows at large Reynolds numbers. The most accurate computational methods available are prohibitively costly for use in engineering applications. As a result, low-cost models using the Reynolds-averaged Navier-Stokes (RANS) equations are often applied to flows far beyond those for which they were originally designed. These methods will regularly reproduce integrated flow quantities within engineering tolerances (e.g, pressure on a surface); however, such metrics are often insensitive to bluff body flow physics, and therefore, poor indicators of simulation fidelity.\n\nUsing concepts borrowed from large-eddy simulation (LES), a two-equation RANS model is modified to simulate a bluff body turbulent wake. This modification only requires the computation of one additional scalar field, adding very little to the overall computational cost. When properly inserted into the baseline RANS turbulence model, this modification mimics the fidelity of LES in a separated wake, yet reverts to the un-modified form at solid surfaces. In this manner, superior predictive capability may be achieved without the additional cost of fine spatial resolution associated with LES near solid boundaries.\n\nIn this presentation we compare simulations of the modified and baseline RANS models to both high-resolution LES and experimental data for the canonical circular cylinder wake at Reynolds number 3900. Mean value and triple-decomposition analysis reveal substantial improvements using the modified system that appear to drive the flow solution toward that of LES, as intended.\n\nIn addition, the parallelized, overset grid algorithms used in this investigation are subject to code verification via the Method of Manufactured Solutions. The presented results confirm both the spatial and temporal order of accuracy.\n SUMMARY:A Novel, Eddy-viscosity-evolution Based Formulation for the Simulation of Turbulent Separated Flows UID:891 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100621T103000 DTEND;TZID=America/Chicago:20100621T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, TCS Conference Center, 1404 & 1405, Argonne National Laboratory DESCRIPTION:Current implementations of MPI are coarse-grained, with a single MPI process per processor, however, there is nothing in the MPI specification precluding a finer-grain interpretation of the standard. We have implemented Fine-grain MPI (FG-MPI), a system that allows execution of hundreds and thousands of MPI processes on one node or communicating between nodes inside a cluster. FG-MPI uses fibers (coroutines) to support multiple MPI processes inside an operating system process. These are full-fledged MPI processes each with their own MPI rank. FG-MPI is based on MPICH2 middleware and uses the Nemesis communication subsystem for intra-node and inter-node communication.\n\nAlan Wagner will give experimental results for applications using thousands of MPI processes and compare its performance with several fine-grain multicore languages. FG-MPI also made it possible to investigate problems related to scaling of MPI to a larger number of processes. I will present the design and evaluation of techniques to support the scalability of communicators and groups in MPI. Performance results are given for the execution of an MPI benchmark program with upwards of 100,000 MPI processes with communicators created for various groups of different sizes and types.\n\nBrief Biography:\n\nAlan Wagner (BSc Dalhousie University, MSc University of Alberta, PhD, University of Toronto) is an Associate Professor in Computer Science at the University of British Columbia. His research interests include MPI middleware and message-passing systems, networking, data-mining, and computational finance. SUMMARY:FG-MPI: a Fine-Grain Implementation of MPI for Multicore and Clusters UID:892 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100611T093000 DTEND;TZID=America/Chicago:20100611T103000 DTSTAMP:20130525T020110 LOCATION:Building 240 Room 1407, Argonne National Laboratory DESCRIPTION:Mike will present a slide show of scientific visualizations that have been created by staff of Argonne and the University of Chicago. The format will be interactive with a discussion of the process and techniques used. Popcorn will be provided. SUMMARY:Scientific Visualization: Slide Show UID:893 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100614T093000 DTEND;TZID=America/Chicago:20100614T103000 DTSTAMP:20130525T020110 LOCATION:Building 240 Room 1407, Argonne National Laboratory DESCRIPTION:In many science areas, the quest for increased computational resources is driven by a need to span a broader range of scales, that is, to capture the interaction of small scales with the large. In transport problems such as electromagnetics and fluid mechanics, this implies a need to propagate small scale features over long times and distances. In numerical simulations, such long-time integrations are most efficiently realized by using high-order discretizations. Here, we present recent advances in spectral element methods designed for the petascale efficient single- and multi-node performance. Application areas include the study of magnetorotational turbulence in accretion disk models, heat transfer in advanced reactor designs, wakefield computations in accelerators, and transition to turbulence in vascular flows. SUMMARY:Numerical Algorithms for Petascale Science UID:895 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100804T133000 DTEND;TZID=America/Chicago:20100804T170000 DTSTAMP:20130525T020110 LOCATION:Building 240, Conference Center, Rm 1404 -> 1407, Argonne National Laboratory DESCRIPTION:The symposium features 11 talks by Summer students at LANS. \n\nSchedule\n=======\n1:30- 1:45 Jorge Castaton\n Computing Projections\n1:45- 2:00 Zhu Tao\n Filter Methods for Augmented Lagrangian\n2:00- 2:15 Anirban Chatterjee\n Improving Clustering of High-Dimensional Data through Algebraic Distances\n2:15- 2:30 Break\n2:30- 2:45 Zhu Wang\n Dimensionality reduction for uncertainty quantification of nuclear engineering models\n2:45- 3:00 Alexander Stovall and Pierre Robinson\n Binary Optimization and Empirical Instruction Scheduling for Autotuning\n3:00- 3:15 Grantland Gray and Corie Wilson\n A Classified Method Based on Support Vector Machine for Network Intrusion Detection\n3:15- 3:30 Break\n3:30- 3:45 Mihai Alexe\n Monty Python and the Holy Grail of fast uncertainty quantification using magic tricks and automatic differentiation.\n3:45- 4:00 Jing Fu\n Parallel I/O approaches for check pointing on massively parallel partitioned solvers\n4:00- 4:15 Shankar Prasad Sastry\n Preconditioner for Optimization in Power Flow Systems\n4:15- 4:25 Break\n4:25- 4:40 Chia-chun Tsai\n Power Grid Models In Application\n4:40- 4:55 Brian Haines\n Numerical homogenization approach for Stokesian suspensions\n\nAbstracts\n======\n\n Jorge Castaton\nTitle: Computing Projections\nAbstract:\nIn this talk we will discuss algorithms for computing projections onto convex sets. These projections are used in a great variety of science applications. Specifically, we use a semi-smooth approach considering both a semi-smooth Newton method and a matrix-free first order method. Numerical tests indicate that as the dimensions of the problem grow, the first order method has a better performance. A diagonal pre-conditioner also improves the first order method. These findings are particularly important because matrix-free methods, unlike interior point methods, require less memory storage.\n\n Mihai Alexe \nTitle: Monty Python and the Holy Grail of fast uncertainty quantification using magic tricks and automatic differentiation.\nAbstract:\nA posteriori uncertainty quantification may benefit from the introduction of derivative information for the outputs or parameters of interest. Automatic differentiation (AD) is a natural choice for computing derivatives of program outputs wrt the control variables. Moreover, it can do so at a significantly lower cost than the naive finite difference approach. The talk will give a quick introduction to the principles of AD. Then, we describe the results obtained with MCS\' own OpenAD tool for the nuclear reactor safety simulation code MATWS, developed at Argonne\'s Nuclear Engineering division. Efficient taping and checkpointing of intermediate program variables enable the fast computation of gradients using the reverse mode of AD.\n\n Zhu Tao\nTitle: Filter Methods for Augmented Lagrangian\nAbstract:\nBound constrained augmented Lagrangian (BCL) method is an appealing way to solve large scale nonlinearly constrained optimization problems because the bound constrained subproblems can usually be solved efficiently and allow large scale implementation. Unfortunately, classic BCL methods suffer from several deficiencies: 1) progress towards the solution is rigidly prescribed via two forcing sequences which control feasibility and optimality of the subproblems; 2) convergence near regular minimizers can be slow; 3) the penalty update may result in slow convergence in the beginning and difficult subproblems in the later iterations. In this project, we investigated a new class of filter BCL methods for nonlinear optimization that overcome these deficiencies. First, the forcing sequences are replaced by a two-dimensional filter which is more flexible in terms of accepting trial steps. Second, an equality constrained quadratic programming (EQP) phase is added to accelerate the convergence. Third, a penalty estimate function is used to ensure convergence of the subproblems. Numerical experiments on a subset of the CUTEr test problems demonstrate the effectiveness of this approach.\n\n Anirban Chatterjee \nTitle: Improving Clustering of High-Dimensional Data through Algebraic Distances\nAbstract:\nMeasuring the connection strength between two entities in high-dimensional space is one of the most vital concerns in data mining. In this project, we adapt recently introduced measure on simple graphs and hypergraphs, namely the algebraic distance, for improving unsupervised classification of high-dimensional text data. In particular, we work on a multilevel approach towards the noise elimination problem for high-dimensional discrete systems obtained in text mining.\n\n Jing Fu \nTitle: Parallel I/O approaches for check pointing on massively parallel partitioned solvers\nAbstract:\nWe present several parallel I/O approaches and compare them with traditional POSIX I/O strategy (1 file per processor). We tackle this problem from different aspects, including I/O library choosing, access concurrency reduction, and separating I/O communicator from computation communicator. These approaches are especially useful for checkpoint-restart on large-scale parallel partitioned solvers. We applied these approaches on two applications, NEKCEM@MCS and PHASTA@RPI, and we will analyze the performance (I/O rates and application run time reduction) on Intrepid at different scales.\n\n Shankar Prasad Sastry \nTitle: Preconditioner for Optimization in Power Flow Systems\nAbstract:\nIn this talk, we explore the use of preconditioners in power flow system optimization. In a power flow system, we optimize the cost of production of electricity under demand, network capacity and other physical constraints. Interior point method is used in such constrained optimization problems. In interior point methods, we incorporate the constraints in the objective function such that the cost of violating the constraints is very high. We lower the weight of the constraints after every iteration. Thus, when the weights are small enough, we optimize the real objective function. The constraints are large as the network. Thus, in every iteration, we solve huge linear systems, which determine the optimal solution. In current implementation in MATPOWER solver, the linear system is solved through LU decomposition. However, the system is sparse and we can use iterative techniques with preconditioners to solve the linear system efficiently. In this talk, we introduce power systems, interior point methods, linear solvers, preconditioners and also propose a preconditioner applicable to this problem.\n\n Alexander Stovall and Pierre Robinson \nTitle: Binary Optimization and Empirical Instruction Scheduling for Autotuning\nAbstract:\nEmpirical performance tuning has been emerging as an effective means to improve performance of application programs. In this presentation, we will talk about experiences of using low-level optimization techniques to improve performance of applications on an AMD Phenom processor. We also explored a new approach where the given instruction scheduling is altered to improve performance based solely on performance measurement. This approach does not suffer from modeling errors found in some techniques such as using integer linear programming solvers.\n\n Grantland Gray and Corie Wilson \nTitle: A Classified Method Based on Support Vector Machine for Network Intrusion Detection\nAbstract:\nIntrusion detection is a critical requirement for enterprise network protection, since one of its necessary tasks is to protect the computers responsible for the infrastructures operational control, and an effective intrusion detection system (IDS) is essential for ensuring network security. Network-based attacks make it difficult for legitimate users to access various network services by sabotaging network resources and services. This is achieved by sending large amounts of network traffic, exploiting well-known flaws in networking services, and by overwhelming network hosts. Intrusion Detection attempts to detect computer attacks by examining various data records observed in processes on the network and it is split into two groups, anomaly detection systems and misuse detection systems. Anomaly detection is an attempt to search for malicious behavior that deviates from established normal patterns. Misuse detection is used to identify intrusions that match known attack scenarios. In this research effort, we focus on anomaly detection and our proposed strategy is an efficient and reliable solution for detecting network based anomalies. We employ a supervised machine learning method, Support Vector Machines (SVM) for classification of abnormal traffic and normal traffic and LiBSVM and LibLinear, as support vector machine tools. The tools provide an effective mechanism to perform cross-validation, parameter selection and training large datasets. Performance evaluation of the proposed method is conducted on diverse publicly available network packet traces. Experimental results show that high average detection rates and low average false positive rates in anomaly detection are achieved by our proposed method.\n\n Chia-chun Tsai\nTitle: Power Grid Models In Application\nAbstract: \nThis project investigate three power system problems: (1)Optimal Power flow(OPF). (2)Transmission Network Expansion(TNE). (3)Optimal Transmission Switching(OTS). Among these models, we concentrate on building efficient reformulation techniques that model the non-convex power flow equations such as Kirchhoff\'s law in AC problem and Ohm\'s law in DC problem. We also investigate the structure and formulation of these models and identify common mathematical components. The DC model in TNE and OTS problems of this type lead to non-convex nonlinear mix-integer optimization problem, we can apply big-M method to get linear mix-integer model and we can also apply complementarity method to get another nonlinear model. Our goal is to develop a number of case studies that illustrate the computational and mathematical challenges, and that can be used to benchmark new global optimization solvers. \n\n Zhu Wang \nTitle: Dimensionality reduction for uncertainty quantification of nuclear engineering models\nAbstract: \nWe address the question of uncertainty quantification of complex nuclear engineering simulation models. Previous research effort has shown that propagation of uncertainty in the inputs can be approximated by polynomial regression, at the cost of very few computationally expensive model evaluations, if derivative information is also used. When dimension of uncertainty space is truly large (we estimate ~100), its sampling does not extract enough information for a good approximation, even if derivative information is used, and regression setup is optimal. Some sort of dimensionality reduction is required. We project the high-dimensional uncertainty space into a reduced representation, using proper orthogonal decomposition (POD) based reduction that is dual-weighted (i.e. individual importance is assigned to each training set point, and to each component of points). The weighting is based on derivative information of the model; comparisons with different weighting schemes are also performed. Our work indicates that it is possible to perform high-precision uncertainty quantification when the dimension of the uncertainty space is ~100, or more, which has practical applications for currently difficult tasks of uncertainty quantification, parametric dependence analysis, verification and validation for nuclear engineering models with many parameters, and sparse description of the mathematical structure.\n\n Brian Haines \nTitle: Numerical homogenization approach for Stokesian suspensions\nAbstract: \nSuspensions of rigid particles present numerical difficulties when they contain many particles, largely due to their complicated moving boundaries. We propose a new approach to studying these problems through numerical homogenization. This is done by introducing a FEM basis that is computed once for the suspension in its initial configuration. This basis can be cheaply advected to produce a basis to be used at later times (in the evolved geometry) with explicitly controlled error. Furthermore, we present error estimates for the approximate solution and discuss ongoing work on localizing the initial basis so that it is cheaper to compute. With a localized basis, the computational complexity is reduced significantly when compared to standard approaches. SUMMARY:SASSy: Student Argonne Summer Symposium UID:894 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100615T093000 DTEND;TZID=America/Chicago:20100615T103000 DTSTAMP:20130525T020110 LOCATION:Building 240 Room 1407, Argonne National Laboratory DESCRIPTION:The Argonne Leadership Computing Facility (ALCF) houses the largest installation of IBM\'s BlueGene/P, a massively parallel supercomputer. This talk will present an overview of the BlueGene/P architecture and why the system is attractive to application developers. SUMMARY:Introduction to Blue Gene P Architecture UID:897 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100617T093000 DTEND;TZID=America/Chicago:20100617T103000 DTSTAMP:20130525T020110 LOCATION:Building 240 Room 1416, Argonne National Laboratory DESCRIPTION:More and more complete genomes for bacteria are becoming available and are providing a series of challenges to bioinformatics. Finding the genes and describing gene function for hundreds of genomes is a complex problem that we currently are in the beginning stages of solving. I will highlight the importance of comparative analysis in the process of analyzing these genomes. SUMMARY:Annotating the First 1000 Genomes UID:898 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100625T103000 DTEND;TZID=America/Chicago:20100625T120000 DTSTAMP:20130525T020110 LOCATION:Building 240, TCS Conference Center, 1406 & 1407, Argonne National Laboratory DESCRIPTION:Due to the explosive growth in the size of scientific data sets, data-intensive computing is an emerging trend in computational science. Many application scientists are looking to integrate data-intensive computing into computational-intensive High Performance Computing facilities, particularly for data analytics. We have observed several scientific applications which must migrate their data from an HPC storage system to a data-intensive one. There is a gap between the data semantics of HPC storage and data-intensive system, hence, once migrated, the data must be further refined and reorganized. This reorganization requires at least two complete scans through the data set and then at least one MapReduce program to prepare the data before analyzing it. Running multiple MapReduce phases causes significant overhead for the application, in the form of excessive I/O operations. For every MapReduce application that must be run in order to complete the desired data analysis, a distributed read and write operation on the file system must be performed. Our contribution is to extend Map-Reduce to eliminate the multiple scans and also reduce the number of pre-processing MapReduce programs. We have added additional expressiveness to the MapReduce language to allow users to specify the logical semantics of their data such that 1) the data can be analyzed without running multiple data pre-processing MapReduce programs, and 2) the data can be simultaneously reorganized as it is migrated to the data-intensive file system. Using our augmented MapReduce system, MapReduce with Access Patterns (MRAP), we have demonstrated up to 33% throughput improvement in one real application, and up to 70% in an I/O kernel of another application.\n SUMMARY:MRAP: A Novel MapReduce-based Framework to Support HPC Analytics Applications with Access Patterns UID:899 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100622T103000 DTEND;TZID=America/Chicago:20100622T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, TCS Conference Center, 1404 & 1405, Argonne National Laboratory DESCRIPTION:This talk introduces new classes of collective operations from an implementation as well as an application programmer\'s perspective. We discuss issues with schedule generation, caching, and progression, and how these influence the application programmer. Then we focus on simple strategies, such as loop tiling, pipelining, and simple code movement, that can be used to optimize application performance with nonblocking collectives. We also discuss how the new semantics can be utilized to design new, asymptotically optimal algorithms for one-level termination detection, which is important for data-driven algorithms. The second part of the talk focusses on sparse collective operations and static binding of communication topologies. We discuss a possible interface for MPI-3, several productivity and performance issues, and show some performance results and potential for future work and architectures. SUMMARY:Nonblocking and Sparse Collective Operations on Petascale Computers UID:900 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100618T093000 DTEND;TZID=America/Chicago:20100618T103000 DTSTAMP:20130525T020110 LOCATION:Building 240 Room 1404, Argonne National Laboratory DESCRIPTION:MPI (Message Passing Interface) is a standard, portable interface for writing message-passing parallel programs. Dave will give an introduction to MPI and describe various MPI-related projects in MCS. SUMMARY:MPI Tutorial UID:901 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100621T093000 DTEND;TZID=America/Chicago:20100621T103000 DTSTAMP:20130525T020110 LOCATION:Building 240 Room 1407, Argonne National Laboratory DESCRIPTION:This will be a highly interactive presentation with a focus on scientific consensus versus what makes a great headline in the news. Doug is the ARM Climate Research Facility (ACRF) Operations Manager. The ACRF is a DOE national user facility with word-wide sites. SUMMARY:Global Warming: The Real Deal or Just a Good Story? UID:902 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100707T150000 DTEND;TZID=America/Chicago:20100707T160000 DTSTAMP:20130525T020110 LOCATION:Building 240, Conference Center 1404-1405, Argonne National Laboratory DESCRIPTION:Anticipating the 2009 analog-to-digital TV transition, the Federal Communications Commission issued a landmark ruling on Nov. 4, 2008, voting unanimously in favor of opening unused television frequencies, called white spaces, for unlicensed use. Regulators around the world want to follow suit and are watching the United States closely. To some, this is the biggest opportunity in wireless communications since GSM. To others, it is an irresponsible act that will interfere with television broadcasts, rock concerts, and church services. To me, white space networking is the first main-stream manifestation of a powerful idea - opportunistic dynamic spectrum access networks. With the advent of cognitive radio technology, opportunistic dynamic spectrum access has the potential to mitigate spectrum scarcity and meet the increasing demand for spectrum. Following topics will be discussed : 1) Introduction to cognitive radio technology and dynamic spectrum access, 2) Technical Challenges pertaining to communications and networking protocols and 3) Spectrum-aware protocol and algorithms to deal with spectrum fragmentation. SUMMARY:Opportunistic Networking via Dynamic Spectrum Access UID:904 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100622T093000 DTEND;TZID=America/Chicago:20100622T103000 DTSTAMP:20130525T020110 LOCATION:Building 240 Room 1407, Argonne National Laboratory DESCRIPTION:In this presentation, Hongjun and Rajeev will review the algorithms commonly used in mesh generation, for both tetrahedral and hexahedral meshes. They will also describe some of the infrastructure common to both cases, for example representation and evaluation of the geometric domain. Also, Hongjun and Rajeev will conclude by describing active areas of research in mesh generation and its application to current problems in scientific computing. SUMMARY:Mesh Generation for Scientific Computing UID:905 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100811T150000 DTEND;TZID=America/Chicago:20100811T160000 DTSTAMP:20130525T020110 LOCATION:Bldg 240 Conference Center 1404-1405, Argonne National Laboratory DESCRIPTION:The rapid move towards the regionalization of climate models leads to the\nquestion: How valid is the atmosphere being parameterized within the model\nat regional scales? Given the massive disparity in spatial scales between\ndependable in-situ measurements and a global model grid cell how do you\nverify that atmospheric processes are being represented accurately? Recent\npublications detail the use of sounding data to drive a variety of climate\nand cloud resolving models for intercomparison but these results need to be\nunderpinned by measurements. This talk will introduce the use of Doppler\nradar in the remote sensing of precipitating cloud systems and the\napplication of variational techniques to generate model-like data from the\nmeasurements they produce for the verification of modeling efforts. SUMMARY:Climate modeling meets remote sensing: How do you measure a storm cloud? UID:906 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100625T093000 DTEND;TZID=America/Chicago:20100625T103000 DTSTAMP:20130525T020110 LOCATION:Building 240 Room 1405, Argonne National Laboratory DESCRIPTION:Empirical performance tuning has been emerging as an attractive means of coping with increasingly complex computer architectures and application programs. At MCS, we have been developing technologies to automate empirical tuning process of leadership class scientific applications. In this lecture, Jaewook will use their current project as an example to introduce the audience to empirical performance tuning. SUMMARY:Empirical Performance Tuning UID:907 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100709T140000 DTEND;TZID=America/Chicago:20100709T150000 DTSTAMP:20130525T020110 LOCATION:Building 240, TCS Conference Center, 1406 & 1407, Argonne National Laboratory DESCRIPTION:In this talk I will motivate the need for better testing tools for distributed and concurrent systems. The first part of the talk will describe recent advancements and successes in testing of distributed and concurrent systems. The second part of my talk will then present the design and implementation of dBug - a tool for systematic evaluation of distributed and concurrent systems. The talk will conclude with an overview of two case studies for the use of dBug: the Parallel Virtual File System and FAWN-KV, a distributed key-value storage based on the FAWN architecture.\n SUMMARY:dBug: Systematic Evaluation of Distributed Systems UID:909 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100701T093000 DTEND;TZID=America/Chicago:20100701T103000 DTSTAMP:20130525T020110 LOCATION:Building 240 Room 1407, Argonne National Laboratory DESCRIPTION:Moving large quantities of data across wide area networks can be a painful and laborious process. The Center for Enabling Distributed Petascale Science (CEDPS) project has developed a service called Globus.org that automates this process. Users hand off data transfer tasks (e.g., move a set of files, mirror a directory) to the hosted Globus.org service, which then does its best to ensure that the transfer completes successfully, monitoring performance and errors, correcting problems automatically whenever possible, and reporting back on status and errors. Thus, we eliminate the need to babysit transfers, freeing users to focus on domain-specific work. This lecture includes an overview of the motivation for Globus.org, a discussion of key design concepts, and a demonstration of current capabilities.\nNOTE: Cookies will be served! SUMMARY:Introduction to Globus.org UID:910 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100702T093000 DTEND;TZID=America/Chicago:20100702T103000 DTSTAMP:20130525T020110 LOCATION:TCS Building 240 Conference Room 1407, Argonne National Laboratory DESCRIPTION:There are many different open source game programming libraries available, but few are truly multiple platform capable. I will focus my discussion on Pygame, which is a python library built on top of SDL to aid in the abstraction from OS and hardware. In this talk I will talk about some topics and fundamentals of 2d game programming and then walk through some demo code to illustrate the topics discussed. The end result is a working 2d scrolling shooter game. SUMMARY:Open Source Game Programming with Pygame UID:911 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100706T093000 DTEND;TZID=America/Chicago:20100706T103000 DTSTAMP:20130525T020110 LOCATION:TCS Building 240 Conference Room 1407, Argonne National Laboratory DESCRIPTION:The presentation will provide an overview of how Cyber Security is implemented at an open science, highly collaborative DOE National Laboratory. Topics discussed will include: \"Worst to First\" - Evolution of the ANL Cyber Program , technical highlights, current collaborative projects and futures. SUMMARY:Cyber Security at a National Laboratory UID:912 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100707T130000 DTEND;TZID=America/Chicago:20100707T140000 DTSTAMP:20130525T020110 LOCATION:Building 240 / Conference Room 4301, Argonne National Laboratory DESCRIPTION:Many proteins, and by extension protein networks and biological processes, are affected by interactions with specific small molecules. Understanding the basis and mechanism of protein small-molecule interactions is also crucial for drug discovery and design. Given the number of novel protein structures solved by groups in structural genomics initiatives, automated methods to mine structural datasets are important. This is especially applicable in the case of shared features that interact with the same molecule. In many instances, the binding sites and the physiologically relevant small molecules that interact with proteins from structural genomics are unknown. Thus, a computational tool that compares potential binding sites against a dataset of proteins that have small molecules bound can be useful to propose candidate ligands for proteins with unknown function. \n\nWhile at Michigan State University, I have designed and implemented a software package, SimSite3D, that searches a dataset of binding sites for those that are similar to a given query site. The initial goal was to develop a robust tool that can quickly perform the searches. By using rigid alignments and coarse samplings of the binding sites, the initial goal has been met, and SimSite3D has now become part of Pfizer Global R&D\'s drug discovery toolkit. \nHowever, the problem of identifying otherwise unrelated proteins that bind the same molecule is challenging. In protein science, binding site shape is known to be an important feature. This has been addressed by adding a triangulated mesh representation of the protein surfaces and using rigid refinements of aligned surface meshes to improve the binding site orientations. The addition of binding site surfaces and refinement of orientations has been shown to capture more remote similarities and increase the accuracy of alignments (at the cost of additional computation). \n\nMost recently, I have added flexible refinement of binding sites using articulated surface and chemistry matching to address the fact that proteins are intrinsically flexible. This method, related to techniques in computer animation and robotics, addresses the question of whether one binding site can be morphed into another, subject to the underlying molecular constraints. The preliminary results are encouraging, but given the challenges of protein flexibility, there are ample opportunities for future work. \n\nClick here to add this seminar to your calendar. SUMMARY:Mining Protein Binding Sites with Flexible Surface and Chemistry Matching UID:913 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100708T130000 DTEND;TZID=America/Chicago:20100708T140000 DTSTAMP:20130525T020110 LOCATION:Building 240 / Conference Room 4301, Argonne National Laboratory DESCRIPTION:Fragment molecular orbital method (FMO)* is one of the most promising approaches to efficiently perform quantum-mechanical calculations of large systems. The advantages over traditional QM calculations are: relatively low CPU and memory requirements, scalability on large number of CPUs, ability to compute properties of fragments and their interactions. FMO method is proven to be useful for biological applications such as studying ligand-protein and protein-protein interactions. However, the method can be further developed to achieve better scaling and greater efficiency.\n\nIn this seminar, I will give a short introduction to FMO method and its applications to drug research. FMO issues and bottlenecks will be covered in detail. A new scalable distributed data FMO algorithm will be presented which will make code peta-scale capable.\n\n*K. Kitaura et al. (1999). Chem. Phys. Lett. 313: 701–706\n\nClick on the link below to add this seminar to your calendar.\n\n SUMMARY:An Introduction to the Fragment Molecular Orbital Method and Its Application to Drug Research UID:914 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100708T093000 DTEND;TZID=America/Chicago:20100708T103000 DTSTAMP:20130525T020110 LOCATION:TCS Building 240 Conference Room 1405, Argonne National Laboratory DESCRIPTION:Basic concepts of parallel programming will be discussed, along with how to get access to Argonne\'s Laboratory Computing Resource Center. Presentation is largely meant for those new to parallel programming. The presentation covers common terminology, basic parallel hardware models, and parallel programming models. A few basic examples will also be discussed. SUMMARY:Introduction to Parallel Computing UID:915 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100709T093000 DTEND;TZID=America/Chicago:20100709T103000 DTSTAMP:20130525T020110 LOCATION:TCS Conference Center Bldg 240; Room 1405, Argonne National Laboratory DESCRIPTION:The Swift parallel scripting language lets users apply parallel composition constructs to existing sequential or parallel programs to express highly parallel scripts.\n\nSwift scripts are flexible and portable, and can run efficiently on platforms ranging from multicore workstations to petascale supercomputers. For performing parameter sweeps and data analysis with exiting application programs, parallel scripting is typically easier and more productive than tightly-coupled parallel programming.\n\nThis talk will provide an overview of Swift and how its used to run scientific applications in parallel on clusters, grids, clouds, and petascale systems.\n \nThe architectural challenges of scripting on large-scale systems will be covered, and case studies will be presented. Swift’s place in the taxonomy of parallel programming languages and environments will be discussed, and speculative ideas for hybrid models for multi-level programming to go beyond the petascale will be considered.\n SUMMARY:Parallel Scripting with Swift for Applications at the Petascale and Beyond UID:916 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100712T093000 DTEND;TZID=America/Chicago:20100712T103000 DTSTAMP:20130525T020110 LOCATION:TCS Conference Center Bldg 240; Room 1407, Argonne National Laboratory DESCRIPTION:Nick will begin with a high-level overview of electronic structure methods and their role in predicting material properties. He will then describe a multi-institute effort in scaling the GPAW code on BlueGene/P. Lastly, he’ll present the challenges that may lie ahead for electronic structure methods, particular density functional theory, at the exascale. SUMMARY:Computational Material Science Using the Blue Gene/P UID:917 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100928T080000 DTEND;TZID=America/Chicago:20100928T170000 DTSTAMP:20130525T020110 LOCATION:Hyatt Regency Atlanta, Atlanta, Georgia DESCRIPTION:The Grace Hopper Celebration of Women in Computing is a series of conferences designed to bring the research and career interests of women in computing to the forefront. Presenters are leaders in their respective fields, representing industrial, academic and government communities. Leading researchers present their current work, while special sessions focus on the role of women in today\'s technology fields, including computer science, information technology, research and engineering.\n\nPast Grace Hopper Celebrations have resulted in collaborative proposals, networking, mentoring, and increased visibility for the contributions of women in computing. SUMMARY:Collaborating Across Boundaries UID:918 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100728T103000 DTEND;TZID=America/Chicago:20100728T113000 DTSTAMP:20130525T020110 LOCATION:Building 240 Seminar Room 4301, Argonne National Laboratory DESCRIPTION:This talk concerns particle methods and has two parts. First we present a treecode algorithm for evaluating multiquadric radial basis function (RBF) approximations. The treecode employs a divide and conquer strategy and replaces particle-particle interactions by particle-cluster interactions which are efficiently computed by a far-field Cartesian Taylor approximation. For the multiquadric RBF, $\\phi(x) = \\sqrt{x^2 + c^2}$, the Laurent series presented in the literature converges only for a limited range of the RBF shape parameter $0 \\le c \\le c^*$, but the Taylor series employed here converges for all $c \\ge 0$. The treecode reduces the computational cost from $O(N^2)$ to $O(N\\log N)$ operations, where $N$ is the size of the system. Second we discuss work in progress on solving the Barotropic Vorticity Equation on a rotating sphere by a vortex method. Vortex methods are Lagrangian techniques in contrast with more conventional Eulerian methods such as finite difference, finite element and spectral methods. The vortex method uses Lagrangian particles and panels to track the flow map and absolute vorticity. The velocity is computed from the Biot-Savart integral on the sphere. An adaptive refinement strategy is implemented to resolve small-scale features. Results are presented for Rossby-Haurwitz waves and vortex patch interactions. The two parts of the talk are related because the treecode can be used to evaluate the Biot-Savart integral on the sphere. SUMMARY:A Cartesian Treecode for Multiquadric Radial Basis Functions and a Vortex Method for Fluid Flow on a Sphere UID:919 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100716T093000 DTEND;TZID=America/Chicago:20100716T103000 DTSTAMP:20130525T020110 LOCATION:TCS Conference Center Bldg 240; Room 1407, Argonne National Laboratory DESCRIPTION:Systems Administration is often thought of as something you fall into, something you do because you have to, or something you do before moving on to something else. But it doesn\'t have to be that way. (This talk will be mostly monkey-free.) SUMMARY:Systems Administration As A Career UID:920 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100722T103000 DTEND;TZID=America/Chicago:20100722T120000 DTSTAMP:20130525T020110 LOCATION:Building 240 Conference Room 1404, Argonne National Laboratory DESCRIPTION:IPCC and other current projections of climate change rely on global models of climate that must be run at a coarse horizontal resolution – approximately 150 km for many of the models used in the recent IPCC AR4. The next IPCC assessment (AR5), underway as of this writing, has only increased that resolution to about 100 km. As stressed by IPCC, results at the global scale are useful for indicating the general nature and large-scale patterns of climate change, but are not very robust at the local or regional scale. This is for two key reasons: 1) the model can only explicitly resolve physical processes operating over several hundred kilometers or larger; and 2) spatial surface heterogeneities, especially regions of complex topography or differing land use patterns, can be large and occur on small spatial scales of 5-20 km. \n\nWe use a high-resolution (order 10 km), limited-domain (i.e., regional) climate model to perform a physically-based downscaling. Because climate is inherently global in nature, a limited-domain model must be driven at its lateral boundaries by either observations or output from a global model. Output from a global model is required if scenarios of future climate change (e.g., due to greenhouse gas warming) are needed, while observations (in reality the proxy reanalyses) can be used if, for example, the effects on regional climate of changes in land use patterns is under investigation. Representative results will be presented, with a focus on the central U.S.\n SUMMARY:Climate Change in your Backyard: The Role of Regional Climate Models UID:921 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100825T100000 DTEND;TZID=America/Chicago:20100825T113000 DTSTAMP:20130525T020110 LOCATION:Bldg 240 Conference Center 1404-1405, Argonne National Laboratory DESCRIPTION:The event features three talks by MCS summer students.\n\n Rajat Garg \nTitle: Performance Implications of Parallel I/O in Adjoint Computations\nAbstract\nAdjoint computations offer the potential to provide insight (e.g., sensitivites with respect to billions of independent variables) at a modest increase in flops, typically only a factor 3-10 relative to simulation. However, these computations require intermediate states in reverse order of their generation. The states are restored through a combination of application-level checkpointing and recomputation from the checkpointed states. Adjoint computations typically assume a cluster-like execution environment and checkpoint to loal disk. We will attempt to use parallel I/O (e.g., MPI-I/O) to store this data to a global parallel file system. We will add parallel I/O to the adjoint code for MITgcm, a general circulation model used for ocean state estimation and related climate studies. We are interested in studying the performance characterisitics of the resulting file access patterns, which are not typical of current simulation codes.\n\n Manu Shantharam \nTitle: Performance Bounds Prediction of Parallel Applications\nAbstract\nOne of the main concerns in high performance computing relates to the\nestimation of application performance bound on a given hardware. For a\ngiven hardware configuration and an application, we estimate the maximum\nperformance that the application can achieve using static program\nanalysis.\n\nIn the talk, first, I will give a basic introduction to the ROSE compiler\nframework and how it is used within the Pbound tool. Next, I will present\nmy work related to improving the efficiency of Pbound and extending its\nfunctionality to parallel applications.\n\nVyacheslav Kungurtsev\nTitle: Augmented Lagrangian Interpolation-Based Derivative-Free Optimization\nAbstract\nThere is a class of general nonlinear programming problems (NLP) for which derivative information is unavailable. So far, algorithms for constrained optimization have used direct-search methods. Using the POUNDer algorithm for interpolating an objective function with a quadratic model, we attempt to solve constrained NLPs using the Augmented Lagrangian framework. In this framework, each subproblem is interpolated and solved, then the subproblems are updated at an outer iteration. Numerical results show the method to be promising. SUMMARY:SASSy : Student Argonne Summer Symposium - Round 2 UID:922 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100818T150000 DTEND;TZID=America/Chicago:20100818T160000 DTSTAMP:20130525T020110 LOCATION:Bldg 240 Rm 4301, Argonne National Laboratory DESCRIPTION:In our larger research effort, we approximate the effect of uncertainty in the inputs of a complex simulation model by a hybrid polynomial regression method that uses first-order derivatives of the model as additional fitting conditions. We shall now discuss dimensionality reduction required for the models with large dimension of the uncertainty space. Two other aspects: automatic differentiation, and selection of best polynomial basis for approximation will also be mentioned. SUMMARY:Dimension reduction in uncertainty quantification of nuclear engineering models UID:923 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100805T120000 DTEND;TZID=America/Chicago:20100805T130000 DTSTAMP:20130525T020110 LOCATION:Cafeteria, Private Dining Rooms A&B, Argonne National Laboratory DESCRIPTION:The South Pole Telescope (SPT) is a 10-m mm-wave telescope located at the Amundsen-Scott South Pole Station. The SPT achieved first light in early 2007 and has been surveying the sky at 3, 2, and 1-mm. SPT utilizes a Transition Edge Sensor bolometer array to make it the most sensitive survey instrument at these observing wavelengths. The first results from this survey include the discovery of a new population of mm-wave sources believed to be distant dusty star forming galaxies, discovery of new galaxy clusters through the Sunyaev-Zeldovich effect, and measuring the power spectrum of distant unresolved galaxy clusters. In this talk, I will review these initial results and discuss the current status of SPT. I will also describe our plans for the next SPT project, SPTpol, which will utilize new bolometer technology developed in collaboration with ANL to conduct the most precise measurement of the polarization of the Cosmic Microwave Background. SUMMARY:Astrophysics and Cosmology with the South Pole Telescope UID:925 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100802T103000 DTEND;TZID=America/Chicago:20100802T120000 DTSTAMP:20130525T020110 LOCATION:Bldg 240 Conference Room 1406-1407, Argonne National Laboratory DESCRIPTION:Title: Depth Analysis of MPI Programs\n\nAbstract:\n\nData-flow analyses that include some model of the data-flow between MPI sends and receives result in improved precision in the analysis results. One issue that arises with performing data-flow analyses on MPI programs is that the interprocedural control-flow graph ICFG is often irreducible due to call and return edges, and the MPI-ICFG adds further irreducibility due to communication edges. To provide an upper bound for iterative data-flow analysis complexity, one needs to have a general idea as to the depth of common flow graphs. Unfortunately, computing the depth of an irreducible graph is NP-complete.\n\nWe compare the depth-based iteration bounds for several MPI benchmarks with the actual number of iterations required for two data-flow analyses. We are able to compute the depth despite the worst-case exponential complexity of the depth-analysis algorithm by first reducing the reducible parts of the flow graph and then explore paths in the reduced graph. Our results show that on average the reduced graphs are 80% smaller than the original graphs and have 90% fewer cycle-free paths, resulting in a 10x faster algorithm. We also observe that although the number of iterations over the flow graph is bounded by the lattice height multiplied by the graph depth, the graph depth is clearly the dominating factor and provides a close approximation to the complexity of iterative\ndata-flow analysis over MPI programs.\n SUMMARY:Depth Analysis of MPI Programs UID:926 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100810T103000 DTEND;TZID=America/Chicago:20100810T120000 DTSTAMP:20130525T020110 LOCATION:Building 240, TCS Conference Center, 1404 & 1405, Argonne National Laboratory DESCRIPTION:Energy efficiency is a major concern in modern high performance computing system design. Today it is not uncommon for large scale computing systems and data centers to consume massive amounts of energy, typically 510 megawatts in powering and cooling. The enormous energy consumption results in large electricity bills and reduced system reliability due to increased heat emissions. The continuing need to increase the size and scale of these data centers means energy consumption potentially limits future deployments of high performance computing systems.\n\nImproving energy efficiency is a challenging task in high performance computing: application performance must not be adversely degraded when energy consumption is reduced. Nevertheless, in reality power reduction normally leads to performance degradation. In this talk, I will present a comprehensive framework that addresses this challenge. This framework enables indepth understanding of power, performance, and their correlations, and exploits opportunities and novel techniques to reduce power while maintaining performance. The results on real systems show that the framework can be used to identify opportunities for energy savings and improve energy efficiency by up to 20% with little to no impact on performance for a large spectrum of applications. \n\nShort Bio\n\nRong Ge is an assistant professor in the Department of Mathematics, Statistics and Computer Science at Marquette University. She received the PhD degree in computer science from Virginia Tech. Her research interests include performance modeling and analysis, parallel and distributed systems, energy efficient computing, high performance computing, and computational science. She is a member of the IEEE and the IEEE Computer Society, and a member of the ACM and Upsilon Pi Epsilon. Contact her at rong.ge@marquette.edu.\n\n\n SUMMARY:Models and Techniques for Energy Efficient High Performance Computing UID:927 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100809T150000 DTEND;TZID=America/Chicago:20100809T160000 DTSTAMP:20130525T020110 LOCATION:Bldg 240 Rm 4301, Argonne National Laboratory DESCRIPTION:This talk will discuss the use of the discrete adjoint for sensitivity analysis and uncertainty quantification. In the first part of the talk, the use of the adjoint for sensitivity analysis in hypersonic flow simulations will be detailed. The simulation of hypersonic flow is a subject of interest in many engineering fields and is particularly important for evaluating the performance of atmospheric re-entry vehicles. Hypersonic flow (roughly defined as Mach number greater than 5) is typically characterized by the presence of strong shocks, chemical reactions and the excitation of internal molecular energy modes, such as vibrational or electronic energy. The modeling of these phenomena depends on elaborate empirical models and experimentally measured constants. Quantifying the effects of these model parameters on relevant engineering quantities, such as drag or surface heating, is the basis of sensitivity analysis and can provide valuable insights for improving the predictive and design capability of a solver. In order to calculate sensitivity derivatives with respect to a large number of model parameters, a discrete adjoint approach is used. By exploiting similarities between the adjoint problem and the flow problem and making use of automatic differentiation, the flow adjoint can be calculated with minimal developer effort and computational expense roughly equivalent to that of the flow solve. With the sensitivity derivatives calculated, the relative importance of parameters for a given output can be compared. Additionally, these derivatives can be used for uncertainty quantification purposes by aiding in the construction of computationally inexpensive surrogate models. The use of gradient information in constructing inexpensive surrogate models will be the focus of the second part of this talk. The traditional approach to uncertainty quantification, particularly for hypersonic flow simulations, is Monte Carlo sampling. In order to collect reliable statistics for the output, thousands of code evaluations are required. For high fidelity simulations, this requirement can be prohibitively expensive and alternative approaches must be considered. One alternative is to replace the expensive code evaluation with an inexpensive model of the design space which can be sampled exhaustively. The particular models examined in this work are extrapolation based models and Kriging based models. For each of these models, the inclusion of gradient information can substantially reduce the number of code evaluations required to accurately build a model of the design space. In this talk, the use of these response models for quantifying uncertainty in hypersonic flows will be demonstrated. Time permitting, the use of combined regression/Kriging based models will be outlined and applications for nuclear engineering will be discussed. SUMMARY:Parameter Sensitivity and Uncertainty quantification for Engineering Simulations using the Discrete Adjoint UID:929 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100811T103000 DTEND;TZID=America/Chicago:20100811T120000 DTSTAMP:20130525T020110 LOCATION:Building 240, TCS Conference Center, 1404 & 1405, Argonne National Laboratory DESCRIPTION:Abstract: In scientific computing highly concurrent bursts of writing create unique challenges for parallel file system designers. In order to fulfill this need, file systems are being geared towards large scales and commodity storage. A common characteristic of parallel file systems is the object storage abstraction layer, a software implementation of the T10 OSD specification. Past work has established replicated object storage systems as effective with highly concurrent bandwidth bound large reads and writes. However, it has not been shown such a system performs well with highly concurrent latency bound metadata operations, which are often quite small.\n\nTo improve performance for the latter type of accesses, we have chosen to implement B-trees designed to exist within the object storage abstraction layer. Because of their efficacy in secondary storage and successful preliminary results we believe they will successfully improve performance for metadata operations. A number of challenges still exist, including the tuning of B-trees to the specific memory hierarchy in question, and difficulties involved when interfacing them with caches, which can further improve the performance of latency bound metadata operations.\n\nBio: Ellis Wilson III is currently an intern at Argonne National Laboratory and a PHD student at the Pennsylvania State University. He received his undergraduate degree from LaSalle University. His research interests include the simulation of highly parallel environments, multi-level cache replacement and partitioning policies, parallel file systems and their underlying metadata storage structures. He is a member of the IEEE and ACM. More details about Ellis Wilson are available at http://www.cse.psu.edu/~ehw111.\n SUMMARY:Improving Metadata Operation Performance on a Replicated Object Storage System UID:930 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100810T153000 DTEND;TZID=America/Chicago:20100810T163000 DTSTAMP:20130525T020110 LOCATION:Building 240, TCS Conference Center, 1404 & 1405, Argonne National Laboratory DESCRIPTION:Abstract:\nHeterogeneous clusters that add hardware accelerators to nodes are becoming more popular, due to trends in hardware as well as constraints in computing towards the exascale. Three of the top ten computers in the Top500 list use hardware accelerators, including the current #2 supercomputer. Achieving optimal performance on such machines requires the use of hybrid parallelization, combining shared memory at the nodal level and message passing in between nodes.\n\nIn this talk, I will motivate the need for hybrid parallelization and address some of the missing pieces in accelerator computing. I will present the results of a case study in hybrid parallelization using an MPI based unstructured mesh, explicit, finite element solver for computational solid mechanics. The domain decomposition used for unstructured FEM provides a natural hierarchy for hardware acceleration at the node level. A performance improvement of 20-30x has been demonstrated using a hybrid CUDA+MPI approach. Issues of data management will be discussed, as well as possible extensions to MPI to facilitate accelerator computing.\n\nBio:\nPiotr Fidkowski is a masters student in aerospace engineering at the Massachusetts Institute of Technology, working under Professor Raul Radovitzky. His research interests are in the field of computational solid mechanics, specifically in discontinuous Galerkin methods and GPGPU computing. SUMMARY:Exploring Hybrid Parallelization through Computational Solid Mechanics UID:931 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100813T153000 DTEND;TZID=America/Chicago:20100813T163000 DTSTAMP:20130525T020110 LOCATION:Building 240, TCS Conference Center, 1404 & 1405, Argonne National Laboratory DESCRIPTION:ABSTRACT:\nA very fast network is an essential part of any supercomputer. A quick look to the latest top 500 list shows that around half the clusters have other interconnect rather than Ethernet. IBM has its own proprietary network and an interface to access it. This library is called Low-level Application Programming Interface (LAPI) and consists in a set of message-passing functions based on a one-sided communication model. Additionally, LAPI provides an active message infrastructure that can accelerate communication between processes by only involving one of them in any transmission. The upcoming Blue Waters supercomputer will offer LAPI as one of the libraries to interact with the network.\n\nIn this talk, we will describe the challenges in designing and constructing a LAPI network layer for MPICH2. At the beginning, we will see how two-sided communication operations are implemented into MPICH2 library and how 4.5 us latency is achieved (compared to bare LAPI 4 us latency). Then, we will cover the opportunities LAPI opens up for improving one-sided communication operations and non-contiguous data transfers.\n\nSHORT BIOGRAPHY:\nEsteban Meneses received a B. Eng. degree in Computing Engineering (2001) and a M.Sc. degree in Computer Science (2007) from the Costa Rica Institute of Technology. Currently, he is a third year PhD student working with Prof. Laxmikant V. Kale at the Parallel Programming Laboratory in the University of Illinois at Urbana- Champaign. From 2007 to 2009, Esteban held a Fulbright-LASPAU scholarship for the first two years of his PhD program. His research interests span the areas of scalable fault tolerance and load balancing for supercomputing applications.\n SUMMARY:Design and Implementation of a LAPI network layer for MPICH2 UID:932 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100816T103000 DTEND;TZID=America/Chicago:20100816T113000 DTSTAMP:20130525T020110 LOCATION:Building 240 Seminar Room 4301, Argonne National Laboratory DESCRIPTION:We present a methodology for exploiting shared-memory parallelism within matrix computations by expressing linear algebra algorithms as directed acyclic graphs. Our solution involves a separation of concerns that completely hides the exploitation of parallelism from the code that implements the linear algebra algorithms. This approach to the problem is fundamentally different since we also address the issue of programmability instead of strictly focusing on parallelization. Using the separation of concerns, we present a framework for analyzing and developing scheduling algorithms and heuristics for this problem domain. As such, we develop a theory and practice of scheduling concepts for matrix computations. SUMMARY:Application of Dependence Analysis and Runtime Data Flow Graph Scheduling to Matrix Computations UID:934 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100813T093000 DTEND;TZID=America/Chicago:20100813T110000 DTSTAMP:20130525T020110 LOCATION:Building 240 Conference Room 4301, Argonne National Laboratory DESCRIPTION:Abstract: Magellan is a research and development effort to establish a nationwide scientific mid-range distributed computing and data analysis testbed. Tao and Phil will discuss the improvements they\'ve accomplished that will enhance the user experience. Tao\'s discussion will focus on his experience as a new Magellan user and his ability to provide solutions to unknown problems with Eucalyptus, as well as explain the effort to administer accounts on Magellan. Phil will discuss changes and improvements he\'s made to the Magellan wiki and provide an investigation of the different online forum solutions that compliment the current workflow of the magellan community. General improvements and how Eucalyptus can be integrated into the Argonne userbase system will also be discussed. SUMMARY:Improving the experience for the Magellan user UID:935 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100819T133000 DTEND;TZID=America/Chicago:20100819T143000 DTSTAMP:20130525T020110 LOCATION:Building 240, TCS Conference Center, 1404 & 1405, Argonne National Laboratory DESCRIPTION:Abstract:\nEnergy efficiency has risen up to be the predominant factor in the design of embedded and mobile hardware in the last decade. Energy management is now a key issue in the design and development of warehouse size machine rooms. Extensive finite element models are deployed to understand hot spots and devise cooling solutions in the machine rooms, often solving complicated set of PDEs using CFD techniques. Simultaneously, there is a growing trend in obtaining real-time temperature and humidity parameters by deploying several sensors. In this talk I will motivate the need for a holistic view of several physical parameters of the machine room beyond the common temperature, humidity and airflow parameters. I will describe our efforts in the design and implementation of a sensing envelope using open-source embedded hardware toolkits including Arduino and Gumstix. Next, I will briefly describe the design decisions involved in the realization of the envelope and discuss the embedded communication and networking mechanisms. Following some visuals of the physical parameters we have collected using the system in the form of color-maps and graphs, I will conclude the talk presenting some of our ideas on the physical representation and manipulation of the collected parameters.\n \nShort Biography:\nRajesh Sankaran is a PhD candidate in the Department of Electrical & Computer Engineering at Louisiana State University, Baton Rouge. He is a member of the Tangible Visualization group at the Center for Computation & Technology. Rajesh is co-advised by Dr. Brygg Ullmer & Dr. J (Ram) Ramanujam, and his research interests are in Embedded Systems and their applications in HCI. He is currently working on a modular, scalable, re-usable, and flexible hardware platform called Blades & Tiles as part of his PhD thesis. Rajesh maintains Cyber presence at http://www.cct.lsu.edu/~rajesh/ .\n\n SUMMARY:Acquisition of physical environment parameters in the machine room using open source embedded hardware. UID:937 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:00000000T103000 DTEND;TZID=America/Chicago:00000000T113000 DTSTAMP:20130525T020110 LOCATION:Building 240 / Conference Room 4301, Argonne National Laboratory DESCRIPTION:Engineering and physics applications frequently give rise to generalized eigenvalue problems of the form A x = \\lambda B x, where A is Hermitian and B is Hermitian positive definite. In structural dynamics, A and B are respectively the stiffness and mass matrices, and for discretizations of the time-independent Schroedinger equation, they are the Hamiltonian and overlap matrices. When a large percentage of the eigenpairs are sought, the first steps in the process are typically to find the Cholesky factor of B and transform the problem into a Hermitian standard eigenvalue problem. Unfortunately, the LAPACK and ScaLAPACK approach for this transformation is inherently unscalable and quickly becomes the dominant portion of the eigensolution. A new approach is demonstrated that yields a 20x speedup on two racks of Blue Gene/P.\n\nClick on the link below to add this seminar to your calendar. SUMMARY:A New Algorithm for Hermitian Generalized Eigenvalue Problems UID:938 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100823T103000 DTEND;TZID=America/Chicago:20100823T113000 DTSTAMP:20130525T020110 LOCATION:Building 240 / Conference Room 4301, Argonne National Laboratory DESCRIPTION:Engineering and physics applications frequently give rise to generalized eigenvalue problems of the form A x = \\lambda B x, where A is Hermitian and B is Hermitian positive definite. In structural dynamics, A and B are respectively the stiffness and mass matrices, and for discretizations of the time-independent Schroedinger equation, they are the Hamiltonian and overlap matrices. When a large percentage of the eigenpairs are sought, the first steps in the process are typically to find the Cholesky factor of B and transform the problem into a Hermitian standard eigenvalue problem. Unfortunately, the LAPACK and ScaLAPACK approach for this transformation is inherently unscalable and quickly becomes the dominant portion of the eigensolution. A new approach is demonstrated that yields a 20x speedup on two racks of Blue Gene/P.\n\nClick on the link below to add this seminar to your calendar. SUMMARY:A New Algorithm for Hermitian Generalized Eigenvalue Problems UID:939 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100902T120000 DTEND;TZID=America/Chicago:20100902T130000 DTSTAMP:20130525T020110 LOCATION:Cafeteria, Private Dining Rooms A&B, Argonne National Laboratory DESCRIPTION:http://www.phy.anl.gov/theory/astrolunch.html SUMMARY:Indirect Dark Matter Searches Using Gamma-Ray Telescopes UID:948 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100909T103000 DTEND;TZID=America/Chicago:20100909T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, Conference Room 4301, Argonne National Laboratory DESCRIPTION:Biological membranes are of critical importance for homeostasis in living cells. Far from being simple impenetrable barriers, such membranes allow the selective permeation of various pharmaceutically-relevant substances. The behavior of membrane-enclosed functional units can also play a role in sophisticated cellular processes that involve the coalescence and fusion of membrane-bound containers. This brief talk will outline a combination of experimental techniques and multi-scale theoretical models, which have been employed in an attempt to gain further insight into such systems. SUMMARY:Permeation and Attractive Forces in Biologically Relevant Membrane Models UID:949 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100908T103000 DTEND;TZID=America/Chicago:20100908T113000 DTSTAMP:20130525T020110 LOCATION:Building 240 Seminar Room 4301, Argonne National Laboratory DESCRIPTION:A three-level domain decomposition is considered. Bodies in contact with each other are divided into subdomains, which in turn are the union of elements. Using an approach based purely on FETI (finite element tearing and interconnecting) algorithms with only Lagrange multipliers as unknowns, which has been developed by the engineering community, does not lead to a scalable algorithm with respect to the number of subdomains in each body. We present a proof that such a method has a condition number which depends linearly on the number of subdomains across each body and logarithmically on the number of elements across each subdomain. We also propose a new method based on the saddle point formulation of the FETI methods with both displacement vectors and Lagrange multipliers as unknowns. The resulting system is solved with a block-diagonal preconditioner which combines the one-level FETI and the BDDC (balancing domain decomposition by constraints) methods. We show that this new method is scalable with respect to the number of subdomains. SUMMARY:Comparison of two domain decomposition methods for a linearized contact problem UID:941 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100907T033000 DTEND;TZID=America/Chicago:20100907T043000 DTSTAMP:20130525T020110 LOCATION:Building 240, TCS Conference Center, 1406 & 1407, Argonne National Laboratory DESCRIPTION:Abstract: As the system size and complexity continue to grow, a critical challenge facing high end computing is fault management of these systems. An increasing attention is to explore coordination among multiple software components to enhance system-wide resilience. Coordinated Infrastructure for Fault Tolerant Systems (CIFTS) enables system software components to share fault information with each other and adapt to faults in a holistic manner. However, currently CIFTS does not provide reliable messaging mechanism, which is critical for the correctness, effectiveness, and robustness of the coordination among multiple software components.\n\nIn this talk, we first describe the reliability mechanism of other messaging frameworks, and discuss the requirements and challenges in the reliable messaging for CIFTS. Then we introduce the design of the FTB reliable/guarantee mechanism. We also discuss the potential extensions to the FTB API to support reliable event publishing. Our mechanism does not need to block the publishers, which provides a simple and efficient manner for reliable messaging. Finally, preliminary experimental results show that our mechanism is helpful to improve the reliability of CIFTS with low overhead for the system software components. \n\nBio: Ziming Zheng is a PhD candidate at Illinois Institute of Technology. He received his BS and MS degrees from the College of Computer Science and Engineering, University of Electronic Science and Technology of China. His research interest is fault resilience for large scale systems. He was an intern at Argonne National Laboratory and Oak Ridge National Laboratory for CIFTS project. More details about Ziming Zheng are available at http://www.iit.edu/~zzheng11/. SUMMARY:Improving Reliability and QoS in CIFTS UID:951 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100901T103000 DTEND;TZID=America/Chicago:20100901T113000 DTSTAMP:20130525T020110 LOCATION:Building 240 / Conference Room 4301, Argonne National Laboratory DESCRIPTION:The Partitioned Global Address Space (PGAS) models bring together the best features of the shared memory programming model and the message passing model. While providing simple operations for one-sided communication, they expose the data distribution information to the applications for them to exploit locality. Though these models provide an efficient interface, their performance largely depends on how their implementations efficiently exploit the features provided by the underlying systems. Some of these implementations use other run-time systems to provide them the abstraction for remote memory access (RMA) operations. In such cases, the API that the run-time system exposes and its implementation play a crucial role in the overall performance of the PGAS model. This work focuses on Global Arrays, a library based PGAS model, and its run-time system, namely ARMCI, on the Blue Gene/P. First, the various features offered by the BG/P system and its low level communication API, DCMF, were explored and evaluated using micro-benchmarks. The impact of shared system resources on performance was studied at scale. Then, the limitations of the interface and implementation of the current AMRCI run-time were analyzed. Finally, the A1 library, a super-set of ARMCI, was designed, providing API that would allow Global Arrays to take advantage of the features and performance offered by the underlying system.\n\nBiography: Sreeram Potluri is a Graduate Student in the Department of Computer Science and Engineering at The Ohio State University. He is a member of the Network-Based Computing Laboratory lead by Dr. D. K. Panda. He had received his Bachelors degree in Computer Science and Engineering from the Jawaharlal Nehru Technological University, Hyderabad, India. His research interests include high speed interconnects, parallel programming models and high-end computing applications. His recent work includes optimizing AWP-ODC, a widely used seismic modeling application, using MPI-1 Non-blocking and MPI-2 RMA semantics on large scale InfiniBand clusters. This work was published at ICS\'10 and is part of the application\'s entry as a finalist for the 2010 Gordon Bell Prize. Sreeram is involved in the design and development of MVAPICH2, an open-source high-performance implementation of MPI-2 over InfiniBand and 10GigE/iWARP. This software is currently used by over 1,100 organizations in 56 countries. SUMMARY:Design and implementation of an optimized one-sided communication run-time for Global Arrays on Blue Gene/P UID:943 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100909T103000 DTEND;TZID=America/Chicago:20100909T113000 DTSTAMP:20130525T020110 LOCATION:Building 240 / Room 1172, Argonne National Laboratory DESCRIPTION:This presentation will focus on large-scale direct simulations, as an essential tool to understand the physics of turbulence in fluid dynamics. An introducing part will be devoted to review the short history of direct numerical simulation, from pioneering works to nowadays computations on supercomputers. Key figures will be established to understand how computational scientists can improve our knowledge of turbulence in the future. Thus, a part of our recent work on wall turbulence will be presented. A new massively parallel Navier Stokes solver, SPLATS, has been developed, allowing computations on Blue Gene/P architectures, using up to 32K cores. We revisit the problem of Perot and Moin [2] as a step towards second-order-closure turbulence modeling. The main purpose of this experiment is to elucidate the energy transfer in the near-wall region, essential to build appropriate closure models. A better-suited configuration has been exploited, in which shearless turbulence is interacting with a solid wall and usual production mechanisms of the turbulence are replaced with a random forcing. Therefore, we isolate mechanisms related to pure-diffusive turbulence. Our approach bring new conclusions: we identify the skewness of the velocity field in the outer region as the main responsible of the energy redistribution in the vicinity of the wall [1], and propose new length scales associated with this transfer. SUMMARY:Understanding Near-Wall Turbulence Using Large Scale Direct Simulations UID:950 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100827T103000 DTEND;TZID=America/Chicago:20100827T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, Conference Room 4301, Argonne National Laboratory DESCRIPTION:Recent advances in storage technologies and high performance interconnects have made possible in the last years to build, more and more potent storage systems that serve thousands of nodes. The majority of storage systems of clusters and supercomputers from Top 500 list are managed by one of three scalable parallel file systems: GPFS, PVFS, and Lustre. Parallel applications currently suffer from a significant imbalance between computational power and available I/O bandwidth. Additionally, the hierarchical organization of current Petascale systems and of the envisioned Exascale platforms contributes to an increase of the I/O subsystem latency. In these hierarchies, file access involves pipelining data through several networks with incremental latencies and higher probability of congestion.\n\nWe present a novel generic parallel I/O architecture for both clusters and supercomputers. Our design is aimed at large-scale parallel architectures with thousands of compute nodes. Besides acting as middleware for existing parallel file systems, our architecture provides on-line virtualization of storage resources. Our solution is based on a multi-tier cache architecture and asynchronous data staging strategies hiding the latency of data transfer between cache tiers. This work targets to reduce the file access latency perceived by the data-intensive parallel scientific applications by multi-layer asynchronous data transfers. In order to accomplish this objective, our techniques leverage the multi-core architectures by overlapping computation with communication and I/O in parallel threads. Prototypes of our solutions have been deployed on both clusters and Blue Gene supercomputers. \n\nBio:\nFrancisco Javier Garcia Blas has been a Teaching Assistant of University Carlos III (Spain) since 2005. He has also cooperated in several projects with researchers from various high performance research institutions including HLRS (funded by HPC-Europe program) and Argonne National Laboratory. He is currently involved in various projects on topics including parallel I/O and parallel architectures. He received the MS degree in Computer Science of University Carlos III of Madrid in 2007. In 2010 he received a PhD in Computer Science from University Carlos III. SUMMARY:Object-based multi level file caching for massively parallel supercomputers UID:944 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20101215T150000 DTEND;TZID=America/Chicago:20101215T160000 DTSTAMP:20130525T020110 LOCATION:Bldg 240 Conference Center 1404-1405, Argonne National Laboratory DESCRIPTION:Graphics in LaTeX can be frustrating. In this talk, I will introduce a LaTeX package called TikZ, created by the author of the popular Beamer presentation package. TikZ can create 2d vector graphics from *within* a LaTeX file, thereby eliminating the need for any external graphics editors. TikZ has an advantage over PSTricks because it can create either PS or PDF files, and also TikZ can interact with Beamer commands. This talk is mainly geared towards intermediate LaTeX users who want to overcome the pain of importing graphics. \n\n\nInterested people can install the latest version (I will be using features from the latest version, so I suggest upgrading) of TikZ from here: \nhttp://www.texample.net/tikz/builds/ \nInstalling LaTeX packages is out of the scope of this talk. SUMMARY:An Introduction to TikZ UID:945 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100922T150000 DTEND;TZID=America/Chicago:20100922T160000 DTSTAMP:20130525T020110 LOCATION:Bldg 240 Rm 4301, Argonne National Laboratory DESCRIPTION:OpenAD is a tool for automatic differentiation (AD) of numerical computer programs. The standardized OpenMP programming interface can be used to define parallel code regions inside of C, C++ or Fortran code. Until now OpenAD takes no advantage of given OpenMP pragmas inside of input code. This work should get OpenAD to process OpenMP pragmas and to exploit the knowledge\nabout the parallelism inside the input code. \n\nWe will go through the different phases wherein OpenAD transforms its input code per forward or per reverse mode. Especially the reverse mode must be considered according to data flow reversal. Therefore we will consider an example to illustrate the concepts. SUMMARY:Shared Memory Multiprocessing and Automatic Differentiation UID:946 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100929T150000 DTEND;TZID=America/Chicago:20100929T160000 DTSTAMP:20130525T020110 LOCATION:Bldg 240 Rm 4301, Argonne National Laboratory DESCRIPTION:Power market problems that are characterized by the presence of several generators competing amongst themselves necessitate modeling by means of a game theoretic approach. In addition, fluctuations in market demand require additional addressing with regard to demand and ramp constraints (dynamics). This work focuses on modeling and computation of equilibria arising in a supply function setting. Congestion and network constraints are modeled by means of the power distribution factor and the presence of the ISO (Independent System Operator) as another agent. The Nash game arising from the above setting is modeled under two levels of rationality. The bounded rationality framework gives rise to a complementarity problem whereas the completely rational framework leads to a more challenging class of problems namely the EPEC (Equilibrium Problem with Equilibrium Constraints). The EPEC is further modeled as a more tractable nonlinear programming problem. Nature of equilibria arising from these instances are studied and comapred. Effects of ramp constraints, planning horizon and wind power penetration on equilibria are studied. \n SUMMARY:Dynamic Power Markets - A game theoretic approach UID:947 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100910T103000 DTEND;TZID=America/Chicago:20100910T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, TCS Conference Center, 1404 & 1405, Argonne National Laboratory DESCRIPTION:Abstract:\nWith the ever increasing number of components in high-end computing systems, component failures have become common place rather than exceptional. Currently, the most prevalent approach to overcome such failures on large scale machines is to periodically checkpoint the applications to stable storage. It has been reported that the overhead imposed by the checkpoint operation is already about 20% of the overall execution time and it is anticipated that in the coming years this ratio may reach over 50%. The substantial overhead introduced by checkpointing stems from the fact that there is a significant imbalance between the computational power and the available I/O bandwidth in current high-end computing systems. Reducing checkpoint image size without significant computational overhead is therefore a major concern.\n\nIn this talk, we first present a similarity based checkpointing mechanism in the context of virtual machine replication and show how the same approach may be adapted to large scale HPC systems. We detail our design and implementation of a distributed checkpoint compression algorithm which aims at finding similar patterns within the memory content of the whole parallel application. We show preliminary results on the degree of similarity we found and the computational overhead of our proposed compression method on various applications running on BlueGene/P.\n\nBio:\nBalazs Gerofi is a second year Ph.D student at The University of Tokyo. His main interest involves operating systems and fault tolerant computing. He received his MS degree in Computer Science of Vrije Universiteit of Amsterdam in 2006.\n SUMMARY:A similarity based distributed checkpointing algorithm for reducing checkpoint image size in parallel computing systems UID:952 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110316T150000 DTEND;TZID=America/Chicago:20110316T160000 DTSTAMP:20130525T020110 LOCATION:Bldg 240 Conference Center 1404-1405, Argonne National Laboratory DESCRIPTION:Nonconvex quadratic programming (QP) is an NP-hard problem that optimizes a general quadratic function over linear constraints. This paper introduces a new global optimization algorithm for this problem, which combines two ideas from the literature--finite branching based on the first-order KKT conditions and polyhedral-semidefinite relaxations of completely positive (or copositive) programs. Through a series of computational experiments comparing the new algorithm with existing codes on a diverse set of test instances, we demonstrate that the new algorithm is an attractive method for globally solving nonconvex QP. SUMMARY:Globally Solving Nonconvex QPs via Completely Positive Programming UID:1017 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110415T150000 DTEND;TZID=America/Chicago:20110415T160000 DTSTAMP:20130525T020110 LOCATION:Bldg 240 Conference Center 1404-1405, Argonne National Laboratory DESCRIPTION:Analysis of large networks is an important exploratory tool for many disciplines such as biology, epidemiology and sociology. However, most analysis algorithms focus on static networks, whereas in reality these systems are dynamic. In this presentation, I will discuss some of the challenges in developing algorithms for dynamic networks in the context of some of our recent research on community detection and evaluating the effect of changes on large networks. SUMMARY:Challenges in Analyzing Large Dynamic Networks UID:1019 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110118T103000 DTEND;TZID=America/Chicago:20110118T113000 DTSTAMP:20130525T020110 LOCATION:Building 240 Seminar Room 4301, Argonne National Laboratory DESCRIPTION:We consider a combinatorial optimization problem that generalizes the minimum disagreement halfspace problem; we seek to minimize the number of misclassifications of a weighted voting classifier, plus a penalty proportional to the density of the vector of weights. We prove that the optimum is at least as hard to approximately compute as minimum disagreement halfspace for a large class of penalty parameters. After formulating the problem as a mixed integer program (MIP), we show that common \"soft margin\" linear programming formulations for constructing weighted voting classifiers are equivalent to the standard LP relaxation of our formulation. We illustrate that this standard LP relaxation can be very weak, with an exponential lower bound on the potential integrality gap. We then prove that augmenting the optimization problem with certain simple valid inequalities tightens the relaxation considerably, yielding a linear upper bound on the gap for all values of the penalty parameter that exceed a sensible lower bound. Finally, we describe a linear programming based boosting algorithm that dynamically generates both cuts and columns for solving our relaxation. We demonstrate the classification performance of our proposed algorithm with experimental results. SUMMARY:Sparse Weighted Voting Classifier Selection and its LP Relaxation UID:1021 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20101110T150000 DTEND;TZID=America/Chicago:20101110T160000 DTSTAMP:20130525T020110 LOCATION:Bldg 240 Conference Center 1404-1405, Argonne National Laboratory DESCRIPTION:Particle systems have been proposed as a means to detect and sample features in 3D image datasets, for example to find airways and lobe fissures in CT scans of the lungs, and white matter in diffusion MRI. Particles move relative to each other and throughout the image domain in order to minimize a carefully designed energy term, so that the converged configuration represents a uniform spatial sampling of the relevant image features. A significant challenge to more widespread use of particle systems in these applications is the long computation time required to run the system to convergence. Poor choices for system parameters and for the energy minimization method (stochastic search vs gradient descent) will increase the total computational cost of finding a converged configuration. Although the system will tend to convergence (in the sense of *energy* minimization) even with a wide range of system parameters, an interesting object of study is minimizing the total computational *cost* of finding the minimization. The total computational cost depends on costs of various individual operations (finding energy, finding gradients, re-applying constraints, etc) whose relative frequency depends on choices of minimization method. I will describe early efforts in modeling computational costs of particle system minimization, with the hope of applying expertise from the optimization research community to the development of faster tools for biomedical image analysis.\n SUMMARY:Optimizing Particle Systems for Image Feature Sampling UID:955 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20101117T150000 DTEND;TZID=America/Chicago:20101117T160000 DTSTAMP:20130525T020110 LOCATION:Bldg 240, 4301, Argonne National Laboratory DESCRIPTION:In this study, we follow the insights we obtain from practical tests with a number of popular NLP solvers on a sample set of infeasible problem instances. We argue that efficient infeasibility detection requires mechanisms that are carefully designed and embedded to the overall solution methodologies. In particular, we examine two classes of nonlinear optimization algorithms, with two distinct approaches to handle infeasibility. First, we consider the interior point methods, and build an infeasibility detection mechanism based on a switching approach in which the algorithm is either in a feasibility mode or in an optimization mode. Second, we focus on the active set methods that gradually change their emphasis from optimality to feasibility through an exact penalty approach. SUMMARY:Nonlinear Programming Algorithms and Infeasibility Detection UID:956 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20101015T103000 DTEND;TZID=America/Chicago:20101015T113000 DTSTAMP:20130525T020110 LOCATION:Bldg 240 Conference Center 1404-1405, Argonne National Laboratory DESCRIPTION:An algorithm for the solution of phase equilibrium at constant pressure and temperature, also known as (P,T) flash, is presented. The solution of this problem corresponds to the global minimum of the system Gibbs free energy, a multi-dimensional, highly non-linear and non-convex function. It has been shown that this can be interpreted using duality theory [1]. In this talk, we present a formulation of (P,T) phase equilibrium as a dual optimisation problem in the volume-composition space, translated away from the Gibbs free energy to the Helmholtz free energy. Working in this space brings considerable benefits [2], especially when using complex equations of state (EOS) that are higher than cubic functions in volume and are formulated in the Helmholtz free energy, such as SAFT (statistical associating fluid theory). In addition, the Helmholtz free energy surface is smooth and therefore numerically better behaved than that for the Gibbs free energy. We develop an algorithm for the solution of the problem that is based on a combination of local and global optimisation, in which the number of subproblems solved globally is kept at a minimum. Using this approach, one is guaranteed to predict the number of phases present at equilibrium, along with their properties, without any need for initial guesses, or indeed any a priori knowledge about the behaviour of the system in question. The method is applicable to multi-component mixtures and to the calculation of any kind of fluid phase behaviour (e.g. vapour-liquid (VLE), liquid-liquid (LLE), vapour-liquid-liquid (VLLE) etc.) and to any EOS explicitly expressed in terms of the Helmholtz free energy. Several algorithmic options are investigated to improve computational efficiency. Results are presented for the VLE and VLLE for mixtures modelled with the augmented van der Waals EOS, a non-cubic EOS that incorporates the Carnahan-Starling Equation to account for the repulsive interactions. Further examples are presented for VLE and VLLE in polymer systems, modelled with the SAFT-HS EOS. Fluid phase equilibrium calculations for polymer systems are notoriously difficult, and convergence problems are often encountered, even with very good initial guesses. The new method is found to be reliable in all cases.\n\nReferences \n[1] A. Mitsos and P. I. Barton, AIChE Journal, (2007), 53, 2131.\n[2] N.R. Nagarajan, A. S. Cullick, and A. Griewank, Fluid Phase Equilibria, (1991), 62, 191.\n SUMMARY:The reliable solution of (P,T) phase equilibrium and stability in the volume-composition space UID:957 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100920T133000 DTEND;TZID=America/Chicago:20100920T150000 DTSTAMP:20130525T020110 LOCATION:Building 240 Conference Room 4313, Argonne National Laboratory DESCRIPTION:Motivated by the industry transition from single to many-core chip designs, this talk outlines a framework that leverages heterogeneous computing to tackle many-body dynamics problems. The heterogeneous computing infrastructure discussed relies on a combination of CPUs and GPUs, the latter regarded as co-processors, or accelerators, capable of delivering substantial efficiency gains in simulating many-body dynamics problems. Examples of such problems include granular material dynamics; fluid-structure interaction; and molecular dynamics analysis. After a brief review of emerging trends in hardware design, the discussion concentrates on a heterogeneous computing template (HCT) that our collaboration is putting together in order to capitalize on the industry shift towards many-core processor designs. The talk highlights several components of the HCT research/implementation project: a Dynamic Data Exchange Protocol (DDEP), a proximity calculation framework, partitioned iterative solvers, and a distributed post-processing capability. Outcomes of this ongoing collaboration are succinctly showcased in the context of three applications: a liquid sloshing problem, a tracked-vehicle mobility on granular terrain analysis, and a collision detection benchmark study. The talk concludes with a brief discussion of experimental efforts meant to validate the simulations enabled by the proposed computational dynamics framework. SUMMARY:A CPU/GPU Heterogeneous Computing Framework for Computational Dynamics Applications UID:958 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100924T103000 DTEND;TZID=America/Chicago:20100924T120000 DTSTAMP:20130525T020110 LOCATION:Building 240 TCS Conference Center, Rm. 1404&1405, Argonne National Laboratory DESCRIPTION:In a reactor environment, intense particle irradiation induces microchemical and microstructural evolution and phase changes in fuel and structural materials. Such lattice-level material changes are strongly influenced by the operating temperature and stress conditions, and they result in severe dimensional changes and mechanical and thermal property degradation. Over the past several decades, researchers have gathered large databases on various aspects of radiation effects in materials, and developed a wide range of models to describe the mechanisms and microstructure processes leading to the observed macroscopic radiation effects. Although these efforts have advanced our knowledge about materials in irradiation environments, a unified view of how to describe radiation effects in materials is still largely missing. In this presentation, I will highlight a field theoretic approach to modeling and understanding the microstructural changes that occur in materials as a result of irradiation. This approach is based on the concurrent defect dynamics and microstructure and microchemical evolution (including phase changes) in irradiated materials. Specific attention will be given to the mesoscale, where a systematic derivation of a phase field model from the principles of non-equilibrium thermodynamics will be shown and demonstrated for void and gas bubble nucleation and growth and microchemical evolution in alloys. The physical, mathematical, and computational aspects of the phase field modeling framework for irradiated materials will be discussed.\n\n\nAnter El-Azab is a Professor of Computational Materials Science at Florida State University. His research interests are in the field of theoretical and computational modeling of defects in materials. His current research areas include microstructure and microchemical evolution in irradiated alloys and ceramics, morphological instabilities in heterogeneous and nanoscale materials, dislocation dynamics and mesoscale deformation of metals, thermal transport in materials, and reduced order models in materials modeling. He worked for six years as a senior scientist at Pacific Northwest National Laboratory; there he joined the computational mechanics, applied mathematics, and the interfacial and nanoscience groups. He joined Florida State University in the fall of 2004 as an Associate Professor of Mechanical Engineering then transitioned to a joint position with the former School of Computational Science and Mechanical Engineering department two years later. Currently, he is a full professor with the Department of Scientific Computing and is a faculty member of Florida State’s graduate program in Materials Science. \n\nAnter obtained his Ph.D. degree in Nuclear Engineering in 1994 at the University of California, Los Angeles, and his B.S. and M.S. degrees also in Nuclear Engineering at the University of Alexandria, Egypt. SUMMARY:Microstructural Stability in Irradiated Materials UID:959 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100924T103000 DTEND;TZID=America/Chicago:20100924T120000 DTSTAMP:20130525T020110 LOCATION:Building 240 TCS Conference Center, Rm. 1404, Argonne National Laboratory DESCRIPTION:In a reactor environment, intense particle irradiation induces microchemical and microstructural evolution and phase changes in fuel and structural materials. Such lattice-level material changes are strongly influenced by the operating temperature and stress conditions, and they result in severe dimensional changes and mechanical and thermal property degradation. Over the past several decades, researchers have gathered large databases on various aspects of radiation effects in materials, and developed a wide range of models to describe the mechanisms and microstructure processes leading to the observed macroscopic radiation effects. Although these efforts have advanced our knowledge about materials in irradiation environments, a unified view of how to describe radiation effects in materials is still largely missing. In this presentation, I will highlight a field theoretic approach to modeling and understanding the microstructural changes that occur in materials as a result of irradiation. This approach is based on the concurrent defect dynamics and microstructure and microchemical evolution (including phase changes) in irradiated materials. Specific attention will be given to the mesoscale, where a systematic derivation of a phase field model from the principles of non-equilibrium thermodynamics will be shown and demonstrated for void and gas bubble nucleation and growth and microchemical evolution in alloys. The physical, mathematical, and computational aspects of the phase field modeling framework for irradiated materials will be discussed.\n\n\nAnter El-Azab is a Professor of Computational Materials Science at Florida State University. His research interests are in the field of theoretical and computational modeling of defects in materials. His current research areas include microstructure and microchemical evolution in irradiated alloys and ceramics, morphological instabilities in heterogeneous and nanoscale materials, dislocation dynamics and mesoscale deformation of metals, thermal transport in materials, and reduced order models in materials modeling. He worked for six years as a senior scientist at Pacific Northwest National Laboratory; there he joined the computational mechanics, applied mathematics, and the interfacial and nanoscience groups. He joined Florida State University in the fall of 2004 as an Associate Professor of Mechanical Engineering then transitioned to a joint position with the former School of Computational Science and Mechanical Engineering department two years later. Currently, he is a full professor with the Department of Scientific Computing and is a faculty member of Florida State’s graduate program in Materials Science. \n\nAnter obtained his Ph.D. degree in Nuclear Engineering in 1994 at the University of California, Los Angeles, and his B.S. and M.S. degrees also in Nuclear Engineering at the University of Alexandria, Egypt. SUMMARY:Microstructural Stability in Irradiated Materials UID:960 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20101006T150000 DTEND;TZID=America/Chicago:20101006T160000 DTSTAMP:20130525T020110 LOCATION:Bldg 240 Rm 4301, Argonne National Laboratory DESCRIPTION:We will discuss the problem of factorizing multi-way data sets with the goal of capturing the underlying latent structure of the data. Such data can be represented using tensors, or multidimensional arrays, and latent structure can be exposed using low-rank factorizations, akin to principal component analysis used in two-way data analysis. Focus will be on one of the most well-known tensor factorizations, CANDECOMP/PARAFAC (CP), one of the higher order analogs of the singular value decomposition. Discussion will include scalability and robustness of various methods for computing CP factorizations, and application of these factorizations in the areas of chemometrics, citation analysis, network traffic analysis, and neuroscience will be demonstrated. This is joint work with Tamara G. Kolda, Evrim Acar , Brett Bader and Morten Mřrup. SUMMARY:Multiway data analysis using tensor decompositions UID:961 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100923T150000 DTEND;TZID=America/Chicago:20100923T160000 DTSTAMP:20130525T020110 LOCATION:Bldg 240 Conference Center 1404-1405, Argonne National Laboratory DESCRIPTION:Photovoltaic energy conversion is arguably the most promising option for supplying renewable, carbon-neutral energy on a global scale. In order to reach grid parity, however, costs must be reduced substantially. Inexpensive materials generally exhibit efficiencies too low for practical application, but by controlling the morphology on the nanoscale there are opportunities to achieve significant improvements in this area. Block copolymers, which naturally self-assemble into periodic ordered nanostructures, can be utilized in diverse ways to control morphology, ranging from active layers to structure directors to a combination of these methodologies. SUMMARY:Controlling Nanostructure in Organic and Hybrid Photovoltaics UID:962 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20101103T150000 DTEND;TZID=America/Chicago:20101103T160000 DTSTAMP:20130525T020110 LOCATION:Bldg 240 Conference Center 1404-1405, Argonne National Laboratory DESCRIPTION:This talk describes a method for constructing Cartesian meshes for embedded boundary algorithms by using a ray-tracing technique. In this approach, each mesh cell is distinguished as being inside, outside, or on the boundary of the input geometry, which is determined by firing rays parallel to x/y/z coordinates. The most expensive process of the embedded boundary mesh generation, an edge-geometry intersection test, is performed for the group of edges on a fired ray line together, which decreases the computational complexity of the whole method significantly. Produced boundary cells also have edge-cut fraction information and volume cut fraction information for each material. This work is implemented to be enable to directly import various CAD-based solid model formats and as an open-source code to be used easily in many engineering simulation fields. SUMMARY:EBMesh: An Embedded Boundary Meshing Tool UID:963 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20101027T150000 DTEND;TZID=America/Chicago:20101027T160000 DTSTAMP:20130525T020110 LOCATION:Bldg 240 Conference Center 1406-1407, Argonne National Laboratory DESCRIPTION:There are two talks: \n\nTalk by Michael McCourt\n===============\nProgress in Parallel Implicit Methods for Tokamak Edge Plasma Modeling\n\nPerformance of prototype tokamak fusion devices depends sensitively on\ncharacteristics of the edge plasma between the hot core and surrounding walls. The edge plasma includes an especially wide range of physical time scales such that implicit numerical algorithms can substantially improve overall computational efficiency. This presentation introduces some of the benefits and challenges of parallel implicit solution strategies, with emphasis on preconditioned Newton-Krylov methods in the UEDGE and BOUT++ applications. We will discuss multiphysics issues in the context of the FACETS project, which is developing a multiphysics, parallel application to enable\nmodeling from the material wall to the plasma core. We also will explain how this fusion research is motivating new capabilities in the PETSc library to better handle strong coupling between two or more distinct PDE-based mathematical models.\n\nTalk by Hong Zhang\n============\nSparse Triangular Solves for ILU Revisited: Data Layout Crucial to Better Performance\n\n A key to good processor utilization for sparse matrix computations is storing the data in the format that is most conducive to fast access by the memory system. In particular, for sparse matrix triangular solves the traditional compressed sparse matrix format is poor, and minor adjustments to the data structure can increase the processor utilization dramatically. Such adjustments involve storing the L and U factors separately and storing the U rows \"backwards\" so that they are accessed in a simple streaming fashion during the triangular solves. Changes to the PETSc libraries to use this modified storage format resulted in over twice the floating-point rate for some matrices. SUMMARY:1) Progress in Parallel Implicit Methods for Tokamak Edge Plasma Modeling and 2) Sparse Triangular Solves for ILU Revisited: Data Layout Crucial to Be UID:964 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110119T150000 DTEND;TZID=America/Chicago:20110119T160000 DTSTAMP:20130525T020110 LOCATION:Bldg 240 Conference Center 1404-1405, Argonne National Laboratory DESCRIPTION:In this talk, we discuss strategies to minimize energy consumption in building systems by integrating sensor information, data-based models, model predictive control techniques. We discuss current challenges and implications of adopting energy management technologies at a large-scale. Finally, we report operational details, status, and expected savings of an energy management system recently deployed in the Theory and Computing Sciences building at Argonne. SUMMARY:Real-Time Optimization for Energy Management of Building Systems UID:965 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20101007T103000 DTEND;TZID=America/Chicago:20101007T113000 DTSTAMP:20130525T020110 LOCATION:Building 240 Seminar Room 4301, Argonne National Laboratory DESCRIPTION:Due to physical constraints and design difficulties, the computer processor architectures have shifted to multi-core and many-core based approaches in recent years. Today, chip multi-processor (CMP)products are replacing uni-processors in every segment of the market, including super computers, with the capability to integrate hundreds of cores onto individual chips. This provides tremendous potential towards peta/exa-scale computing platforms.\nHowever, people in both academia research and industry are still seeking proper ways to make efficient and effective use of these processor architectures. The impact to the user software and system software development is largely under exploration. However, little has been settled for large scale CMPs architecture, system software, and applications, which also means research opportunities in this field.\nIn this talk, I will briefly go over the development and architecture issues of CMP architectures and its applications. Through the example of Tilera processors, I would like to share some insights about software development and performance optimization onto mid-large scale CMPs. I will also present a cache partitioning technique for reducing memory bandwidth requirement on CMP platforms. SUMMARY: Cross-Layer Customization onto High-Performance Multi-Core/Many-Core Processors UID:967 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110210T150000 DTEND;TZID=America/Chicago:20110210T160000 DTSTAMP:20130525T020110 LOCATION:Bldg 240 Conference Center 1404-1405, Argonne National Laboratory DESCRIPTION:It is well known that grid overlapping methods represent a powerful tool for simulating engineering problems in complex geometries, since they alleviate mesh generation process, provide a convenient way of resolution control and even allow for modeling different physics in multiphysics systems. In this talk, we concentrate on stability of grid overlapping methods, namely how the choice of temporal discretization of interface terms influences stability properties. Matrix stability analysis is performed on a model problem of one-dimensional heat equation on overlapping grids. We obtain the estimates of the spectral radius of the corresponding discrete matrix operators, allowing us to judge the stability of different order schemes. Influence of iterations on stability properties is also investigated. Numerical experiments are then presented relating obtained stability bounds to the observed numerical values. We also describe the implementation of grid overlapping techniques into the spectral element solver NEK5000 and introduce some interesting problems which can be efficiently solved with the multidomain version of the code.\n SUMMARY:On stability of grid overlapping methods UID:968 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20101011T103000 DTEND;TZID=America/Chicago:20101011T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, TCS Conference Center, 1406 & 1407, Argonne National Laboratory DESCRIPTION:For some years ahead, peta-scale high performance computing systems which have over peta-FLOPS performance, are being built and installed in US, Japan and Europe. In order to make use of such system, the software to support parallel programming in peta-scale system is indispensable. We are carrying on a project for parallel programming language for a petascale distributed memory system. In this project, we are designing a directive-based language extension and programming model, called XcalableMP (XMP for short), which allows users to develop parallel programs for distributed memory systems easily and tune the performance by having minimal and simple notations. XcalableMP provides a global view programming partially inherited from HPF, and also includes Co-array feature for local view programming. In this talk, I will give our experience from HPF (High Performance Fortran), and describe the idea behinds XcalableMP. \n\nBio:\nMitsuhisa Sato received the M.S. degree and the Ph.D. degree in information science from the University of Tokyo in 1984 and 1990. He was a senior researcher at Electrotechnical Laboratory from 1991 to 1996, and a chief of Parallel and distributed system performance laboratory in Real World Computing Partnership, Japan, from 1996 to 2001, leading the Omni OpenMP compiler project. Currently, he is a professor of Graduate School of Systems and Information Engineering, University of Tsukuba. He is a director of Center for Computational Sciences, University of Tsukuba since 2007. His research interests include computer architecture, compilers and performance evaluation for parallel computer systems, OpenMP and parallel programming. SUMMARY:XcalableMP: A performance-aware scalable parallel programming language for distributed memory system --- beyond PGAS models --- UID:969 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20101013T130000 DTEND;TZID=America/Chicago:20101013T140000 DTSTAMP:20130525T020110 LOCATION:Argonne National Lab, TCS/Room 5C2 (5172), The University of Chicago, Searle 240A, 5735 S. Ellis Ave. DESCRIPTION:Abstract:\nUnderstanding the factors that drive land-use change, and developing better methods for projecting future land-use change, become vitally important in the climate change context. The 3rd Assessment Report of the UN Intergovernmental Panel on Climate Change (IPCC) noted that \"the emissions scenarios considered in future climate change studies need to integrate high resolution representations of land use change\" and that increased coupling among the various relevant components—such as mitigation and adaptation responses to climate change, and climate response to land use—should be included in a consistent framework for integrated assessment (Jones et al. AR-3 2001, Chapter 3). While much work has been done to include global forecasts of natural land cover types such as forests and deserts at high spatial resolutions, little has been done to incorporate into these forecasts the socioeconomic drivers of land-use change, such as agriculture, forestry, and urbanization on a global scale. To address this urgent need for improved global land-use forecasts, we are developing a versatile high-resolution companion to the CIM-EARTH computable general equilibrium (CGE) global economics framework. The Partial-Equilibrium Economic Landuse (PEEL) model generates consistent forecasts of the land-use change (LUC) that results from changing growing conditions, technology, resource availability, and demands.\n\nWe illustrate the usefulness of the model by applying the prototype to the important and highly topical question of the implications of an aggressive 1st generation biofuels policy on LUC. We note that the high resolution (5 arcminutes, or approximately 10km on a side) of the model elements allows for the direct integration of climate and socioeconomic factors at the scale of relevant climate variation, without the need for aggregation (and the associated potential for significant information loss). We conclude with a discussion of model limitations and future directions, focusing especially on data needs and potential new data sources.\n SUMMARY:Integrating the Socio-economic and Physical Drivers of Land-use Change at Climate relevant Scales: an Example with Biofuels UID:970 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100930T130000 DTEND;TZID=America/Chicago:20100930T140000 DTSTAMP:20130525T020110 LOCATION:Building 240, Conference Center Rooms 1404-1405, Argonne National Laboratory DESCRIPTION:A New Approach to Distance Visualization of Large Volume Data\n\nI will introduces a new approach to distance visualization of large-scale data anywhere, anytime, on any display device, enabling exploration in multi-dimensional space without re-rendering and access to the raw data. This new approach is based on the concept of explorable image, which is a compact representation of data. Explorable images may be generated in situ during the simulation or in a postprocessing step, and then used as a previewing solution, a remedy to data compression, or a solution for making visualization that would be impossible/impractical to make after the simulation. SUMMARY:A New Approach to Distance Visualization of Large Volume Data UID:971 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20101201T150000 DTEND;TZID=America/Chicago:20101201T160000 DTSTAMP:20130525T020110 LOCATION:Bldg 240 Conference Center 1404-1405, Argonne National Laboratory DESCRIPTION:We present a software package, called ColPack, encapsulating an array of fast and effective algorithms for a variety of graph coloring and associated problems arising in the efficient computation of large, sparse Jacobian and Hessian matrices using automatic differentiation. Several of the coloring problems supported by ColPack also find applications in many areas outside derivative computation; a few examples include: concurrency discovery in parallel computing, frequency assignment in wireless networks, scheduling, facility location, and compiler design. In this presentation, we will give an overview of the functionalities available in ColPack, describe the package\'s major algorithms with a focus on the common framework within which they are designed, and present performance results, including comparisons of solutions yielded by the various coloring algorithms with appropriate lower bounds. As an example of an application that benefited from sparsity exploitation using ColPack, we will discuss a large-scale PDE-constrained optimization problem in a chromatographic separation process in chemical engineering. Finally, we will mention related efforts on the development of parallel algorithms for a subset of the coloring problems in ColPack. SUMMARY:Algorithms and Software for Sparse Derivative Computation and Beyond UID:972 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110420T150000 DTEND;TZID=America/Chicago:20110420T160000 DTSTAMP:20130525T020110 LOCATION:Bldg 240 Conference Center 1404-1405, Argonne National Laboratory DESCRIPTION:The performance mismatch between computing and I/O components of current-generation HPC systems has made I/O a bottleneck for scientific applications. It is therefore critical to make data movement as efficient as possible, and, to facilitate simulation-time data analysis and visualization to reduce the data volume written to storage. Enabling these will be of paramount importance to glean insights from simulations. \n\nIn this talk, we will elucidate the analytics and I/O challenges faced by applications at extreme-scale and present various promising approaches. Next, we will give an overview our work in GLEAN - A flexible framework for data-analysis and I/O at extreme scale which leverages data semantics of applications and fully exploits the diverse system topologies and characteristics. We will discuss the performance of GLEAN for simulation-time analysis and I/O with DOE INCITE and ESP simulations at scale on leadership class system SUMMARY:Towards simulation-time data analysis and I/O acceleration on leadership-class systems UID:973 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20101020T103000 DTEND;TZID=America/Chicago:20101020T113000 DTSTAMP:20130525T020110 LOCATION:Building 240 / Conference Room 4301, Argonne National Laboratory DESCRIPTION:The environmental impact of aviation is a subject of academic research and practical interest that has caught the attention of scientists since the late 90’s because of its implication in atmospheric and climate change. This research encompasses diverse scientific disciplines that are based on different physical models and characteristic temporal and spatial scales. Recent evaluations indicate that links are missing today between these disciplines to evaluate with precision the effects of aircraft emissions on the physical and chemical state of the atmosphere. This talk summarizes the scientific activity carried out by the author in this subject over the last few years. The focus is laid on the modeling and numerical simulation of aircraft emissions, as well as on the methods developed to represent and quantify the atmospheric perturbations at different scales, and to incorporate the effects of such perturbations into global models.\n\n[schedule.ics] SUMMARY:Modeling and Simulation of the Environmental Impact of Aircraft Emissions UID:974 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20101108T150000 DTEND;TZID=America/Chicago:20101108T160000 DTSTAMP:20130525T020110 LOCATION:Bldg 240 Conference Center 1404-1405, Argonne National Laboratory DESCRIPTION:The recent emergence of clouds with large, virtualized pools of compute and storage resources raises the possibility of a new compute paradigm for scientific research. With virtualization technologies, consolidation of scientific workflows presents a promising opportunity for energy and resource cost optimization, while achieving high performance. We have developed pSciMapper, a power-aware consolidation framework for scientific workflows. We view consolidation as a hierarchical clustering problem, and introduce a distance metric that is based on interference between resource requirements. A dimensionality reduction method (KCCA) is used to relate the resource requirements to performance and power consumption. We have evaluated pSciMapper with both real-world and synthetic scientific workflows, and demonstrated that it is able to reduce power consumption by up to 56%, with less than 15% slowdown. Our experiments also show that scheduling overheads of pSciMapper are low, and it can scale well for workflows with hundreds of tasks. SUMMARY:Power-Aware Consolidation of Scientific Workflows in Virtualized Environments UID:975 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20101022T103000 DTEND;TZID=America/Chicago:20101022T113000 DTSTAMP:20130525T020110 LOCATION:Building 240 / Conference Room 4301, Argonne National Laboratory DESCRIPTION:During the last decades, the studies on molecular interactions at microscopic level of biological systems have emigrated to technological applications beyond the life sciences, as to nanomaterials areas to develop efficient devices in transport and storage of energy. Particularly, the interaction of small biomolecules with metals has caught the interest of various research groups. For example, the binding of a single metal atom to DNA strands might induce changes in the local reactivity, bond breaking and variations in optical properties, among others. \n\nIn this seminar, I will show advances on my investigations regarding the complexation of metal atoms to organic molecules that have moieties commonly present in biomolecules. These studies include the application of non-linearoptical properties and vibrational spectra approaches that I implemented in the NWChem.\n SUMMARY:Biomolecules Complexed to Metal Clusters and Their Optical Activity UID:976 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20101026T150000 DTEND;TZID=America/Chicago:20101026T160000 DTSTAMP:20130525T020110 LOCATION:Bldg 240 Conference Center 1404-1405, Argonne National Laboratory DESCRIPTION:The dynamics of micro and nanoscale objects in fluid are stochastic and correlated due to inherent Brownian fluctuations and the long-range motion of the surrounding fluid. This has been exploited in a number of technologies including atomic force microscopy, microrheometry, and in the development of biosensors. In this talk a thermodynamic approach using the fluctuation-dissipation theorem is discussed. It is shown that the stochastic dynamics of elastic objects in fluid can be quantified from straightforward deterministic calculations that can be done on a single workstation. Using this approach, the stochastic dynamics of micro and nanoscale cantilevers are quantified for experimentally relevant conditions including cantilever arrays, complex geometries, and finite sized domains. Absolute predictions of the auto and cross-correlations of the equilibrium fluctuations in cantilever displacement are used to yield limits on the force and time sensitivity of fluctuation based\ndetection methods. The cross-correlations in cantilever displacement reveal interesting dynamics resulting from the interplay between the viscous and potential fluid dynamics around an oscillating object at low Reynolds number. SUMMARY:The stochastic dynamics of micro and nanoscale elastic objects in fluid UID:978 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20101101T150000 DTEND;TZID=America/Chicago:20101101T160000 DTSTAMP:20130525T020110 LOCATION:Bldg 240 Conference Center 1404-1405, Argonne National Laboratory DESCRIPTION:We are interested in numerical methods for identifying models that describe optimal processes in nature, such as human gait. Our goal is to identify a model that describes the gait of Cerebral Palsy (CP) patients. A CP gait model provides the possibility to classify and evaluate gaits, and most importantly, it affords a basis for treatment planing. Mathematically, a CP model can be described by an optimal control problem (OCP) which includes parameters that cannot be derived theoretically and need to be determined form observation data. This leads to a challenging parameter estimation problem whose constraints include an optimal control problem. We present three different approaches (a derivative-free optimization technique, a bundle method, and an SQP approach) for solving this problem and provide numerical results for two benchmark problems. SUMMARY:Estimating Parameters in Optimal Control Models UID:981 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20101018T081500 DTEND;TZID=America/Chicago:20101018T173000 DTSTAMP:20130525T020110 LOCATION:Building 240 Conference Room 1405, Argonne National Laboratory DESCRIPTION:The ALCF Early Science Program Kick-Off Workshop was our first gathering of the key players in the exciting endeavor to build the future of computational science.\n\n * IBM representatives presented a first look at Blue Gene/Q hardware and software, including compilers and messaging.\n\n * ALCF staff provided hands-on help for those new to Blue Gene to facilitate the development efforts key to the ESP.\n\n * ESP project teams mapped out and discussed plans for their first six months of development.\n\n * IBM, ALCF, and project representatives presented performance measurements.\n\n * The Blue Gene/Q Tools, Libraries, and Programming Models project presented their efforts and participated in detailed planning sessions with IBM.\n SUMMARY:See complete details: http://workshops.alcf.anl.gov/esp10/agenda/ UID:982 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20101104T110000 DTEND;TZID=America/Chicago:20101104T120000 DTSTAMP:20130525T020110 LOCATION:Building 240 / Conference Room 4301, Argonne National Laboratory DESCRIPTION:With the new generation of semiconductors based on nanoscaled heterostructures, modelling the electronic transport in such devices using the Boltzmann classical approaches is no longer available. The contacts effects in such components cannot be neglected. Moreover, classical or semi-classical methods do not take into account the energy levels modification when under a high electrical field. All these effects can be taken into consideration using the Density Functional Theory (DFT) within the Non-Equilibrium Green Functions Formalism (NEGFF) and effective mass approximation. \n\nWe will discuss the electronic transport in an AlGaN/GaN heterostructure in a ballistic regime where no carrier is subject to the scattering centres, by briefly presenting the principal aspects of the DFT and the NEGF formalism. Then we will show results of the implementation of this formalism in a sequential FORTRAN code, the main approximations, the main algorithm, the compute kernel and finally our strategy in parallelizing this code in order to study larger systems. \n\nClick on this link to add this event to your calendar.\n\n[schedule.ics] SUMMARY:Electronic Transport in Mesoscopic Devices Using the Non-Equilibrium Green Functions Formalism: From Theory to Parallel Code UID:983 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:00000000T130000 DTEND;TZID=America/Chicago:00000000T140000 DTSTAMP:20130525T020110 LOCATION:Building 240/TCS Conference Center/Room 1407, Argonne National Laboratory DESCRIPTION:In this talk we will present two analysis tools that have been developed and deployed for exploring large scale simulations. Common to both tools is their ability to provide interactive results to the application scientist. \n\nThe first tool was designed for topological analysis of toroidal magnetic fields commonly found in confined magnetic fusion simulations and utilizes a combination of on demand and parallel generation integral curves to obtain its interactivity. While the second tool was designed for query based exploration of particle based codes commonly found in fusion, combustion, and laser acceleration and obtains its interactivity by utilizing FastBit to preprocess the data for fast look up. \n\nOur talk will focus on the science behind the tools as well as the hurdles to full deployment to the application scientist.\n\n\n SUMMARY:Deployment of Analysis Tools for Exploring Large Scale Simulations UID:984 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20101110T130000 DTEND;TZID=America/Chicago:20101110T140000 DTSTAMP:20130525T020110 LOCATION:Building 240 Conference Room 1407, Argonne National Laboratory DESCRIPTION:With the advent of the era of petascale supercomputing and the drive to exascale, there is a pressing need to address the problem of analyzing and visualizing massive peta/exascale-sized results. \n\nIn this presentation, I will discuss progress on a number of approaches we are pursuing at Los Alamos including science-driven feature extraction integrated with in-situ analysis, multi-resolution out-of-core data streaming and interactive rendering on the supercomputing platform. \n\nThese approaches are placed in context by the emerging area of data-intensive supercomputing.\n SUMMARY:Data-Intensive Analysis and Visualization on Numerically-Intensive Supercomputers UID:985 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20101111T133000 DTEND;TZID=America/Chicago:20101111T143000 DTSTAMP:20130525T020110 LOCATION:Building 240 / Conference Room 4301, Argonne National Laboratory DESCRIPTION:The King Abdullah University of Science and Technology (KAUST) is a research-oriented graduate-level institution that houses seven core research labs. One is these fundamental research facilities is the KAUST Supercomputing Laboratory (KSL) which currently hosts the most powerful supercomputer in the Middle East. KSL\'s mission is to further the ambitious research goals of KAUST by providing researchers world-class supercomputing hardware and human support. \n\nThis presentation will overview two capability projects currently over development at KSL. One project is the development of a turbulent combustion code traditionally used as a capability code into a capability application. The second project is the design and creation of a highly scalable acoustics wave propagation application by KSL computational scientists. Both projects produced interesting results in terms of performance of the Blue Gene/P platform and commonly used libraries used in the BG/P platform.\n\nClick on this link to add the seminar to your calendar. SUMMARY:Developing capability applications on the Blue Gene/P UID:986 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20101202T110000 DTEND;TZID=America/Chicago:20101202T120000 DTSTAMP:20130525T020110 LOCATION:Building 240, Room 4301, Argonne National Laboratory DESCRIPTION:Seminar Title: Integrative Modeling of Genetic, Transcriptional Regulatory, and Metabolic Networks\n\nABSTRACT: Biological networks are highly useful to integrate data of diverse types across multiple scales of space and time. In my talk, I will address three major classes of networks and the biology they help elucidate. First, I will discuss our new method for integrating statistically learned transcriptional regulatory networks with biochemically detailed metabolic networks, called probabilistic regulation of metabolism (PROM) (Chandrasekaran and Price, PNAS, 2010). We have used PROM to reconstruct state-of-the-art integrated regulatory-metabolic models for the widely studied Escherichia coli and Saccharomyces cerevisiae, as well as for the pathogen M. tuberculosis. I will also briefly discuss implications for metabolic modeling from emerging data on single-cell protein copy number differences (Kim and Price, Physical Review Letters, 2010). Then, I will present our recent reconstruction of the first genome-scale, quantitative predictive model of a transcriptional regulatory network for a brain, done for the honey bee Apis mellifera. Finally, I will discuss the derivation of a network of putative epistatic relationships by studying highly enriched co-occurring orthologs (comologs) across all sequenced bacteria, and demonstrate how such information can be useful for interpreting datasets such as protein-proteininteractions (manuscript currently under review at Science). Taken together, network analysis – whether in individual cells or across the time scale of evolution – provides a powerful tool for harnessing high-throughput data to drive biological discovery. SUMMARY:Integrative Modeling of Genetic, Transcriptional Regulatory, and Metabolic Networks UID:991 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20101203T130000 DTEND;TZID=America/Chicago:20101203T140000 DTSTAMP:20130525T020110 LOCATION:Building 240 Seminar Room 4301, Argonne National Laboratory DESCRIPTION:Porting data-intensive, componentized applications on large scale distributed computing infrastructures is not trivial. Many different approaches continue to be developed, implemented and practiced in this area. As the application demands and infrastructure capabilities evolve, the approaches taken to automate face challenge to keep up and a gap forms between the two.\n\nBridging such a gap poses challenges at different levels. The challenge at the end-user level is a need to express the application\'s logic and data flow requirements from a non-technical domain. At the infrastructure level, it is a challenge to port the application such that a maximum exploitation of the underlying resources can takes place.\n\nThe workflow technology is a promising means to bridge such a gap. Workflows enable distributed application deployment by recognizing the application component\'s inter-connections and the flow among them.\n\nHowever, workflow expressions and engines need enhancements to meet the challenges outlined. Facilitation of a concise expression of parallelism, data combinations and higher level data structures in a coherent fashion is required. This work targets to fullfil these requirements. The work is driven by the use-cases in the field of medical image processing domain. Various strategies are developed to efficiently express asynchronous and maximum parallel execution of complex flows by providing concise expression and enactments interfaced with large scale ditributed computing infrastructures.\n\nThe main contributions of this research are: a) A rich workflow language with two-way expression and fruitful results from the experiments carried out on enactment of medical image processing applications workflows on the European Grid Computing Infrastructure; and b) Extension of an existing workflow environment (Taverna) to interface with the Grid Computing Infrastructures.\n SUMMARY:DataFlow-Intensive Enactments on Distributed Computing Infrastructures UID:993 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20101209T103000 DTEND;TZID=America/Chicago:20101209T113000 DTSTAMP:20130525T020110 LOCATION:Building 240 Conference Room 1416, Argonne National Laboratory DESCRIPTION:Three decades of Moore’s law and dramatic advances in mathematics, algorithms, software, and parallelism have enabled the computing sciences’ rapid growth in importance to the advancement of science and engineering. It is now widely accepted that computational methods have joined theory and experiment as the “third pillar of science”. Computing has transformed the economy and society with equal vigor, driving business efficiency, globalization and whole new forms of economic and social activities. \nComputation – and more broadly information technology -- is poised to remain an engine of dramatic change. However, indications are the coming decade will be characterized by disruptive changes quite different than the past decade. We outline a few of the major disruptive changes (compute, storage, IOT, heterogeneity, cloud/large-scale, etc.) which are on the horizon. We will discuss how these changes create deep research challenges, and exciting new technical and scientific opportunities. SUMMARY:Disruptive Challenges and Opportunities for the Computing Sciences UID:996 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20101216T103000 DTEND;TZID=America/Chicago:20101216T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, TCS Conference Center, 1406 & 1407, Argonne National Laboratory DESCRIPTION:Parallel programming models like MapReduce have allowed companies to easily process information at scales of hundreds of petabytes and beyond. MapReduce and other popular industrial approaches, however, do not suffice for scientific datasets and analysis problems. The increasing size of datasets creates a pressing need in the visualization community for a general and scalable framework. In this talk, I discuss recent advances in a data analysis framework that aid in progressing towards the goal of a parallel programming model. Some key components that I discuss are the I/O techniques that are used by this framework along with how they are also being used in other large scale applications like parallel particle tracing. I also will cover the parallel sorting functionality that is currently supported in the framework and how I managed to scale it to 32 K processes of BGP. I show how the current framework can be applied to various large scale problems like climatic analysis (using over a TB of MODIS satellite data) and flow field analysis.\n\nWes Kendall is a graduate research assistant under Jian Huang at the University of Tennessee, Knoxville. His primary interest is large data analysis and visualization. He has interned at Oak Ridge National Labs, spent two summers with Argonne National Labs, and recently spent the summer with the Google MapReduce team. He has been in close collaboration with Argonne National Labs since his internships and has authored publications with ---them at Supercomputing and the Eurographics Symposium on Parallel Graphics and Visualization.\n SUMMARY:Towards Generalized Parallel Programming Models for Large Scientific Data Analysis and Visualization UID:999 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110216T150000 DTEND;TZID=America/Chicago:20110216T160000 DTSTAMP:20130525T020110 LOCATION:Bldg 240 Conference Center 1404-1405, Argonne National Laboratory DESCRIPTION:Reverse Monte Carlo (RMC) simulations are widely used to develop models of the atomistic structure of disordered materials from x-ray and neutron scattering data. Since experimental information is collected from different length-scales by various techniques (total scattering, small angle scattering, x-ray absorption fine structure, etc) it is desirable to develop multiscale models that unify all this information into a single very large atomistic model. Traditional RMC is typically limited to short lengthscales of a few tens of angstroems, while multiscale models are aimed to include lengthscales upto one micron. Algorithmic and computational challenges on the way of such an extension of RMC will be discussed in the presentation. SUMMARY:Algorithmic and computational challenges of very large Reverse Monte Carlo simulations UID:1005 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110112T150000 DTEND;TZID=America/Chicago:20110112T160000 DTSTAMP:20130525T020110 LOCATION:Bldg 240 Conference Center 1404-1405, Argonne National Laboratory DESCRIPTION:An introduction to PETSc\'s GPU implementation, the PETSc Matlab interface, as well as new splitting-based preconditioner infrastructure. SUMMARY:New Additions to PETSc UID:1015 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110112T103000 DTEND;TZID=America/Chicago:20110112T113000 DTSTAMP:20130525T020110 LOCATION:Building 240 Seminar Room 4301, Argonne National Laboratory DESCRIPTION:We present an algorithm for preprocessing a class of\nstochastic shortest-path problems on networks that have no negative cost cycles, almost surely. Our method adds utility to existing frameworks by significantly reducing input problem sizes and thereby increasing computational tractability. Given random costs with finite lower and upper bounds on each edge, our algorithm removes edges that cannot be in any optimal solution to the deterministic shortest-path problem, for any realization of the random costs. Although this problem is NP-complete, our algorithm efficiently preprocesses nearly all edges in a given network. We provide computational results both on sparse networks from PSPLIB - a well-known project evaluation and review technique library [Kolisch, R., A. Sprecher. 1996. PSPLIB—A project scheduling problem library. Eur. J. Oper. Res. 96(1) 205–216] - and dense synthetic ones: on average, less than 0.1% of the edges in the PSPLIB instances and 0.5% of the edges in the dense instances remain unclassified after preprocessing. SUMMARY:Preprocessing Stochastic Shortest-Path Problems UID:1023 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110223T150000 DTEND;TZID=America/Chicago:20110223T160000 DTSTAMP:20130525T020110 LOCATION:Bldg 240, Room 4301, Argonne National Laboratory DESCRIPTION:Recent advances in quantitative imaging allow unprecedented views into whole organisms of humans or test animals, specific organs or cellular chemistry in vivo. Novel imaging modalities make possible the scientific investigation of spatio-temporal distribution of therapeutic drugs or velocity fields of body fluids like blood or cerebrospinal fluid. Our group has integrated the accelerating capabilities of quantitative imaging techniques with rigorous computational fluid mechanics methods to study complex reaction and transport phenomena occurring in biological systems. \nThis presentation will demonstrate the need for massive computing power for addressing open questions in intracranial dynamics and drug transport. The first example fuses advanced medical imaging modalities with rigorous computational techniques to quantify the interaction of cerebrospinal fluid (CSF) dynamics with cerebral blood and soft deformable brain tissues. A fully coupled dynamic model of the entire human brain predicts intracranial pressures, CSF flow pulsations as well as brain tissue displacements in normal and pathological conditions such as hydrocephalus. The second example discusses large scale CFD simulations of invasive delivery of drugs in anisotropic brain tissue for the treatment of neurodegenerative diseases. Finally, we present our approach for simulation and control of cerebral blood flow based in a massive network model of the entire cerebral vasculature. \nImage reconstruction techniques are demonstrated to create a seamless bridge from real patient-specific anatomical spaces to large computational meshes to perform realistic simulations of blood and cerebrospinal fluid flow, or drug distribution in anisotropic tissues. The case studies will also highlight limitations of current computational power and the potential knowledge gain that could be achieved with more massive computations. SUMMARY:The need for super-computations in biomedical simulation of the brain UID:1027 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110208T150000 DTEND;TZID=America/Chicago:20110208T160000 DTSTAMP:20130525T020110 LOCATION:Bldg 240 Conference Center 1404-1405, Argonne National Laboratory DESCRIPTION:Tuning code for a specific system has often been necessary, but rarely enjoyable or easy. With many variants of a given architecture now being used in machines, for example the many implementations of the x86 instruction set, tuning has become even more difficult. In this talk, I will outline our work on the Active Harmony Auto-tuning system. I will present some results showing that we can improve the performance of real codes, and that the best configuration can vary quite a bit even among very similar CPUs. Our tool is capable of both online (tuning for a single execution) and offline (tuning using training runs) operation. I will also describe our core search algorithm, Parallel Rank Order. SUMMARY:Friends Don\'t Let Friends Tune Code UID:1029 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110120T103000 DTEND;TZID=America/Chicago:20110120T113000 DTSTAMP:20130525T020110 LOCATION:Bldg 240, Room 4301, Argonne National Laboratory DESCRIPTION:Tone languages like Mandarin Chinese use tones (specific pitch patterns) to distinguish syllables which are otherwise ambiguous. Tones in Mandarin Chinese have been shown to carry as much information as vowels [1]. However, several contextual factors in continuous speech make it challenging to achieve successful tone recognition. First, coarticulation between adjacent tones can compromise the realization of underlying tone targets. Second, phrase, sentence and topic boundaries can also affect pitch; pitch variation has been successfully employed to perform sentence and story segmentation. Third, speaker differences, especially gender differences, make it necessary to scale tone targets to compensate for individual variation.\n\nI present two approaches to model contextual effects on tone production in continuous speech. These approaches achieve state-of-the-art tone recognition performance on Mandarin Chinese Broadcast News data. The first approach focuses on modeling the coarticulation between adjacent tones. It manipulates a landmark-based vowel detection system to locate the most reliable tone production regions and to remove those regions affected by coarticulation. This approach shows a 15% improvement over two previously published tone recognition frameworks. The second approach employs sequential graphical models with Conditional Random Fields (CRF) to encode tone variation due to phrase, sentence and topic level intonation. We found that not only do different tones vary under each of these intonational conditions, but the choice of graphical model structure can also impact performance. Finally, I will briefly talk about possible application of these approaches to general machine learning challenges, especially structural learning on network behavior and attack pattern analysis [2,3].\n\n[1] Dinoj Surendran, Gina-Anne Levow, and Yi Xu. Tone recognition in mandarin using focus. Proceedings of Interspeech/ICSLP 2005, 2005.\n[2] Detecting Anomalies in Network Traffic Using Maximum Entropy Estimation. Yu Gu, Andrew McCallum and Don Towsley. Internet Measurement Conference, 2005\n[3] Layered approach using conditional random fields for intrusion detection. Kapil Kumar Gupta, Baikunth Nath and Ramamohanarao Kotagiri IEEE Trans. Dependable and Secure Computing, Vol. 7, No. 1, Jan-Mar 2010 SUMMARY:Contextual Modeling in Continuous Speech Tone Recognition UID:1031 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110117T103000 DTEND;TZID=America/Chicago:20110117T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, TCS Conference Center, 1404 & 1405, Argonne National Laboratory DESCRIPTION:Petascale machines with hundreds of thousands of cores are being built. These machines have varying interconnect topologies and large network diameters. Computation is cheap and communication on the network is becoming the bottleneck for scaling of parallel applications. Network contention, specifically, is becoming an increasingly important factor affecting overall performance. Most parallel applications have a certain communication topology. Mapping of tasks in a parallel application based on their communication graph, to the physical processors on a machine can potentially lead to performance improvements. Placement of communicating tasks on nearby physical processors can minimize the distance traveled by messages and reduce the chances of contention.\n\nPerformance improvements through topology aware placement for applications such as NAMD and OpenAtom are used to motivate this work. Building on these ideas, I will present algorithms and techniques for automatic mapping of parallel applications to relieve the application developers of this burden. The hop-bytes metric is proposed for the evaluation of mapping algorithms as a better metric than the previously used maximum dilation metric. The main focus of this work is on developing topology aware mapping algorithms for parallel applications with regular and irregular communication patterns. The automatic mapping framework is a suite of such algorithms with capabilities to choose the best mapping for a problem with a given communication graph. More details on my research available at: http://charm.cs.illinois.edu/~bhatele/phd/\n\nBio:\nAbhinav received a B. Tech. degree in Computer Science and Engineering from I.I.T. Kanpur (INDIA) in May 2005 and M. S. and Ph. D. degrees in Computer Science from the University of Illinois at Urbana-Champaign in 2007 and 2010 respectively. He is a post doctoral research associate working with Professors Kale and Gropp at Illinois. His research is centered around topology aware mapping and load balancing for parallel applications. He is also interested in performance analysis and optimization of parallel applications and studying algorithmic feasibility to exascale. Abhinav is an ACM/IEEE George Michael HPC Fellow (2009). He received the David J. Kuck Outstanding MS Thesis Award in 2009, a Distinguished Paper Award at Euro-Par 2009 for his mapping work and the David J. Kuck Outstanding PhD Thesis Award for 2011.\n\n SUMMARY:Automating Topology Aware Mapping for Supercomputers UID:1033 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110118T103000 DTEND;TZID=America/Chicago:20110118T113000 DTSTAMP:20130525T020110 LOCATION:Building 240 Seminar Room 4301, Argonne National Laboratory DESCRIPTION:We consider a combinatorial optimization problem that generalizes the minimum disagreement halfspace problem; we seek to minimize the number of misclassifications of a weighted voting classifier, plus a penalty proportional to the density of the vector of weights. We prove that the optimum is at least as hard to approximately compute as minimum disagreement halfspace for a large class of penalty parameters. After formulating the problem as a mixed integer program (MIP), we show that common \"soft margin\" linear programming formulations for constructing weighted voting classifiers are equivalent to the standard LP relaxation of our formulation. We illustrate that this standard LP relaxation can be very weak, with an exponential lower bound on the potential integrality gap. We then prove that augmenting the optimization problem with certain simple valid inequalities tightens the relaxation considerably, yielding a linear upper bound on the gap for all values of the penalty parameter that exceed a sensible lower bound. Finally, we describe a linear programming based boosting algorithm that dynamically generates both cuts and columns for solving our relaxation. We demonstrate the classification performance of our proposed algorithm with experimental results. SUMMARY:Sparse Weighted Voting Classifier Selection and its LP Relaxation UID:1035 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110120T150000 DTEND;TZID=America/Chicago:20110120T160000 DTSTAMP:20130525T020110 LOCATION:Building 240, Conference Room 4301, Argonne National Laboratory DESCRIPTION:A unifying goal of the research done within the Brookhaven National Laboratory\'s (BNL) Atmospheric Sciences Division is to understand the role that aerosols and clouds play in the Earth\'s climate system and to better represent aerosol-cloud-climate interactions in models predicting possible climate change. Towards this goal, BNL has undertaken an end-to-end effort beginning with the development of a theoretical framework describing the growth of cloud droplets from nucleation through to the remote sensing of cloud and precipitation processes near the end of cloud lifecycle. This presentation will emphasize remote sensing of the cloud-to-drizzle transition and parameters of well-developed precipitation through the use of millimeter cloud radar Doppler spectra observations and newer scanning cloud radar opportunities. SUMMARY:Cloud Processes Research UID:1039 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110131T133000 DTEND;TZID=America/Chicago:20110131T143000 DTSTAMP:20130525T020110 LOCATION:Building 240 / Conference Room 4301, Argonne National Laboratory DESCRIPTION:With the increasing demand to reduce carbon dioxide as a greenhouse gas emitted from energy systems such as pulverized-coal power plants, attention is paid to retrofitting these energy systems with carbon capture technologies. Among these technologies, oxy-fuel combustion (where the fuel is burned in an atmosphere of oxygen and recycled flue gas rather than air) is a promising ‘easy-to-capture’ option. It is important, however, to predict and minimize any performance penalty due to oxy-fuel retrofitting before actual implementation. Computational fluid dynamics (CFD) simulations with accurate modeling enable us to predict the expected characteristics of combustion, mixing, and heat transfer, and therefore optimize the retrofitting process. These combustion simulations are challenging due to the large number of coupled phenomena and consequently the large number of models needed, which include turbulence, gas chemistry (turbulence-chemistry interaction, reaction kinetics, and chemistry solver), surface chemistry (heterogeneous reactions), gas-particle dynamics (particle tracking and turbulent dispersion), phase change (evaporation and devolatilization), and thermal radiation (radiative transfer equation and absorption coefficient). These models, in turn, warrant assessment/verification studies to identify applicability limitations, necessary modifications, and possible improvements. This is the main topic of the seminar.\n\nFrom the various modeling aspects in combustion, the seminar considers gas chemistry and radiation. The seminar presents an assessment/verification study of reaction kinetics and chemistry solvers through the simulations of a turbulent syngas (carbon monoxide and hydrogen) flame, where two chemistry solvers (embedded Runge-Kutta and semi-implicit Bulirsch-Stoer) are used and predictions of eight reaction mechanisms (global and elementary) are compared. The seminar also presents a study analyzing four new weighted-sum-of-gas-gases models (WSGGMs) to evaluate the radiative absorption coefficient in oxy-fuel combustion, and how they perform differently from a classical air-fuel WSGGM. Two techniques (stepwise and interpolation) to implement the WSGGM are compared, through test cases with rectangular-box geometry and non-uniform temperature distribution.\n\nClick on this link to add this event to your calendar.\n SUMMARY:Gas Chemistry and Radiation Modeling for Syngas and Oxy-Fuel Combustion UID:1055 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110127T150000 DTEND;TZID=America/Chicago:20110127T160000 DTSTAMP:20130525T020110 LOCATION:TCS Building 240, Room 4301, Argonne National Laboratory DESCRIPTION:Abstract\nRevealing the ecological principles that shape natural communities is a major challenge of the post-genomic era. The recent publication of 118 metabolic models of bacterial species across the tree of life allows inferring and analyzing the whole set of 6903 pairwise competitive and cooperative interactions predicted to occur between these species. By crossing computational predictions with ecological data derived from 2801 different natural samples we conduct the first systematic study of the interactions between ecologically co-occurring, randomly-distributed, and mutually-exclusive pairs of species. We find that the level of competition between two species, their prospects to be involved in cooperative interactions and the similarity of their positioning in the communal-network of interactions are associated with their prospects to co-exist in nature. In support of the (controversial) role of competition in community assembly, we show that niche-exclusion and competition are associated features. Most predicted cooperative interactions are asymmetrical give-and-take relations, corresponding to the documented prevalence of syntrophic interactions within bacterial consortia. They are found to form significantly many close cooperative multi-species cycles between community members, indicating that they are beneficial at the community level. Finally, experiments inducing co-growth shifts in co-culture testify that our computational framework can be used for the design of synthetic consortia optimized towards a given application, including biodegradation and probiotics. SUMMARY:Towards revealing the design principles of bacterial communities UID:1059 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110323T150000 DTEND;TZID=America/Chicago:20110323T160000 DTSTAMP:20130525T020110 LOCATION:Bldg 240, 4301, Argonne National Laboratory DESCRIPTION:This talk presents a new response surface methodology, dynamic trees, for learning, optimization, and sequential design -- applications where Gaussian processes (GPs) have reigned supreme. Dynamic trees are thrifty on space and time: no need to store or invert big matrices. They are flexible: natively accommodating nonstationarity or heteroskedasticity. And they are inherently sequential, which means sequential design decisions (like for optimization) can be turned around quickly. Other benefits include the ability to deal with categorical predictors and responses, and to decompose partial dependencies by main, first order, and total effects. The talk will focus on the the dynamic tree methodology, inference by sequential Monte Carlo, and finally on design/optimization heuristics by active learning with illustrations. It will also highlight an R package implementing the methods, called dynaTree, which is available on CRAN. For more details see: http://arxiv.org/abs/0912.1586 SUMMARY:Dynamic Trees for Response Surface Learning and Optimization UID:1057 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110126T150000 DTEND;TZID=America/Chicago:20110126T160000 DTSTAMP:20130525T020110 LOCATION:Bldg 240, 4301, Argonne National Laboratory DESCRIPTION:For decades parallel computing has been the focus of intense research and development in selected fields, and numerous large-scale parallel applications have been developed. SPMD via MPI has been a dominant approach to parallelism to date, but this approach alone will be insufficient going forward. Presently we are on the threshold of mass deployment of parallelism across most application areas, but the path to developing these applications is uncertain. There are many programming models, languages and architectures from which to pick, and the number of choices is growing. In this presentation we discuss some of the principles of parallel application development that have produced today’s codes, and how we can address these principles going forward. We also discuss what much change in order to move forward and give ideas for developing parallel applications now that will have sustained value in the future. SUMMARY:Building the Next Generation of Parallel Applications UID:1061 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110215T103000 DTEND;TZID=America/Chicago:20110215T113000 DTSTAMP:20130525T020110 LOCATION:Building 240 Seminar Room 4301, Argonne National Laboratory DESCRIPTION:Rule-based program transformation employs term rewriting strategies to algorithmically control rewrite sequences. The primary abstractions of rewriting strategies -- rewrite rules, combinators and term traversals -- provide basic yet powerful constructs enabling complex program transformation. Due to the programmability of rewrites, errors are often made because of incorrect compositions or incorrect applications of rewrites to a term within a strategic rewriting program. In practice, ill-formed strategies are the Achilles heel of strategic rewriting and present a major obstacle to the effective application of strategies to large complex problems. We propose a type analysis geared towards term rewriting strategies to detect and remove errors statically. Novel aspects of this analysis are (1) more precise type-checking enabled by the data type constructor-based typing of terms and patterns, and (2) integration of compositional behavior of strategies to model arguments and results of rewrite rule compositions. The primary benefit of such analysis is automated detection of ill-formed strategies, which in turn can significantly reduce the development and testing of program transformations. The analysis is demonstrated on TL -- a representative transformation language supporting all of the strategic abstractions. We also discuss several applications of program transformation such as automated adaptation of Java libraries toward non-standard JVM profiles and annotation-based transformation for performance optimization. SUMMARY:Type Analysis of Rewrite Strategies UID:1075 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110204T133000 DTEND;TZID=America/Chicago:20110204T143000 DTSTAMP:20130525T020110 LOCATION:Building 240 / Conference Room 4301, Argonne National Laboratory DESCRIPTION:Quasi-classical trajectory calculations using quantum mechanical energies and forces generated by the Venus and Gaussian programs, provide for the first time a detailed dynamical picture of 1,3-dipolar cycloadditions and carbene cycloadditions to alkenes on the (U)B3LYP/6-31G* surface. For 1,3-dipolar cycloaddtions, the previously reported linear correlation between activation barriers and the energy required to distort reactant to their TS geometries is understandable in terms of the requirements for vibrational excitation. The timing of bond formation and relative reactivities of different 1,3-dipoles are discussed. For carbene cycloadditions, the range of geometries sampled in productive trajectories, and the timing of bond formation were explored.\n\nClick on this link to add this event to your calendar.\n\n SUMMARY:Mechanisms and Dynamics of Cycloadditions UID:1065 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110204T103000 DTEND;TZID=America/Chicago:20110204T113000 DTSTAMP:20130525T020110 LOCATION:TCS Conference Center Rooms 1404 & 1405, Argonne National Laboratory DESCRIPTION:We currently have over 1000 genomes in the public domain, hundreds of metagenomes and many more gene marker data sets, and these numbers are rapidly increasing. Next-generation sequencing technologies promise to further fill the public databases with a bounty of information unthinkable even a few years ago. Each data set represents an organism, or in the case of metagenomes and marker gene sequences a community, with a biological history, sampling location, environmental context and particular set of biologically interesting set of traits. Hence, each of these data sets makes a unique contribution to the ongoing creation of our public online catalogue of life.\n\nWe are now witnessing the rapid democratization of access to sequencing capacity - an immense opportunity for the global community - if proper stewardship of these data keeps pace. This must include enriching the biological context of these sequences within public databases, which will in turn necessitate the adoption of a fresh attitude to reporting results, both in our papers and our submissions to the public databases. The discrete and large genome, metagenome and marker gene data sets (e.g. ribosomal gene surveys) provide ideal opportunities for comparison and contrasting using computational means to solve a wide-range of questions in biology (including questions in medicine, physiology, developmental biology, biogeochemistry, evolution, ecology, etc). \n\nTo exploit fully the promise of these data we need both scientific innovation and community agreement on how to provide appropriate stewardship of these resources for the benefit of all. In this talk, I will focus on describing why community-level approaches can help us ‘tool up’ for this data bonanza. In particular, I will describe the motivations behind the formation of the Genomic Standards Consortium and overview the current activities of this international community in developing and implementing genomic standards and consensus-driven projects. I will also focus on the next frontier in this field – that of determining the function and extent of the “Unknown Genome”, the large fraction of sequenced genetic material found within genomic and metagenomic data sets that has yet to be characterised. SUMMARY:The Era of Genomics: Tooling Up for the Data Bonanza UID:1071 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110208T130000 DTEND;TZID=America/Chicago:20110208T140000 DTSTAMP:20130525T020110 LOCATION:Building 240 Conference Center Rooms 1404 and 1405, Argonne National Laboratory DESCRIPTION:Researchers are currently investigating ways in which various microbes can be leveraged for industrial and medicinal purposes. The diversity of the microbial world holds much promise for the future of research in chemical and biomedical engineering. However biological systems evolve, and it is undesirable for engineered or synthesized microbes to change, particularly in ways we can not control. We need to be able to anticipate how microbes evolve. Our ability to do so is greatly hampered by the diversity and complexity of the microbial world and the stochasticity that underlies natural selection.\n\nIn the work outlined, I take steps towards enabling researchers to understand the evolution of microbes in two areas: metabolic networks and gene regulation. I use a complex systems approach of developing simple models that are placed in the context of evolution. The universality of evolution means that these models apply to a wide range of microbial species and enables researchers to make predictions about the evolution of microbes. SUMMARY:Modeling of gene regulation and metabolism in microbial organisms. UID:1073 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110208T103000 DTEND;TZID=America/Chicago:20110208T113000 DTSTAMP:20130525T020110 LOCATION:Building 240 / Conference Room 4301, Argonne National Laboratory DESCRIPTION:There are many approaches to probing physics beyond the Standard Model. One direct and natural strategy is testing the Standard Model as precisely as possible, and Flavor physics can provide stringent tests on the flavor sector. Many experiments in the flavor sector are related with hadronic matrix elements in the low energy QCD regime where applying perturbative methods is difficult. Thus, lattice gauge theory, which is a successful non-perturbative method, can play an important role in this area. \n\nSeveral interesting topics in Flavor physics and current status for Lattice calculations will be presented.\n\nClick on this link to add this event to your calendar. SUMMARY:Flavor physics from Lattice QCD UID:1077 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110208T133000 DTEND;TZID=America/Chicago:20110208T143000 DTSTAMP:20130525T020110 LOCATION:Building 240 / Conference Room 4301, Argonne National Laboratory DESCRIPTION:Recent advances in computational power make the combination of theory, modeling and simulations a powerful tool for deepening our understanding of life science and material science. Classical molecular dynamics (MD) simulations of large-scale with advanced phase sampling strategy and accurate potential energy functions are crucial for exploring the behavior of biomolecules in physiological environment. In our recent studies, we explore the methodologies and applications of MD simulations on the leadership level platform Blue Gene/P in biological system, force field development, as well as in material science.\n\nIn biomolecular system, the calcium binding protein Calbindin D9k is chosen as a model system for studying the molecular mechanism of cooperativity(1). Binding is characterized in terms of a potential of mean force (PMF) as a function of two variables: the distance r between the ion, and the binding pocket, and the root-mean-square deviation (RMSD) of the conformation of the EF-hand relative to its ion-bound structure. The PMF is calculated using a novel two-dimensional replica-exchange molecular dynamics (MD) umbrella sampling scheme, which is developed and implemented in the program CHARMM to increase the configuration space sampling. Using 2048 replicas on Blue Gene/P, the 2D-PMF/REMD calculation for the binding the second calcium ion converges within 800 ps. The absolute binding free energy of a second ion to a singly occupied calbindin calculated from the 2D-PMF following the statistical mechanical formulation of noncovalent association(2) is -9.4 kcal/mol, in excellent agreement with the experimental value(3). The 2D-PMF/REMD simulations will be extended to provide important information about the molecular basis of calcium binding cooperativity. In the force field development, a novel methodology for osmotic pressure calculation is developed and applied to both nonpolarizable and polarizable Drude model(4). The study shows that the osmotic pressure is a powerful route for validating and parametrizing atomic force fields. In particular, the key thermodynamic and transport properties of ionic Drude models are calculated in large-scale system using dual-Langevin thermostat scheme with the high-performance scalable program NAMD(5). In the material science, the allocation on Intrepid for the IonPermeation project has allowed us to advance significantly our research on Reverse osmosis (RO) membranes. RO membranes, based on aromatic polyamide thin-film composites (TFC), are extensively used for desalination and purification of sea water and brackish water. In our research, we use large-scale MD simulation to probe and reveal the polymerization process and transport properties of the RO membranes at the atomic level. The result shows that computer simulation is an effective tool to investigate the relationship between the microscopic structure of materials and their macroscopic properties, thus can help to advance the new material design.\n\n(1)Marchand S. and Roux B. Protein Struct. Funct. Genet. 1998, 33, 265-284. \n(2)Woo H. and Roux B., PNAS, 2005, 19, 6825. \n(3)Linse S. et al. Biochemistry 1991, 30, 154.\n(4)Luo Y. and Roux B. J. Phys. Chem. Lett., 2010, 1, 183.\n(5)Jiang, W., et al. J. Phys. Chem. Lett. 2011, 2, 87.\n\nClick on this link to add this event to your calendar.\n SUMMARY:MD Simulation Study on Ligand Binding, Ion Permeation and Ionic Force Field on Leadership Level Platform UID:1079 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110207T103000 DTEND;TZID=America/Chicago:20110207T113000 DTSTAMP:20130525T020110 LOCATION:Building 240 / Conference Room 4301, Argonne National Laboratory DESCRIPTION:Inflation has become a well-accepted scheme for explaining the three large mysteries of big-bang cosmology: the observed homogeneity of the universe, its extreme flatness, and the nearly-scale-free power-spectrum of initial density perturbations. The study of models in which the inflation field(s) experience parametric resonance, a popular method for causing particle production and, thus, the resumption of the normal big-bang time line will be discussed. Several distinguishing observables have been investigated using a new C++ pseudo-spectral code named PSpectRE: first, the spectrum of gravitational radiation produced by each model has been calculated. In addition, many models tend to produce oscillons, a kind of long-lived but non-symmetry-protected soliton, and these can profoundly affect the evolution history of the universe prior to their\neventual decay. PSpectRE performs competitively and exhibits excellent energy conservation.\n\nClick on this link to add this event to your calendar. SUMMARY:Simulating Preheating After Inflation UID:1081 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110413T150000 DTEND;TZID=America/Chicago:20110413T160000 DTSTAMP:20130525T020110 LOCATION:Building 240, 4301, Argonne National Laboratory DESCRIPTION:To be announced. SUMMARY:Multigrid methods with applications to MHD and structural mechanics UID:1083 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110209T103000 DTEND;TZID=America/Chicago:20110209T113000 DTSTAMP:20130525T020110 LOCATION:Building 240 Seminar Room 4301, Argonne National Laboratory DESCRIPTION:We present an algorithm for preprocessing a class of\nstochastic shortest-path problems on networks that have no negative cost cycles, almost surely. Our method adds utility to existing frameworks by significantly reducing input problem sizes and thereby increasing computational tractability. Given random costs with finite lower and upper bounds on each edge, our algorithm removes edges that cannot be in any optimal solution to the deterministic shortest-path problem, for any realization of the random costs. Although this problem is NP-complete, our algorithm efficiently preprocesses nearly all edges in a given network. We provide computational results both on sparse networks from PSPLIB - a well-known project evaluation and review technique library [Kolisch, R., A. Sprecher. 1996. PSPLIB—A project scheduling problem library. Eur. J. Oper. Res. 96(1) 205–216] - and dense synthetic ones: on average, less than 0.1% of the edges in the PSPLIB instances and 0.5% of the edges in the dense instances remain unclassified after preproces sing. SUMMARY:Preprocessing Stochastic Shortest-Path Problems UID:1085 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110318T140000 DTEND;TZID=America/Chicago:20110318T150000 DTSTAMP:20130525T020110 LOCATION:Building 240, Seminar Room 4301, Argonne National Laboratory DESCRIPTION:Abstract:\nThe data explosion in computational sciences and engineering requires visualization algorithms to be at their core high-performance data processing methods. The corresponding need for high compute power combined with the vital role played by user interaction in an effective visualization experience puts a natural emphasis on techniques that can harness the ever growing power of the GPU. While visualization research has long been focussing its efforts on GPU-friendly solutions, the mapping of a number of visualization techniques that have gained popularity in recent years continues to pose a significant challenge.\n\nThis talk will discuss recent and ongoing research done in my group on innovative data structures and parallel algorithms that enable advanced scalar, vector, and tensor visualization techniques on modern GPUs. The topics will include a new flexible data representation and its application to volume rendering and ray tracing of dynamic scenes, a new hierarchical data structure that allows for the interactive exploration of transient flow structures at arbitrary resolution, and a hybrid CPU-GPU method for the extraction of ridge meshes of high quality at unprecedented rates with applications in engineering and medical imaging. SUMMARY:GPU Friendly Data Structures and Algorithms for Advanced Visualization UID:1147 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110330T150000 DTEND;TZID=America/Chicago:20110330T160000 DTSTAMP:20130525T020110 LOCATION:Building 240, Conference Center 1404-1405, Argonne National Laboratory DESCRIPTION:We consider the notion of complex analyticity. We will first define it and then point out some differences from real analyticity. We give examples of some practical applications of complex analysis in one variable. Next, we briefly outline how complex analysis in several variables is distinct from that in one variable. We will finish with a discussion on some metrics over pseudoconvex domains in n-complex dimensions. SUMMARY:A simple introduction to complex analysis UID:1093 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110215T143000 DTEND;TZID=America/Chicago:20110215T163000 DTSTAMP:20130525T020110 LOCATION:Searle 240A, University of Chicago, 5735 South Ellis Avenue, Chicago, IL DESCRIPTION:Join the Computation Institute and IT Services at a Jeopardy! Viewing party where the Watson supercomputer (developed by IBM) will make its debut as the first non-human contestant to appear on the award-winning quiz show. Watson will take on all-time Jeopardy! Champions, Ken Jennings and Brad Rutter, in a bid to become a new breed of champion.\n\nThe event, moderated by Ian Foster, Computation Institute Director, will also include perspectives from Klara Jelinkova, Chief Information Technology Officer within IT Services and University faculty members, John Goldsmith (Computer Science and Linguistics) and James Evans (Sociology), on the impact of natural language processing.\n\nRepresentatives from IBM will also be on hand to explain the technology behind Watson and what they hope to accomplish with it.\n\nRefreshments will be served. SUMMARY:Watson on Jeopardy! Cast your votes at the CI’s viewing party UID:1095 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110307T103000 DTEND;TZID=America/Chicago:20110307T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, Conference Center 1404-1405, Argonne National Laboratory DESCRIPTION:Will blocking become as important in tensor computations as it is in matrix computations? \nI will address this question from several perspectives. Topics include (a) unfolding block tensors into block matrices, (b) a block-level connection between general tensor singular values and symmetric tensor eigenvalues, and (c) low-rank tensor approximation using the Kronecker product SVD. SUMMARY:Block Tensor Computations UID:1099 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20100427T150000 DTEND;TZID=America/Chicago:20100427T160000 DTSTAMP:20130525T020110 LOCATION:Building 240, Conference Center 1404-1405, Argonne National Laboratory DESCRIPTION:To be announced. SUMMARY:Title to be announced UID:1101 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110427T150000 DTEND;TZID=America/Chicago:20110427T160000 DTSTAMP:20130525T020110 LOCATION:Building 240, 4301, Argonne National Laboratory DESCRIPTION:Global Arrays (GA) is popular high-level parallel programming model that provides the substrate on which the NWChem computational chemistry suite has been built. GA\'s global-view model is supported by the ARMCI partitioned global address space (PGAS) runtime system, which is implemented natively on each supported platform in order to provide the best performance. Because of its sophistication, significant time and effort are required to implement ARMCI on a new system. The industry standard Message Passing Interface (MPI) also provides one-sided functionality and is available on virtually every supercomputing system. However, it is believed that the MPI one-sided model is not rich enough to support a high-level model like GA/ARMCI. In this presentation, I will describe the first high-performance, portable implementation of ARMCI using MPI one-sided communication. Existing GA infrastructure is interfaced with ARMCI-MPI and NWChem is used to evaluate application-level performance for MPI versus native implementations of the GA runtime system. SUMMARY:High-Level, One-Sided Programming Models on MPI: A Case Study with Global Arrays UID:1103 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110216T103000 DTEND;TZID=America/Chicago:20110216T113000 DTSTAMP:20130525T020110 LOCATION:Building 240 Seminar Room 4301, Argonne National Laboratory DESCRIPTION:Tone languages like Mandarin Chinese use tones (specific pitch patterns) to distinguish syllables which are otherwise ambiguous. Tones in Mandarin Chinese have been shown to carry as much information as vowels [1]. However, several contextual factors in continuous speech make it challenging to achieve successful tone recognition. First, coarticulation between adjacent tones can compromise the realization of underlying tone targets. Second, phrase, sentence and topic boundaries can also affect pitch; pitch variation has been successfully employed to perform sentence and story segmentation. Third, speaker differences, especially gender differences, make it necessary to scale tone targets to compensate for individual variation.\n\nI present two approaches to model contextual effects on tone production in continuous speech. These approaches achieve state-of-the-art tone recognition performance on Mandarin Chinese Broadcast News data. The first approach focuses on modeling the coarticulation between adjacent tones. It manipulates a landmark-based vowel detection system to locate the most reliable tone production regions and to remove those regions affected by coarticulation. This approach shows a 15% improvement over two previously published tone recognition frameworks. The second approach employs sequential graphical models with Conditional Random Fields (CRF) to encode tone variation due to phrase, sentence and topic level intonation. We found that not only do different tones vary under each of these intonation conditions, but the choice of graphical model structure can also impact performance. These techniques can also be extended to other machine learning challenges, i.e modeling human social processes in dyad conversations of three cultures (American English, Mexican Spanish and Iraqi Arabic). Finally, I will briefly talk about how to apply structural learning on network behavior and attack pattern\nanalysis [2,3].\n\n[1] Dinoj Surendran, Gina-Anne Levow, and Yi Xu. Tone recognition in mandarin using focus. Proceedings of Interspeech/ICSLP 2005, 2005.\n[2] Detecting Anomalies in Network Traffic Using Maximum Entropy Estimation. Yu Gu, Andrew McCallum and Don Towsley. Internet Measurement Conference, 2005.\n[3] Layered approach using conditional random fields for intrusion detection. Kapil Kumar Gupta, Baikunth Nath and Ramamohanarao Kotagiri IEEE Trans. Dependable and Secure Computing, Vol. 7, No. 1, Jan-Mar\n2010 SUMMARY:Contextual Modeling in Continuous Speech Tone Recognition UID:1105 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110218T103000 DTEND;TZID=America/Chicago:20110218T113000 DTSTAMP:20130525T020110 LOCATION:Building 240 Conference Center Rooms 1404 and 1405, Argonne National Laboratory DESCRIPTION:Tone languages like Mandarin Chinese use tones (specific pitch patterns) to distinguish syllables which are otherwise ambiguous. Tones in Mandarin Chinese have been shown to carry as much information as vowels [1]. However, several contextual factors in continuous speech make it challenging to achieve successful tone recognition. First, coarticulation between adjacent tones can compromise the realization of underlying tone targets. Second, phrase, sentence and topic boundaries can also affect pitch; pitch variation has been successfully employed to perform sentence and story segmentation. Third, speaker differences, especially gender differences, make it necessary to scale tone targets to compensate for individual variation.\n\nI present two approaches to model contextual effects on tone production in continuous speech. These approaches achieve state-of-the-art tone recognition performance on Mandarin Chinese Broadcast News data. The first approach focuses on modeling the coarticulation between adjacent tones. It manipulates a landmark-based vowel detection system to locate the most reliable tone production regions and to remove those regions affected by coarticulation. This approach shows a 15% improvement over two previously published tone recognition frameworks. The second approach employs sequential graphical models with Conditional Random Fields (CRF) to encode tone variation due to phrase, sentence and topic level intonation. We found that not only do different tones vary under each of these intonation conditions, but the choice of graphical model structure can also impact performance. These techniques can also be extended to other machine learning challenges, i.e modeling human social processes in dyad conversations of three cultures (American English, Mexican Spanish and Iraqi Arabic). Finally, I will briefly talk about how to apply structural learning on network behavior and attack pattern\nanalysis [2,3].\n\n[1] Dinoj Surendran, Gina-Anne Levow, and Yi Xu. Tone recognition in mandarin using focus. Proceedings of Interspeech/ICSLP 2005, 2005.\n[2] Detecting Anomalies in Network Traffic Using Maximum Entropy Estimation. Yu Gu, Andrew McCallum and Don Towsley. Internet Measurement Conference, 2005.\n[3] Layered approach using conditional random fields for intrusion detection. Kapil Kumar Gupta, Baikunth Nath and Ramamohanarao Kotagiri IEEE Trans. Dependable and Secure Computing, Vol. 7, No. 1, Jan-Mar 2010 SUMMARY:Contextual Modeling in Continuous Speech Tone Recognition UID:1107 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110304T120000 DTEND;TZID=America/Chicago:20110304T130000 DTSTAMP:20130525T020110 LOCATION:Building 240, TCS Room 1406, Argonne DESCRIPTION:Globus Online is high-performance file transfer Software-as-a-Service (SaaS). Launched by a CI-lead team in November 2010, Globus Online allows researchers to securely and reliably move large volumes of data between HPC resources and local computing environments. This event will provide an overview of the service\'s capabilities and enable attendees to start using the service for their file transfer needs. SUMMARY:Globus Online: Reliable File Transfer Software-as-a-Service UID:1115 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110225T133000 DTEND;TZID=America/Chicago:20110225T143000 DTSTAMP:20130525T020110 LOCATION:Building 240 / Conference Room 4301, Argonne National Laboratory DESCRIPTION:One of the most challenging problems in modern electronic structure theory is the development of affordable ab initio methods that can provide an accurate description of ground- and excited-state molecular potential energy surfaces. Of all approaches to the many-electron correlation problem, the coupled-cluster (CC)(1) methods are regarded as the best methods for high accuracy calculations. Standard single-reference (SR) CC and equation-of-motion CC (EOMCC)(2,3) approaches, such as CCSD(4), CCSD(T)(5), and EOMCCSD(3), provide an excellent description of closed-shell systems and dynamic correlation effects but fail when potential energy surfaces involving bond breaking and biradicals are examined. It has been demonstrated that many of the problems encountered in SRCC or EOMCC calculations can be eliminated by adding non-iterative corrections to CC or EOMCC energies based on the method of moments of CC equations (MMCC)(6). On the other hand, it is well known that the low-order multi-reference many-body perturbation theory (MRMBPT) approaches can provide very good description of electronic quasi-degeneracies with relatively small computer effort. MRMBPT methods also provide straightforward access to ground and excited states. The purpose of this work is to develop a new class of MMCC approaches in which information about the most essential non-dynamic and dynamic correlation effects that are relevant to electronic quasi-degeneracies is extracted from MRMBPT, referred to as MMCC/PT(7,8).\n\nThe performance of the basic MMCC/PT approximation, in which inexpensive non-iterative corrections due to triples (MMCC(2,3)/PT) and triples and quadruples (MMCC(2,4)/PT) are added to the ground- and excited state energies obtained with CCSD/EOMCCSD, is illustrated by the results of a few benchmark calculations including bond breaking in HF, F(2), and H(2)0, and excited states of CH+. The test calculations show that at least for single bond breaking and some cases of multiple bond dissociation, the MMCC/PT method eliminates the failures of the conventional CC/EOMCC approaches at large internuclear distances. It also eliminates large errors in EOMCCSD results for excited states dominated by two-electron transition without invoking the expensive steps of high-order iterative EOMCC methods.\n \nIn addition to describing the theoretical details behind MMCC/PT and discussing the results from the test calculations, the computational details, specifically the diagram factorization techniques which lead to efficient computer implementation, will be presented in this talk as well.\n\nReferences:\n(1). J. Čížek, J. Chem. Phys. 45, 4256 (1966). \n(2). K. Emrich, Nucl. Phys. A 351, 379 (1981). \n(3). J. F. Stanton and R. J. Bartlett, J. Chem. Phys. 98, 7029 (1993). \n(4). G. D. Purvis III and R. J. Bartlett, J. Chem. Phys. 76, 1910 (1982). \n(5). K. Raghavachari, G. W. Trucks, J. A. Pople, and M. Head-Gordon, Chem. Phys. Lett. 157, 479 (1989). \n(6). K. Kowalski and P. Piecuch, J. Chem. Phys. 113, 18 (2000); K. Kowalski and P. Piecuch, J. Chem. Phys. 115 2966 (2001); P. Piecuch, K. Kowalski, I.S.O. Pimienta, P-D. Fan, M. D. Lodriguito, M. J. McGuire, S. A. Kucharski, T. Kuś, and M. Musiał, Theor. Chem. Acc. 122, 349 (2004). \n(7). M. D. Lodriguito, K. Kowalski, M. Włoch, and P. Piecuch, J. Mol. Struct: THEOCHEM 771, 89 (2006). \n(8). M. Włoch, M.D. Lodriguito, and P. Piecuch, J.R. Gour, Mol. Phys. 104, 2149 (2006). \n\nClick on this link to add this event to your calendar. SUMMARY:Non-Iterative Coupled-Cluster Methods Employing Multi-Reference Perturbation Theory Wave Functions UID:1117 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110301T103000 DTEND;TZID=America/Chicago:20110301T113000 DTSTAMP:20130525T020110 LOCATION:Building 240 Seminar Room 4301, Argonne National Laboratory DESCRIPTION:Researchers are currently investigating ways in which various microbes can be leveraged for industrial and medicinal purposes. The diversity of the microbial world holds much promise for the future of research in chemical and biomedical engineering. However biological systems evolve, and it is undesirable for engineered or synthesized microbes to change, particularly in ways we can not control. We need to be able to anticipate how microbes evolve. Our ability to do so is greatly hampered by the diversity and complexity of the microbial world and the stochasticity that underlies natural selection.\n\nIn the work outlined, I take steps towards enabling researchers to understand the evolution of microbes in two areas: metabolic networks and gene regulation. I use a complex systems approach of developing simple models that are placed in the context of evolution. The universality of evolution means that these models apply to a wide range of microbial species and enables researchers to make predictions about the evolution of microbes. SUMMARY:Modeling of gene regulation and metabolism in microbial organisms. UID:1127 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110224T133000 DTEND;TZID=America/Chicago:20110224T143000 DTSTAMP:20130525T020110 LOCATION:Bldg 240, Rm 4301, Argonne National Laboratory DESCRIPTION:Quantum Chromodynamics (QCD) is the fundamental theory that describes the interactions of quarks and gluons. In the lattice formulation of QCD, the equations governing these interactions are solved numerically on a four-dimensional space-time grid, a process that it is carried out in two basic stages. The first stage, \"lattice generation,\" typically demands leadership-class machines, while the subsequent \"analysis\" stage is often farmed out to smaller resources, each sustaining perhaps a Tflop/s. Recently, graphics processing units (GPUs) have proven to be extremely cost-effective for many workloads in lattice QCD, especially the latter \"analysis\" stage of large-scale computations. As a result, it is now possible to tackle much more challenging problems than would have been possible using more traditional architectures.\n\nThis talk will focus on lattice QCD as a case study in GPU computing, drawing lessons that might have relevance for other fields. In particular, I will discuss \"QUDA,\" a library of linear solvers tailored for QCD and developed using NVIDIA\'s \"C for CUDA\" API. Recently, QUDA has been enhanced to support the use of multiple GPUs in parallel with communications handled via MPI. I will discuss some of the strategies we employed to obtain high performance on a single GPU and maintain efficiency on GPU-enabled clusters. Finally, I will describe some of the challenges we face as week seek to scale up to systems consisting of hundreds or even thousands of GPUs, with possible consequences for the design of future heterogeneous systems on the road to the exascale. SUMMARY:Leveraging Many GPUs in Parallel for Lattice QCD UID:1129 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110309T103000 DTEND;TZID=America/Chicago:20110309T113000 DTSTAMP:20130525T020110 LOCATION:Building 240 Seminar Room 4301, Argonne National Laboratory DESCRIPTION:The talk is broadly divided into two parts. First part discusses the work of the speaker in the development of structured and unstructured preconditioners and their analysis (classical and Fourier). \n\nIn the second part, the speaker reviews the nonlinear algorithms with his own insights, in particular, several ingredients like full approximation scheme, nonlinear acceleration techniques their advantages and disadvantages are outlined. An understanding of the advantages and disadvantages of Homotopy solvers over Newton solvers is discussed. A brief discussion of Groebner basis and its effectiveness in solving the non linear equations is also \ndiscussed. SUMMARY:Exploring Non Linear Algorithms For The Solution Of Partial Differential Equations UID:1131 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110314T103000 DTEND;TZID=America/Chicago:20110314T113000 DTSTAMP:20130525T020110 LOCATION:Building 240 Conference Room 1404-1405, Argonne National Laboratory DESCRIPTION:Inelastic electron tunneling via molecular-scale junctions can induce a variety of fascinating dynamical processes in the molecular moiety. These include vibration, rotation, inter-mode energy flow and reaction. Potential applications of current-driven dynamics in heterojunctions range from new forms of molecular machines and new modes of conduction, to new directions in surface nanochemistry and nanolithography. \n\nIn the first part of the talk, I will discuss the qualitative physics underlying current-driven dynamics in molecular-scale devices, briefly outline the theory we developed to explore these dynamics, describe the results of ongoing research on surface nanochemistry and molecular machines, and sketch several of our dreams and plans in these areas. \n\nThe application of light to control molecular motions and electronic transport in junctions is intriguing, since photonic (by contrast to electronic) sources offer (sub)femtosecond time resolution and tunable phase and polarization properties. One of several challenges, however, is the requirement of coherent light sources that are tightly localized in space. It is here that plasmonics offer an opportunity. \n\nIn the second part of the talk, we will combine plasmonics physics with concepts and tools borrowed from coherent control of molecular dynamics with two goals in mind. One is to introduce new function into nanoplasmonics, including ultrafast elements and broken symmetry elements. The second is to develop coherent nanoscale sources and apply them to coherent control of both molecular dynamics and electric transport in the nanoscale. \n\nSeveral simple elements in what we envision developing into coherently controlled nanoplasmonics are schematically illustrated in Fig. 1. The T-junction of Fig. 1A guides electromagnetic energy traveling down the leg into one or the other of the two symmetry-equivalent arms of the junction. Figure 1B depicts a hybrid construct, which combines elements that provide local enhancement with elements that provide long distance propagation in order to minimize losses. The structural parameters of the construct are optimized using a genetic algorithm. Fig. 1C depicts a plasmonic nanocrystal, developed to separate an incident plane wave into two frequency components and funnel each component in a different direction normal to the direction of incidence. To conclude the talk, we will return to nanoelectronics, and illustrate the application of plasmonics to control of transport in the nanoscale, with a view to ultrafast electric switches. \n\n SUMMARY:Toward Coherent Control in the Nanoscale UID:1133 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110302T103000 DTEND;TZID=America/Chicago:20110302T113000 DTSTAMP:20130525T020110 LOCATION:Building 240 Seminar Room 4301, Argonne National Laboratory DESCRIPTION:My presentation will consist of two parts. During the first part, I will present results of work done on multi-scale filamentary analysis of the 3D images. This research focuses on a very specific type of data which is typically very noisy and may contain filamentary faint structures such as line segments and smooth curves. The proposed tools mainly rely on the 3-D Beamlet Transform offering the collection of line integrals along a strategic multi-scale set of line segments. Such tools and methods can be applied in a wide variety of applications which involve 3D imaging. In this work, we focus on applying Beamlet methods for the problem of dim target multi-frame detection and develop specialized tools for this application. We use tools from graph theory and apply them to the special graph generated by the beamlets set.\n\nDuring the second part of this discussion, I will present an overview of ongoing research in 2D-3D face recognition by the Computational Biomedicine Lab. Specifically, I will be focusing on the methods for facial landmarks detection. I will present a fully automated framework for the facial component-landmarks detection based on multi-resolution isotropic analysis and adaptive bag-of-words descriptors incorporated into a cascade of boosted classifiers. The advantage of our approach is that it has robustness to pose as well as illumination. Our method has a failure rate lower than that of commercial software. Additionally, we demonstrate that using our method for the initialization of a point landmark detector results in performance comparable with that of state-of-the-art methods. SUMMARY:Facial Component Landmark Detection UID:1135 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110310T103000 DTEND;TZID=America/Chicago:20110310T113000 DTSTAMP:20130525T020110 LOCATION:Building 240 Seminar Room 4301, Argonne National Laboratory DESCRIPTION:My presentation will consist of two parts. During the first part, I will present results of work done on multi-scale filamentary analysis of the 3D images. This research focuses on a very specific type of data which is typically very noisy and may contain filamentary faint structures such as line segments and smooth curves. The proposed tools mainly rely on the 3-D Beamlet Transform offering the collection of line integrals along a strategic multi-scale set of line segments. Such tools and methods can be applied in a wide variety of applications which involve 3D imaging. In this work, we focus on applying Beamlet methods for the problem of dim target multi-frame detection and develop specialized tools for this application. We use tools from graph theory and apply them to the special graph generated by the beamlets set.\n\nDuring the second part of this discussion, I will present an overview of ongoing research in 2D-3D face recognition by the Computational Biomedicine Lab. Specifically, I will be focusing on the methods for facial landmarks detection. I will present a fully automated framework for the facial component-landmarks detection based on multi-resolution isotropic analysis and adaptive bag-of-words descriptors incorporated into a cascade of boosted classifiers. The advantage of our approach is that it has robustness to pose as well as illumination. Our method has a failure rate lower than that of commercial software. Additionally, we demonstrate that using our method for the initialization of a point landmark detector results in performance comparable with that of state-of-the-art methods. SUMMARY:Facial Component Landmark Detection UID:1137 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110311T103000 DTEND;TZID=America/Chicago:20110311T113000 DTSTAMP:20130525T020110 LOCATION:Building 240 / Room 4301, Argonne National Laboratory DESCRIPTION:The scarcity of near-source recordings for large earthquakes motivates the use of numerical simulations for the prediction of possible ground motion from future events. The relevant phenomena (frictional breakdown, shear heating, effective normal-stress fluctuations, material damage, etc.) controlling rupture are strongly interacting and span many orders of magnitude in spatial scale, requiring high-resolution simulations that couple disparate physical processes (e.g., elastodynamics, thermal weakening, pore-fluid transport, and heat conduction). The capacity to perform 3D rupture simulations that couple these processes will provide guidance for constructing appropriate source models for high-frequency ground motion simulations. Compounding the computational challenge, for the case physics-based probabilistic seismic hazard analysis (PSHA), is the need to perform simulations for multiple variations of all statistically significant possible events. A goal of the Southern California Earthquake Center (SCEC) during the ALCF Early Science Program is to calculate a 1Hz PSHA hazard map for California using improved rupture models from our multi-scale dynamic rupture simulations. This calculation will provide numerous important seismic hazard results, including a state-wide extended earthquake rupture forecast with rupture variations for all significant events, a synthetic seismogram catalog for thousands of scenario events and more than 5000 physics-based seismic hazard curves for California. I will highlight recent milestones toward reaching this goal, as well as discuss the current state of SCEC simulations codes as we begin preparation for upcoming ALCF systems. SUMMARY:Petascale Earthquake Simulations UID:1139 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110318T103000 DTEND;TZID=America/Chicago:20110318T113000 DTSTAMP:20130525T020110 LOCATION:Building 240 Seminar Room 4301, Argonne National Laboratory DESCRIPTION:Embedded systems that are collections of several electronic modules are relatively widespread. In this heterogeneous ecology of multi-module systems, a class of embedded systems distinguish themselves by the scale and distribution of their processing power. Our background research suggests that this class of systems is largely underexplored with potential applications including, for example, HCI, environmental sensing, robotics and fractionated systems.\nWe call them Embedded Aggregates. In this talk I will motivate the need for the aggregates, outline the requirements of the aggregate class of systems and argue that there is a gap in the embedded communication protocol spectrum that makes developing aggregates a formidable challenge. Next, I will describe our efforts in building the underlying communication fabric and hardware infrastructure from ground up by defining new protocols, mechanisms and specifications. Finally, I will discuss Blades and Tiles, an embedded aggregate for HCI researchers for use in mid to late stages of physical user interface (UI) prototyping and pilot UI deployments. SUMMARY:Embedded Aggregates UID:1145 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110323T103000 DTEND;TZID=America/Chicago:20110323T113000 DTSTAMP:20130525T020110 LOCATION:Building 240 Seminar Room 4301, Argonne National Laboratory DESCRIPTION:First, we will describe a new approach to the $p$ and $hp$ versions of finite element method. Our approach is characterized by the use of nodal basis functions, and a posteriori error estimates based on derivative recovery. The special basis functions defined for transition elements allow us to accommodate elements of different degrees, avoid hanging nodes and maintain the continuity of the finite element space. Our error estimates are very attractive as they are robust and independent of the PDE.\n\nSecond, we will discuss the use of an approximate Newton\'s method in conjunction with GMRES to solve large nonlinear systems in IWFM, a water resources management and planning (finite element) model developed by California State Department of Water Resources. Strategies to adaptively control GMRES accuracy and formulate damping factors are proposed.\n\nNumerical results will be provided in both parts of the talk.\n SUMMARY: $hp$-Adaptive Finite Elements and Approximate Newton-Krylov Solver UID:1153 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110517T103000 DTEND;TZID=America/Chicago:20110517T113000 DTSTAMP:20130525T020110 LOCATION:Building 240 Seminar Room 4301, Argonne National Laboratory DESCRIPTION:The numerical treatment of interesting models for physical systems is often hobbled by two major factors: The implementation cost of exploring the space of potentially relevant models, and the computational cost of solving these models. My research has attempted to combine automated scientific computing methods with fast solution algorithms in order to allow for the efficient solution of novel problem formulations in a few key domains of interest. I will describe three efforts: geometric multigrid methods for general finite element problems, the treatment of models for implicit solvation using automated scientific computing methods, and the development of a fast approximation technique for an important model quantity arising in the classical density functional theory treatment of ion channels. SUMMARY:Fast Solvers and Electrochemical Problems UID:1233 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110331T103000 DTEND;TZID=America/Chicago:20110331T113000 DTSTAMP:20130525T020110 LOCATION:Building 240 Conference Center Room 1404, Argonne National Laboratory DESCRIPTION:The Toolkit for Accurate Scientific Software (TASS, http://vsl.cis.udel.edu/tass) is a new symbolic execution framework for verifying C/MPI programs. TASS can verify that a program is free of deadlocks, buffer overflows, memory leaks, null pointer dereferences, divisions by 0, and assertion violations, within a bounded region of the program\'s input space. In addition, it can compare two programs to determine whether they are functionally equivalent. This allows the developer to use a simple sequential program as a specification and use TASS to show that a more complex, parallel program is a correct implementation of that specification. TASS is being used to explore other specification mechanisms, such as \"collective assertions\" that span multiple processes. Unlike previous tools, such as MPI-Spin, TASS works directly from C source code, augmented with a minimal amount of user annotations. A number of novel model checking optimizations have also been incorporated, allowing TASS to scale much further than previous approaches. SUMMARY:Formal Specification and Verification of MPI Programs Using TASS UID:1159 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110328T103000 DTEND;TZID=America/Chicago:20110328T113000 DTSTAMP:20130525T020110 LOCATION:Bldg. 240, Room 4301, Argonne National Laboratory DESCRIPTION:Abstract:\n\nWith the increasing scale and complexity of HPC systems, reliability is becoming critical for these systems. System logs are the primary source of information to understand and analyze system problems. Nevertheless, little study has been done on automated log analysis for HPC systems. In this talk, I will summarize our study on system logs collected from production HPC systems by exploiting data mining and statistical learning technologies.\n\nOur work can be broadly divided into four parts: log pre-processing, online failure prediction, automatic root cause diagnosis, and reliability modeling. The work can greatly improve our understanding of faults/errors/failures arising from hardware/software components and their interactions in HPC systems, and can further facilitate the resilience research for large-scale systems. SUMMARY:Log Analysis for Reliability Management in Large-Scale Systems UID:1165 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110504T150000 DTEND;TZID=America/Chicago:20110504T160000 DTSTAMP:20130525T020110 LOCATION:Bldg 240 Conference Center 1404-1405, Argonne National Laboratory DESCRIPTION:With the ever increasing computing power of supercomputers, nowadays computational sciences and engineering demand numerical solutions for problems in a larger and larger scale. One important problem that finds a wide range of applications in such as physical simulations and machine learning, is in a stochastic process the generation of random data from a prescribed covariance rule, and the inverse question of fitting the covariance rule given experimented data. This problem gives rise to a number of numerical linear algebra challenges, where one needs to deal with dense and irregularly structured covariance matrices of mega-, giga- or even much larger sizes. In this talk, I will illustrate specific encountered challenges, including computing the square root of the matrix, estimating the diagonal, solving linear systems, and preconditioning the matrix. For some of these challenges, we have developed efficient and scalable methods that are capable of dealing with matrices of size at least in the mega-scale, on a single desktop machine. As a natural extension, high performance codes run on supercomputers are being developed; however, there remain other unsolved challenging tasks along the line, which call for innovative algorithms as well as theory. SUMMARY:Numerical Linear Algebra Challenges in Very Large Scale Data Analysis UID:1169 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110330T130000 DTEND;TZID=America/Chicago:20110330T140000 DTSTAMP:20130525T020110 LOCATION:Building 240 / Conference Room 4301, Argonne National Laboratory DESCRIPTION:The landscape of computational chemistry software is dominated by legacy codes, some decades old and designed for hardware architectures long since retired. In this talk we will analyze the performance of state-of-the-art quantum chemistry codes for massively parallel computers - NWChem and MPQC - on current supercomputers. The analysis will focus on both the communication patterns of these codes and their utilization of multicore processors. Algorithms and software components for next-generation quantum chemistry codes will be discussed. In particular, we have implemented a new one-sided communication system targeting forthcoming multi-petaflop architectures such as Blue Gene/Q and Blue Waters. To address the other axis of increasing parallelism - multicore and heterogenous processors - we have implemented a mixed GPU and CPU coupled-cluster code, the performance of which is significantly better than the best CPU codes. SUMMARY:Software Architecture of Current and Future High-Performance Computational Chemistry Codes UID:1171 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110607T083000 DTEND;TZID=America/Chicago:20110607T170000 DTSTAMP:20130525T020110 LOCATION:Building 240 Conference Room 1416, Argonne National Laboratory DESCRIPTION:This workshop is especially geared for current INCITE and Discretionary project teams that have already scaled to at least two racks on our Blue Gene/P, Intrepid.\n\nThe bulk of this specialized workshop will be devoted to tuning applications. In addition:\n\n• Our skilled performance engineers and computational scientists will be at your side to provide hands-on support.\n• We’ll have special queues to accommodate full-scale runs.\n• Tool and debugger developers will be on hand to assist you.\n• Presentations on system architecture, tools, and debuggers will be made as needed.\n SUMMARY:Leap to Petascale 2011 - Formal registration is required UID:1175 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110412T103000 DTEND;TZID=America/Chicago:20110412T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, 4301, Argonne National Laboratory DESCRIPTION:The ability to optimize complex systems using high-fidelity simulations can yield systems with unprecedented performance. Complex systems range from aircraft to integrated circuits to power systems, all of which may now be accurately simulated on a supercomputer. However, the cost of these simulations is too high for optimization to be tractable. Furthermore, the sensitivity of a system\'s performance to its design variables is often not available, and the system performance estimate may vary between simulations due to the presence of uncertainties. This presentation will suggest a multifidelity framework to reduce the cost of designing systems with the most accurate simulations available. This framework progressively calibrates less expensive, lower-fidelity simulations to accelerate finding designs that are optimal with respect to the high-fidelity simulations. Novel aspects of this research are that the framework can be proven to converge to a high-fidelity optimal design even in the absence of high-fidelity sensitivity information, and that the system optimization can be parallelized.\n SUMMARY:Multifidelity Methods for Multidisciplinary System Design UID:1177 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110511T150000 DTEND;TZID=America/Chicago:20110511T160000 DTSTAMP:20130525T020110 LOCATION:Building 240, Conference Center 1404-1405, Argonne National Laboratory DESCRIPTION:Vortex sheets are used in fluid dynamics to model thin shear layers in slightly viscous flow. Examples include mixing layers subject to Kelvin-Helmholtz instability and airplane trailing wakes. One of the earliest simulations in computational fluid dynamics used the point vortex method to compute vortex sheet motion. The results seemed to confirm Prandtl\'s idea that vortex sheets roll up smoothly into concentrated spirals, but later simulations with higher resolution encountered difficulty due to ill-posedness and singularity formation. I\'ll describe the fundamental contributions on this topic by Louis Rosenhead, Garrett Birkhoff, and Derek Moore, and then discuss more recent regularized simulations past the critical time. The results support a conjecture by Dale Pullin on self-similarity, but chaos intervenes unexpectedly. Then I\'ll describe a new panel method for vortex sheet motion in 3D flow which uses a Cartesian treecode for efficiency. An application to vortex rings will be presented. Finally I\'ll indicate briefly how the treecode is being used in other problems involving long-range particle interactions. SUMMARY:Computing Vortex Sheet Motion UID:1179 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110414T150000 DTEND;TZID=America/Chicago:20110414T160000 DTSTAMP:20130525T020110 LOCATION:Bldg 240, 4301, Argonne National Laboratory DESCRIPTION:Ice sheets are a key component of the earth system, and consequently need to be taken into consideration in numerical predictions of climate change, both because of their involvement in internal climate feedbacks, and because of the societal risks posed by sea level change. This talk will explore some of the computational challenges posed by ice sheet modelling over different temporal and spatial scales, and outline work currently being undertaken to address these challenges, both at Swansea and elsewhere. SUMMARY:Numerical modelling in Glaciology: computational challenges and opportunities UID:1185 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110418T110000 DTEND;TZID=America/Chicago:20110418T120000 DTSTAMP:20130525T020110 LOCATION:Building 240 Conference Room 1404, Argonne National Laboratory DESCRIPTION:Abstract. A seismic hazard model computes the probability that earthquake ground motions at a geographic site will exceed some specified shaking intensity during a fixed interval of time. The National Seismic Hazard Model (NSHM), which is regularly updated by the USGS for all U.S. territories, is the basic source of seismological information used by decision-makers at the local, state, and federal levels for earthquake risk assessment, seismic safety engineering, and disaster preparedness. The NSHM is currently limited by uncertainties in long-term earthquake rupture forecasts, the paucity of near-field recordings of large earthquakes, and the variability arising from fault rupture complexity and wave propagation through highly heterogeneous crustal structures. This presentation will describe how large-scale, physics-based simulations of earthquakes can improve seismic hazard mapping by addressing these limitations.\nInterdisciplinary teams organized by the Southern California Earthquake Center (SCEC) have used the Oak Ridge and Argonne Leadership Computing Facilities, as well as NSF supercomputers, to model a complete spectrum of earthquakes in Southern California. The M8 production runs on NCCS Jaguar have simulated magnitude-8 earthquakes on the San Andreas fault with 4D outer/inner scale ratios approaching 1017, fully representing the strong seismic shaking from a “wall-to-wall” rupture up to frequencies of 2 Hz. The computational advantages of seismic reciprocity, which expresses the reversal symmetries of linear wave propagation, have been employed to develop a new type of physics-based seismic hazard model called CyberShake. In its current implementation, CyberShake can generate and manipulate suites of synthetic time-histories large enough (~108 seismograms) for the probabilistic mapping of shaking intensities throughout the Los Angeles region. The physics-based model predicts long-period shaking intensities in the highly-populated sedimentary basins that are substantially higher than empirical models such as the NSHM, primarily due to the strong coupling between rupture directivity and basin excitation. SCEC plans to scale CyberShake up to a 1-Hz statewide seismic hazard model—an exascale computational problem—to coincide with the next release of the NSHM in 2013. \nIn addition to their utility for long-term hazard mapping, numerical simulations can be used to improve operational earthquake forecasting, which provides short-term earthquake probabilities using seismic triggering models, and earthquake early warning systems, which predict imminent shaking during an event. These applications offer new and urgent computational challenges, including requirements for robust, on-demand supercomputing and rapid access to very large data sets.\n_________\n SUMMARY:From M8 to CyberShake: Using Large-Scale Numerical Simulations to Improve Seismic Hazard Models UID:1189 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:00000000T100000 DTEND;TZID=America/Chicago:00000000T110000 DTSTAMP:20130525T020110 LOCATION:Building 240 - Confernence Room 4301, Argonne National Laboratory DESCRIPTION:Traditional numerical techniques for solving time-dependent partial-differential-equation (PDE) initial-value problems (IVPs) store a truncated representation of the function values and some number of their time derivatives at each time step. What if spatial derivatives were also stored? Although redundant in the dx --> 0 limit, this paper demonstrates that stored spatial derivatives can be propagated in an efficient and self-consistent manner. They can effectively contribute to the evolution procedure in a way which can confer several advantages, including aiding solution verification. Specifically, three novel contributions will be presented: First, a concept for constructing verifiably-self-consistent numerical evolution schemes for PDE IVPs; second, an iterated, multipoint differential transform method (IMDTM) for numerically evolving PDE IVPs (the IMDTM can be used to efficiently implement verifiably-self-consistent PDE evolution); and finally, in order to efficiently implement the IMDTM scheme, a generalized finite-difference stencil formula is derived which can take advantage of multiple higher-order spatial derivatives when computing even-higher-order derivatives. As will be demonstrated, the performance of these techniques compares favorably to other explicit evolution schemes in terms of speed, memory footprint and accuracy. SUMMARY:An Iterated, Multipoint Differential Transform Method for Numerically Evolving PDE IVPs UID:1195 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:00000000T100000 DTEND;TZID=America/Chicago:00000000T110000 DTSTAMP:20130525T020110 LOCATION:Building 240 - Confernence Room 4301, Argonne National Laboratory DESCRIPTION:Traditional numerical techniques for solving time-dependent partial-differential-equation (PDE) initial-value problems (IVPs) store a truncated representation of the function values and some number of their time derivatives at each time step. What if spatial derivatives were also stored? Although redundant in the dx --> 0 limit, this paper demonstrates that stored spatial derivatives can be propagated in an efficient and self-consistent manner. They can effectively contribute to the evolution procedure in a way which can confer several advantages, including aiding solution verification. Specifically, three novel contributions will be presented: First, a concept for constructing verifiably-self-consistent numerical evolution schemes for PDE IVPs; second, an iterated, multipoint differential transform method (IMDTM) for numerically evolving PDE IVPs (the IMDTM can be used to efficiently implement verifiably-self-consistent PDE evolution); and finally, in order to efficiently implement the IMDTM scheme, a generalized finite-difference stencil formula is derived which can take advantage of multiple higher-order spatial derivatives when computing even-higher-order derivatives. As will be demonstrated, the performance of these techniques compares favorably to other explicit evolution schemes in terms of speed, memory footprint and accuracy. SUMMARY:An Iterated, Multipoint Differential Transform Method for Numerically Evolving PDE IVPs UID:1197 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:00000000T100000 DTEND;TZID=America/Chicago:00000000T110000 DTSTAMP:20130525T020110 LOCATION:Building 240 - Confernence Room 4301, Argonne National Laboratory DESCRIPTION:Traditional numerical techniques for solving time-dependent partial-differential-equation (PDE) initial-value problems (IVPs) store a truncated representation of the function values and some number of their time derivatives at each time step. What if spatial derivatives were also stored? Although redundant in the dx --> 0 limit, this paper demonstrates that stored spatial derivatives can be propagated in an efficient and self-consistent manner. They can effectively contribute to the evolution procedure in a way which can confer several advantages, including aiding solution verification. Specifically, three novel contributions will be presented: First, a concept for constructing verifiably-self-consistent numerical evolution schemes for PDE IVPs; second, an iterated, multipoint differential transform method (IMDTM) for numerically evolving PDE IVPs (the IMDTM can be used to efficiently implement verifiably-self-consistent PDE evolution); and finally, in order to efficiently implement the IMDTM scheme, a generalized finite-difference stencil formula is derived which can take advantage of multiple higher-order spatial derivatives when computing even-higher-order derivatives. As will be demonstrated, the performance of these techniques compares favorably to other explicit evolution schemes in terms of speed, memory footprint and accuracy. SUMMARY:An Iterated, Multipoint Differential Transform Method for Numerically Evolving PDE IVPs UID:1199 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:00000000T100000 DTEND;TZID=America/Chicago:00000000T110000 DTSTAMP:20130525T020110 LOCATION:Building 240 - Confernence Room 4301, Argonne National Laboratory DESCRIPTION:Traditional numerical techniques for solving time-dependent partial-differential-equation (PDE) initial-value problems (IVPs) store a truncated representation of the function values and some number of their time derivatives at each time step. What if spatial derivatives were also stored? Although redundant in the dx --> 0 limit, this paper demonstrates that stored spatial derivatives can be propagated in an efficient and self-consistent manner. They can effectively contribute to the evolution procedure in a way which can confer several advantages, including aiding solution verification. Specifically, three novel contributions will be presented: First, a concept for constructing verifiably-self-consistent numerical evolution schemes for PDE IVPs; second, an iterated, multipoint differential transform method (IMDTM) for numerically evolving PDE IVPs (the IMDTM can be used to efficiently implement verifiably-self-consistent PDE evolution); and finally, in order to efficiently implement the IMDTM scheme, a generalized finite-difference stencil formula is derived which can take advantage of multiple higher-order spatial derivatives when computing even-higher-order derivatives. As will be demonstrated, the performance of these techniques compares favorably to other explicit evolution schemes in terms of speed, memory footprint and accuracy. SUMMARY:An Iterated, Multipoint Differential Transform Method for Numerically Evolving PDE IVPs UID:1201 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110429T100000 DTEND;TZID=America/Chicago:20110429T110000 DTSTAMP:20130525T020110 LOCATION:Building 240 - Confernence Room 4301, Argonne National Laboratory DESCRIPTION:Traditional numerical techniques for solving time-dependent partial-differential-equation (PDE) initial-value problems (IVPs) store a truncated representation of the function values and some number of their time derivatives at each time step. What if spatial derivatives were also stored? Although redundant in the dx --> 0 limit, this paper demonstrates that stored spatial derivatives can be propagated in an efficient and self-consistent manner. They can effectively contribute to the evolution procedure in a way which can confer several advantages, including aiding solution verification. Specifically, three novel contributions will be presented: First, a concept for constructing verifiably-self-consistent numerical evolution schemes for PDE IVPs; second, an iterated, multipoint differential transform method (IMDTM) for numerically evolving PDE IVPs (the IMDTM can be used to efficiently implement verifiably-self-consistent PDE evolution); and finally, in order to efficiently implement the IMDTM scheme, a generalized finite-difference stencil formula is derived which can take advantage of multiple higher-order spatial derivatives when computing even-higher-order derivatives. As will be demonstrated, the performance of these techniques compares favorably to other explicit evolution schemes in terms of speed, memory footprint and accuracy. SUMMARY:An Iterated, Multipoint Differential Transform Method for Numerically Evolving PDE IVPs UID:1203 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110629T150000 DTEND;TZID=America/Chicago:20110629T160000 DTSTAMP:20130525T020110 LOCATION:Building 240, Conference Center 1404-1405, Argonne National Laboratory DESCRIPTION:Mixed Integer Nonlinear Programming (MINLP) problems consist of minimizing a nonlinear function subject to nonlinear constraints and the integrality of a subset of variables. These problems can be solved by branch-and-bound algorithms equipped with procedures that reduce the solution space and allow to find good lower bounds on the optimal solution.\n\nFinding good feasible solutions is equally important and just as difficult, as one deals with two types of nonconvexity: nonlinear constraints and integrality of some variables. Inspired by a technique that proved successful in Mixed Integer Linear problems, we have developed a variant of the Feasibility Pump for MINLPs.\n\nUnlike some previous extensions of the Feasibility Pump to MINLP, our version uses a valid relaxation of the MINLP problem provided by Couenne, an Open-Source solver for MINLPs available from the COIN-OR initiative. We present results on mid-size and large MINLP instances available from well-known instance libraries.\n\n(Joint work with Timo Berthold, Zuse Institute Berlin) SUMMARY:Heuristics to find feasible solutions of Mixed Integer Nonlinear Problems UID:1207 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110824T150000 DTEND;TZID=America/Chicago:20110824T160000 DTSTAMP:20130525T020110 LOCATION:Building 240, 4301, Argonne National Laboratory DESCRIPTION: Modeling the dynamics of polar ice sheets is critical for projections of future sea level rise. Yet, there remain large uncertainties in the basal boundary conditions and in the non-Newtonian constitutive relations employed within ice sheet models. Here, we formulate an inverse problem which involves the solution of a nonlinear full Stokes equation to infer the basal slipperiness and the rheological exponent parameter fields from surface flow velocities. For this purpose, we minimize a regularized misfit functional between observed and modeled surface flow velocities. The resulting least-squares optimization problem is solved using an adjoint-based inexact Newton-conjugate-gradient method. This method requires only Hessian-vector applications rather than computing the full Hessian matrix, which renders it attractive for large-scale inverse problems. Results show that the inexact Newton method is significantly more efficient than the nonlinear conjugate gradient method, and that it is insensitive to the number of inversion parameters. In addition, a numerical study for the reconstructibility of variations in basal slipperiness from surface data shows that the reconstructions converge to the exact slipperiness field as the noise in the synthetic measurements decreases, and that the nonlinear rheology makes the retrieval of the basal slipperiness more difficult. For the inversion of the stress exponent in Glen\'s flow law we find that horizontally constant or smoothly varying volume fields can be reconstructed satisfactorily from noisy surface measurements. SUMMARY:An adjoint-based inexact Newton method for inverse problems governed by nonlinear full Stokes models of ice sheet flows UID:1209 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110506T103000 DTEND;TZID=America/Chicago:20110506T113000 DTSTAMP:20130525T020110 LOCATION:Bldg 240 Conference Rm: 1404 & 1405, ANL DESCRIPTION:Aggressive technological development in nucleic acid sequencing platforms bring unprecedented growth in data volumes and an associated broadening of the range of biological applications to which sequencing is put. The European Nucleotide Archive (http://www.ebi.ac.uk/ena/), a long-standing project under which primary public domain nucleic acid sequencing information is collected and made available in perpetuity, is no stranger to exponential growth in data. The advent of second generation technologies, however, required some substantial new strategic and technical developments. In the talk, I will describe the design and implementation of the service under which ENA archives raw data from next generation platforms. I will outline service that is currently available and will touch upon ongoing developments that aim to make the service more useful for a broader userbase, including improved data submission tools, locus coordinate-based data retrieval and sequence similarity search. Finally, I will outline the approach that we take to building a sustainable and affordable service into the future, based on our compression technology and an outward-looking collaborative working model. SUMMARY:The European Nucleotide Archive: Life on the log scale UID:1217 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110513T100000 DTEND;TZID=America/Chicago:20110513T110000 DTSTAMP:20130525T020110 LOCATION:Building 240, TCS Conference Center, 1404 & 1405, Argonne National Laboratory DESCRIPTION:Current programming models for Graphics Processing Units (GPUs), such as OpenCL and CUDA, require each computational node to be equipped with one or more local GPUs for applications to be able to use them. Recent advances in cloud computing systems, on the other hand, have advocated using virtualization techniques to decouple the application view of \"local hardware resources\" from the physical hardware itself, thus allowing applications to transparently utilize remote hardware. In this seminar, I will talk about VOCL (standing for Virtual OpenCL), which is a new implementation of the OpenCL programming model that provides the OpenCL API while allowing an application to view all GPUs in the system (including remote GPUs) as local virtual GPUs. I will describe the overall idea of VOCL, its expected role in cloud computing systems such as Magellan, the research ideas that went into designing and optimizing it, and some performance numbers. I will also briefly talk about some of the extensions for VOCL that we are currently working on to allow efficient coordinated resource management with virtualized GPU resources. \n\nShucai Xiao is a PhD candidate in the Bradley Department of Electrical and Computer Engineering at Virginia Tech, under the supervision of Prof. Wu-chun Feng. He received his BS and MS degrees from the Beijing University of Posts and Telecommunications and Tsinghua University, respectively, of China. His research interest is programming abstractions and optimizations for general purpose computation on graphics processing units (GPUs). He was an intern at Argonne National Laboratory for transparent virtualization of GPUs in cloud computing systems. SUMMARY:Transparent Virtualization of Graphics Processing Units on Cloud Computing Systems UID:1223 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110710T080000 DTEND;TZID=America/Chicago:20110710T170000 DTSTAMP:20130525T020110 LOCATION:The Brown Palace Hotel, Denver, Colorado DESCRIPTION:Each year, scientists participating in the Scientific Discovery through Advanced Computing Program (SciDAC}, along with other researchers from the computational science community gather at the annual SciDAC conference to present scientific results, discuss new technologies and discover new approaches to collaboration. SUMMARY:SciDAC 2011 Conference UID:1225 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110517T130000 DTEND;TZID=America/Chicago:20110517T140000 DTSTAMP:20130525T020110 LOCATION:Building 240 - Room 4301, Argonne National Laboratory DESCRIPTION:Research often requires transferring large amounts of data among widely distributed resources including supercomputers, instruments, Web portals, local servers, HPC clusters and laptops/desktops. Traditional methods such as FTP and SCP are ill-suited to data movement on this scale due to poor performance and reliability, and custom solutions are costly to develop and operate. Globus Online is a new Software-as-a-Service solution that provides a robust, reliable, secure, and highly monitored environment for file transfers, with powerful yet easy-to-use interfaces. Globus Online simplifies large-scale data movement without requiring construction of custom, end-to-end systems. With Globus Online, the robust file transfer capabilities that were traditionally available only on the largest scale grid implementations are now accessible to all users ‚Äě even those without the benefit of well-funded, technically savvy IT organizations. In this presentation, we will introduce Globus Online, walk through the process of signing up, and show audience members how to use both GUI and CLI interfaces to move data between two endpoints. We will also cover the Globus Connect feature, which allows users to transfer files between a GridFTP server and their local servers or laptops, even if behind a firewall, without the complexity of a full Globus install. SUMMARY:Globus Online: A Cloud-Hosted Service For Secure, Reliable, High-Performance Data Movement UID:1229 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110525T100000 DTEND;TZID=America/Chicago:20110525T110000 DTSTAMP:20130525T020110 LOCATION:Bldg 240 Conference Center 1404-1405, Argonne National Laboratory DESCRIPTION:The lab is starting to roll out a postdoc mentoring program. I will provide you all with some information about it. Both mentors and mentees are invited to attend. The talk will be short and some time is set aside for postdocs to mingle with scientists. SUMMARY:Postdoc Mentoring meeting UID:1237 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110601T150000 DTEND;TZID=America/Chicago:20110601T160000 DTSTAMP:20130525T020110 LOCATION:Bldg 240 Conference Center 1404-1405, Argonne National Laboratory DESCRIPTION:Both the architecture and application software are changing in complexity and heterogeneity. A system can be configured for shared memory, distributed memory, or both. Algorithms can be implemented using multiple parallel programming models, singly or in various combinations. This can potentially result in many possible solutions to the same problem. Performance models are needed to condense the programming and design space and guide programmers and tools to more efficiently develop optimized code using very low overhead techniques. In this talk, I will present some examples in designing performance models and show how the models can be used in improving software design, performance analysis, and optimization. SUMMARY:Using Performance Models to Provide Efficient Programming Solutions on Multicores UID:1241 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110524T100000 DTEND;TZID=America/Chicago:20110524T120000 DTSTAMP:20130525T020110 LOCATION:Bldg 240 Conference Rm: 4301, ANL DESCRIPTION:Soil contains one of the most diverse and unexplored microbial populations on the planet. As part of the Great Prairie DOE project, the JGI has generated several terabases (10**12) of whole metagenome shotgun sequence from a number of sites in the mid-west. Due to the lack of reference genomes, the short reads produced by the Illumina sequencers, and the volume of data, we have chosen to apply a de novo assembly approach to this data. This in turn has been extremely challenging because existing assemblers are poorly equipped to deal with large volumes of metagenomic data. We have developed a novel approach to assembly in order to solve these issues, and are also applying this approach to massive-scale mRNAseq assembly. SUMMARY:Soil metagenome assembly: diverse data, diverse challenges UID:1243 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110526T103000 DTEND;TZID=America/Chicago:20110526T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, TCS Conference Center, 1406 & 1407, Argonne National Laboratory DESCRIPTION:This talk will introduce the audience to HPC activities at Center for Development of Advanced Computing (C-DAC) and will focus on systems developed in Hardware Technology Development group (HTDG). Indigenously built High Performance Cluster Interconnect, Reconfigurable Computing Systems and the associated System Software will be discussed. Additionally it will provide configuration details of PARAM Yuva, a 54TF cluster housed at National PARAM Supercomputing Facility (NPSF) in Pune, India.\n\nBio:\nMr. Yogeshwar Sonawane is working as a Member Technical Staff (MTS) with Hardware Technology Development Group (HTDG) of Center for Development of Advanced Computing (C-DAC) at Pune, India since 2004. He is associated with System Software development in the areas of High Performance Interconnect for Cluster Computing and Reconfigurable Computing Systems. His chief expertise lies in development of light weight communication stacks for RDMA capable interconnects and software interfaces for Reconfigurable Computing hardware. His research interests include High Speed Networking Systems, efficient protocol stacks, MPI Scalability and Embedded Systems. Yogeshwar has a Bachelor of Engineering (B.E.) degree in Electronics & Telecommunications from Govt. College of Engineering, Pune, India. SUMMARY:HPC Activities at C-DAC, India: An overview UID:1245 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110617T100000 DTEND;TZID=America/Chicago:20110617T110000 DTSTAMP:20130525T020110 LOCATION:Building 240 Conference Room 1406, Argonne National Laboratory DESCRIPTION:With advances in algorithms and growing computing powers, quantum Monte Carlo (QMC) methods have become a leading contender for high accuracy calculations for the electronic structure of realistic systems.\n \nIn this talk, I present recent development in QMC theories and numerical algorithms which have allowed us to reach the accuracy and system sizes at unprecedented time-to-solution. I discuss QMC implementations to overcome the important efficiency and scalability bottlenecks encountered with the HPC systems based on the multi/many-core architecture of today and present state-of-art QMC calculations of solid-state and molecular systems using tens or hundred thousand cores on the petascale computers.\n\nFinally, I examine the readiness of QMC for the future HPC architectures to harness ever-increasing computing powers to tackle outstanding materials and chemical problems. SUMMARY:QMC, harnessing computing powers of today and beyond UID:1247 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110617T093000 DTEND;TZID=America/Chicago:20110617T103000 DTSTAMP:20130525T020110 LOCATION:TCS Conference Center Rms 1404 & 1405, Argonne National Laboratory DESCRIPTION:MPI (Message Passing Interface) is a standard, portable interface for writing message-passing parallel programs. Dave will give an introduction to MPI and describe various MPI-related projects in MCS. SUMMARY:MPI Tutorial UID:1303 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110620T093000 DTEND;TZID=America/Chicago:20110620T103000 DTSTAMP:20130525T020110 LOCATION:TCS Conference Center Rms 1404 & 1405, Argonne National Laboratory DESCRIPTION:This will be a highly interactive presentation with a focus on scientific consensus versus what makes a great headline in the news. Doug is the ARM Climate Research Facility (ACRF) Operations Manager. The ACRF is a DOE national user facility with word-wide sites. Check it out: www.arm.gov SUMMARY:Global Warming: The Real Deal or Just a Good Story? UID:1305 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110620T093000 DTEND;TZID=America/Chicago:20110620T103000 DTSTAMP:20130525T020110 LOCATION:TCS Conference Center Rms 1404 & 1405, Argonne National Laboratory DESCRIPTION:This will be a highly interactive presentation with a focus on scientific consensus versus what makes a great headline in the news. Doug is the ARM Climate Research Facility (ACRF) Operations Manager. The ACRF is a DOE national user facility with word-wide sites. Check it out: www.arm.gov SUMMARY:Global Warming: The Real Deal or Just a Good Story? UID:1307 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110720T150000 DTEND;TZID=America/Chicago:20110720T160000 DTSTAMP:20130525T020110 LOCATION:Bldg 240 Conference Center 1404-1405, Argonne National Laboratory DESCRIPTION:Problems governed by linear operators that are symmetric positive definite (SPD) enjoy a natural association with an energy norm, which finds many uses both in analysis and application. For example, Galerkin discretizations are optimal in the energy norm (a consequence of Cea\'s lemma), as is the conjugate gradient iteration. In the first part of this talk I will discuss a generalization of the energy norm to nonsymmetric operators. In the second part I will discuss how these ideas enable the convergence analysis of alebraic multigrid for SPD matrices, where the usual energy norm features heavily, to be generalized to the nonsymmetric case. SUMMARY:An \"energy norm\" for nonsymmetric problems with application to algebraic multigrid UID:1301 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110606T093000 DTEND;TZID=America/Chicago:20110606T103000 DTSTAMP:20130525T020110 LOCATION:TCS Conference Center Rms 1404 & 1405, Argonne National Laboratory DESCRIPTION:Tisha will introduce the Argonne Leadership Computing Facility (ALCF), its mission, its current systems, and systems expected to arrive at Argonne in the near future. The ALCF (http://www.alcf.anl.gov/) is a national leadership computing facility designed to provide resources that make computationally intensive projects of the largest scales possible. Currently the ALCF supports 36 INCITE computational science projects covering disciplines as diverse as protein folding to modeling aircraft engines SUMMARY:Overview of the Argonne Leadership Computing Facility UID:1255 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110606T103000 DTEND;TZID=America/Chicago:20110606T113000 DTSTAMP:20130525T020110 LOCATION:Building 240 Conference Room 4301, Argonne National Laboratory DESCRIPTION:The Laplacian matrices of graphs are large sparse symmetric matrices fundamental to numerous scientific computing problems. In addition to linear algebra and graph-theoretic applications, they arise in classification and machine learning; spectral clustering; dimensionality reduction and data mining; discretization of elliptic partial-differential equations on unstructure grids; interior-point algorithms for transportation network flows; and electrical resistor networks. \n\nWe present Lean Algebraic Multigrid (LAMG), a fast solver of the graph Laplacian linear system. LAMG consists of a setup phase that requires O(m) operations, and an iterative solve phase using multigrid cycles, requiring O(m log(1/eps)) operations for eps-accuracy. LAMG is an aggregation-based algebraic multigrid variant that relies on lean, efficient components: caliber-1 interpolation operators, and relaxation-guided aggregation. We also employ a coarse-level energy correction to maintain good asymptotic convergence without excessive fill-in at coarser levels of the multigrid hierarchy. \n\nWe present numerical experiments for over 1600 real-world graph instances with up to several millions of nodes and edges. LAMG maintained its optimal efficiency for all those graphs. Although these are preliminary results, to the best of our knowledge LAMG is the first practical linear-scaling solver of the graph Laplacian, and can potentially speed-up the associated computational applications by orders of magnitudes. The developed multiscale methodology can also be extended to non M-matrices and to other computational graph problems. SUMMARY:Lean Algebraic Multigrid (LAMG): Fast Graph Laplacian Linear Solver UID:1259 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110607T093000 DTEND;TZID=America/Chicago:20110607T103000 DTSTAMP:20130525T020110 LOCATION:TCS Conference Center Rms 1404 & 1405, Argonne National Laboratory DESCRIPTION:A programming model is the high-level structure within which an algorithm is implemented. The complexity of the programming model required for a given application depends greatly on the type of parallelism to be expressed. In the first part of this talk, I will highlight some of the most common programming models (e.g. master-worker) and describe their implementation within MPI. The second part will focus on asynchronous programming models using one-sided communication and the utility of this model within application codes in biology and chemistry. SUMMARY:Programming Models for High Performance Scientific Computing UID:1263 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110907T150000 DTEND;TZID=America/Chicago:20110907T160000 DTSTAMP:20130525T020110 LOCATION:Bldg 240 Conference Center 1404-1405, Argonne National Laboratory DESCRIPTION:We analyze the convergence properties of a parallel Newton scheme for differential systems. The scheme concurrently solves the time-coupled nonlinear systems arising\nfrom the application of implicit discretization schemes. We have found that the scheme acts as a tracking algorithm that converges to a moving manifold given by the solution of the nonlinear system at the current time step parameterized in the iterating solution\nof the previous step. This property explains why the method can significantly reduce the number of iterations compared with sequential Newton methods. The method exhibits a theoretical lower bound on the number of iterations equal to the number of discretization points. A numerical study using a detailed dynamic power grid model is provided to demonstrate the scalability of the method. SUMMARY:Convergence Analysis of a Parallel Newton Scheme for Dynamic Power Grid Simulations UID:1265 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:00000000T150000 DTEND;TZID=America/Chicago:00000000T160000 DTSTAMP:20130525T020110 LOCATION:Bldg 240 Conference Center 1404-1405, Argonne National Laboratory DESCRIPTION:To be announced. SUMMARY:Title to be announced UID:1267 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:00000000T150000 DTEND;TZID=America/Chicago:00000000T160000 DTSTAMP:20130525T020110 LOCATION:Bldg 240 Conference Center 1404-1405, Argonne National Laboratory DESCRIPTION:To be announced. SUMMARY:Title to be announced UID:1269 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110728T103000 DTEND;TZID=America/Chicago:20110728T113000 DTSTAMP:20130525T020110 LOCATION:Bldg 240 Conference Center 1416, Argonne National Laboratory DESCRIPTION:Given an undirected graph G, the vertex separator problem is to find the smallest number of nodes whose removal disconnects the graph into disjoint subsets A and B, where A and B are subject to size constraints. We will show how this problem can be formulated as a continuous quadratic program. We use the QP as a local processor in a multilevel scheme for solving large scale instances of the problem. Numerical results will be presented. SUMMARY:A Continuous Multilevel Solver for the Vertex Separator Problem UID:1271 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110609T093000 DTEND;TZID=America/Chicago:20110609T103000 DTSTAMP:20130525T020110 LOCATION:TCS Conference Center Rms 1404 & 1405, Argonne National Laboratory DESCRIPTION:In this talk I will discuss the role that high-performance computing and advanced data systems play in accelerating the transition of biology from a science primarily focused on description and explanation to a new science focused on systems level understanding and data driven predictive theories. I will discuss the role that genomics driven biological models are playing in reshaping what is possible in biological analysis and how high-performance computing technologies are playing a critical role in enabling new types of biological investigations. I’ll outline how we go from raw DNA sequence data to predictive whole genome models of biological phenotype and how automated annotation and analysis systems are enabling small biological labs to do things that were not possible at any scale just a few years ago. I will project how the increases in computing capability coupled with the increases in biological sequence generation throughput and the rise of data integration methods will result in a rapid expansion of our understanding. If time permits I will discuss the impact that predictive biological models can have on health, energy, the environment and science. SUMMARY:High-Performance Computing and Biology: The Quest for a Predictive Biological Theory UID:1273 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110623T103000 DTEND;TZID=America/Chicago:20110623T113000 DTSTAMP:20130525T020110 LOCATION:Bldg 240 Conference Center 1406-1407, Argonne National Laboratory DESCRIPTION:This talk is a summary of our recent research on a rigorous treatment of the continuous-time dynamic user equilibrium (DUE) problem, based on its formulation as a differential complementarity system (DCS). After a synopsis of the DUE problem to date and a highlight of the mathematical and computational challenges of the latter DCS for the general DUE problem, we present two important special cases of this problem: the single-bottleneck problem as a linear complementarity system (LCS) and an instantaneous model as a DCS with constant delays. Both cases are based on a modified point-queue model that has some desirable mathematical properties and is consistent with practical requirements. Numerical solution of the resulting differential systems by time-stepping methods and finite-dimensional complementarity methods is discussed and results are presented.\n\nSupported by the National Science Foundation, our research is based on joint work with Xueguang (Jeff) Ban at Rensselaer Polytechnic Institute, Henry Liu at the University of Minnesota, Lanshan Han and Satish Ukkusuri at Purdue University, and Ramadurai Gitakrishnan at the Indian Institute of Technology at Madras. SUMMARY:Continuous-time dynamic user equilibrium via differential complementarity systems UID:1297 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110614T103000 DTEND;TZID=America/Chicago:20110614T113000 DTSTAMP:20130525T020110 LOCATION:Building 240 / Room 4301, Argonne National Laboratory DESCRIPTION:The numerical solution of the Vlasov equation is usually performed by the particle-in-cell (PIC) method. However, it is well known that, in some cases, the PIC method has difficulty in having an accurate description for the distribution function in phase space due to numerical noise, the inherent drawback of particle-based methods. In this talk, I will present an accurate and efficient PIC method for computing the dynamics of kinetic plasmas. The method overcomes the numerical noise by periodically remapping the distribution function on a hierarchy of locally refined grids in phase space. The positivity of the distribution function is enforced by redistributing excess phase space density in a local neighborhood. Remapping on phase space grid also provides an opportunity to integrate a collisional model and an associated grid-based solver. At the end of this talk, I will show our numerical results on a set of standard plasma physics problems; e.g., Landau damping and the two-stream instability in both 1D and 2D cases. It is shown that remapping reduces the numerical noise significantly and results in a more consistent second order method than the standard PIC method. \n SUMMARY:A Particle-in-Cell Method with Adaptive Phase-Space Remapping for Kinetic Plasmas UID:1279 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110616T093000 DTEND;TZID=America/Chicago:20110616T103000 DTSTAMP:20130525T020110 LOCATION:TCS Conference Center Rms 1404 & 1405, Argonne National Laboratory DESCRIPTION:Developing for HPC challenges the functionality and utility of any development environment. Luckily, Python is supported as a programming language just about everywhere else these days. This talk is meant to give a brief introduction on how to get started writing parallel applications in Python, some of the advantages and disadvantages, and examples of success stories. For bonus, we will use the ALCF Blue Gene /P to try and set a sortbenchmark.org for joule sort. SUMMARY:Using Python for Scientific and High Performance Computing UID:1295 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110610T080000 DTEND;TZID=America/Chicago:20110610T170000 DTSTAMP:20130525T020110 LOCATION:Brown Palace Hotel, Denver, Colorado DESCRIPTION:Each year, scientists participating in the Scientific Discovery through Advanced Computing Program (SciDAC), along with other researchers from the computational science community gather at the annual SciDAC conference to present scientific results, discuss new technologies and discover new approaches to collaboration.\n\nIn addition to highlighting successes from the SciDAC program, the conference is a general celebration of computational science. We bring together computational scientists from different nations, agencies, programs, and application domains to highlight recent advances in computational science in important areas: from understanding our universe on its largest and smallest scales, to understanding Earth’s climate change and its ramifications for humankind, to developing new energy sources. Application talks will focus on performance and scaling issues. Enabling technologies talks will focus on petascale applications and architectures. SUMMARY:SciDAC 2011 Conference UID:1281 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110613T093000 DTEND;TZID=America/Chicago:20110613T103000 DTSTAMP:20130525T020110 LOCATION:TCS Conference Center Rms 1404 & 1405, Argonne National Laboratory DESCRIPTION:In many science areas, the quest for increased computational resources is driven by a need to span a broader range of scales, that is, to capture the interaction of small scales with the large. In transport problems such as \nelectromagnetics and fluid mechanics, this implies a need to propagate small scale features over long times and distances. In numerical simulations, such long-time integrations are most efficiently realized by using high-order discretizations. Here, we present recent advances in spectral element methods designed for the petascale efficient single- and multi-node performance. Application areas include the study of magnetorotational turbulence in accretion disk models, heat transfer in advanced reactor designs, wakefield computations in accelerators, and transition to turbulence in vascular flows. SUMMARY:Numerical Algorithms for Petascale Science UID:1285 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110614T093000 DTEND;TZID=America/Chicago:20110614T103000 DTSTAMP:20130525T020110 LOCATION:TCS Conference Center Rms 1404 & 1405, Argonne National Laboratory DESCRIPTION:Joe will present a slide show of scientific visualizations that have been created by staff of Argonne and the University of Chicago. The format will be interactive with a discussion of the process and techniques used. SUMMARY:Scientific Visualization: A Slide Show UID:1287 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110706T103000 DTEND;TZID=America/Chicago:20110706T113000 DTSTAMP:20130525T020110 LOCATION:Building 240 Seminar Room 4301, Argonne National Laboratory DESCRIPTION:White noise is a very common way for accounting for randomness in the inputs to partial differential equations, especially in cases where little is know about those inputs. On the other hand, pink noise, or more generally, colored noise having a power spectrum that decays as $1/f^\\alpha$, where $f$ denotes the frequency and $\\alpha\\in(0,2]$, has been found to accurately model many natural, social, economic, and other phenomena. Our goal is to study, the effects of modeling random inputs as $1/f^\\alpha$ random fields, including the white noise ($\\alpha=0$), pink noise ($\\alpha=1$), and brown noise ($\\alpha=2$)cases. We show how such random fields can be approximated so that they can be used in computer simulations. We then show that the solutions of the differential equations exhibit a strong dependence on $\\alpha$, indicating that further examination of how randomness in partial differential equation is modeled and simulated is warranted. SUMMARY:Colored Noise and its Effect On the Solutions to Differential Equations UID:1289 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110616T140000 DTEND;TZID=America/Chicago:20110616T150000 DTSTAMP:20130525T020110 LOCATION:Building 240 Conference Room 1404, Argonne National Laboratory DESCRIPTION:Machine learning has recently become a key computational tool for the analysis and understanding of scientific data. This talk describes our efforts in providing state of the art techniques in data analysis by doing research that lies at the intersection of machine learning, statistics, data mining, physics and astronomy. The talk discusses two projects. The first is on the development of an efficient pattern recognition tool for the automated identification and classification of tracks of ionizing radiation as measured by a pixel detector system. Such software has many applications including dosimeters to assess the risk of human exposure to radiation, and area monitors to characterize the general background radiation environment harmful to humans and electronic equipment. The second application is on the automatic geomorphic mapping of planet Mars; the goal is to divide a landscape (represented by an image, digital elevation model DEM, or other spatial datasets) into a set of landscape elements having specific surface patterns. These elements are later grouped into a set of clear landforms (e.g., craters, valleys, ridges, etc.). SUMMARY:Machine Learning in Space and Planetary Science UID:1293 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110623T150000 DTEND;TZID=America/Chicago:20110623T160000 DTSTAMP:20130525T020110 LOCATION:Building 240, Conference Room 1404/1405, Argonne National Laboratory DESCRIPTION:In this presentation, an overview of the Global Precipitation Mission\'s ground validation (GPM-GV) program will be discussed, including its scientific rationale and implementation strategy. GPM is planned to launch in mid-2013, and \"day 1\" retrieval algorithms are currently being developed for the new Ku-Ka dual-frequency precipitation radar and high frequency passive microwave sensors (>95 GHz) to be deployed on the core satellite. Validation data are being collected pre-launch in a series of GV experiments spanning a range of precipitation regimes over both land and ocean. The Mid-Latitude Continental Convective Clouds Experiment (MC3E), conducted jointly between the Department of Energy Atmospheric Systems Research (ASR) program and the NASA-GPM project from April-June 2011, successfully collected over a dozen cases of precipitation proximate to the Atmospheric Radiation Measurement (ARM) Southern Great Plains (SGP) facility in north-central Oklahoma, as well as coordinated aircraft sampling over the broader southern and high plains of the United States. Constituting a major field campaign in the GPM-GV program, preliminary results from the MC3E field campaign will be presented, as well as a perspective on how continued collaboration between DOE ASR and NASA-GPM can mutually benefit both programs. SUMMARY:The Mid-Latitude Continental Convective Clouds Experiment from a NASA Global Precipitation Measurement Perspective UID:1309 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110623T093000 DTEND;TZID=America/Chicago:20110623T103000 DTSTAMP:20130525T020110 LOCATION:TCS Conference Center Rms 1404 & 1405, Argonne National Laboratory DESCRIPTION:Stochastic and dynamic optimization are key paradigms arising in multi-stage decision making under uncertainty. In this lecture, we present fundamentals on theory and algorithms and discuss available software tools. We will motivate these concepts through applications arising in the electrical power grid. We will also motivate some discussion by summarizing some of the open research questions in the area SUMMARY:Stochastic and Dynamic Optimization: Theory, Algorithms, and Applications UID:1311 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110629T103000 DTEND;TZID=America/Chicago:20110629T113000 DTSTAMP:20130525T020110 LOCATION:Building 240 Seminar Room 4301, Argonne National Laboratory DESCRIPTION:A Higher Order Multiscale Method for Polymer-Laden Flows.\nPolymer solutions exhibit fluid dynamic behavior not easily modeled with continuum stress constitutive laws. To simulate such fluids, and to perform numerical experiments aimed at improving continuum constitutive closures, we are working toward a multiscale approach where the macroscopic stress is derived directly from molecular-scale models coupled to the macroscopic equations. Our approach begins with Kramers freely-jointed polymer model. This model is a coarse graining of atomistic representations that is known to capture many essential features of real polymers, and a wealth of analysis provides verification and validation tests for our numerical method. We derive a new second order accurate discretization of the stochastic Langevin equations for this model, demonstrate second-order coupling\nof this molecular model to continuum solvers, and show progress toward the goal of a full multiscale rheology calculation. SUMMARY:A Higher Order Multiscale Method for Polymer-Laden Flows UID:1313 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110629T103000 DTEND;TZID=America/Chicago:20110629T113000 DTSTAMP:20130525T020110 LOCATION:Building 240 Seminar Room 4301, Argonne National Laboratory DESCRIPTION:A Higher Order Multiscale Method for Polymer-Laden Flows.\nPolymer solutions exhibit fluid dynamic behavior not easily modeled with continuum stress constitutive laws. To simulate such fluids, and to perform numerical experiments aimed at improving continuum constitutive closures, we are working toward a multiscale approach where the macroscopic stress is derived directly from molecular-scale models coupled to the macroscopic equations. Our approach begins with Kramers freely-jointed polymer model. This model is a coarse graining of atomistic representations that is known to capture many essential features of real polymers, and a wealth of analysis provides verification and validation tests for our numerical method. We derive a new second order accurate discretization of the stochastic Langevin equations for this model, demonstrate second-order coupling\nof this molecular model to continuum solvers, and show progress toward the goal of a full multiscale rheology calculation. SUMMARY:A Higher Order Multiscale Method for Polymer-Laden Flows UID:1315 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110624T093000 DTEND;TZID=America/Chicago:20110624T103000 DTSTAMP:20130525T020110 LOCATION:TCS Conference Center Rms 1404 & 1405, Argonne National Laboratory DESCRIPTION:In this talk, I will discuss some applications of numerical optimization and provide an overview of some methods for solving them. SUMMARY:Numerical Optimization and Application UID:1317 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110706T150000 DTEND;TZID=America/Chicago:20110706T160000 DTSTAMP:20130525T020110 LOCATION:Bldg 240 Conference Center 1404-1405, Argonne National Laboratory DESCRIPTION:We present a scalable approach and implementation for solving stochastic programming problems, with application to the optimization of complex energy systems under uncertainty. Stochastic programming incorporates a model of uncertainty about future events, which in our case relates to the output of highly variable renewable energy sources such as wind. The Sample Average Approximation (SAA) is used to obtain a very large deterministic problem, which necessitates the use of parallel computing to solve.\n\nOur code, PIPS, is based on primal-dual interior-point methods (IPMs) and uses a classical Schur-complement technique to obtain a sample- or scenario-based decomposition at the linear algebra level of IPMs. We review the Schur-complement decomposition and present our novel method for solving the Schur-complement system using distributed dense linear algebra, allowing, for the first time, problems with up to 100,000 first-stage variables (and ~2 billion total variables) to be solved efficiently. We discuss our experience with implementing a hybrid parallel model and porting PIPS to Blue Gene/P, where we obtained strong scaling efficiency of 90%+ on 32 racks (131,072 cores) of Intrepid. Our application demonstrates the potential of efficiently using High Performance Computing resources for mathematical optimization. SUMMARY:Scalable Stochastic Optimization of Complex Energy Systems UID:1319 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110714T150000 DTEND;TZID=America/Chicago:20110714T160000 DTSTAMP:20130525T020110 LOCATION:Bldg 240 Conference Center 1416, Argonne National Laboratory DESCRIPTION:The gap between modern computer architectures and software performance speed rate has been ever increasing rapidly. The reason being is that compilers often cannot generate efficient object codes. In this presentation, we will re-visit Jim Cody\'s (Argonne Scientist) concepts of subroutine library functions. And a LMG(lean, mean, and green) algorithm that can produce loop-less computer codes to reach peak performance on any targeted computer systems for any well-defined computational tasks will be presented. The algorithm can be applied to produce codes for several kernel functions to improve the performance of some DOE scientific application packages. Some performance results will be shown. SUMMARY:Is there an easy route to reach the peak of performance? UID:1321 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110727T150000 DTEND;TZID=America/Chicago:20110727T160000 DTSTAMP:20130525T020110 LOCATION:Bldg 240 Conference Center 1404-1405, Argonne National Laboratory DESCRIPTION:I present two-phase matrix splitting methods for solving bound-constrained quadratic programs (BQPs) and linear complementarity problems (LCPs). The method for solving BQPs uses matrix splitting iterations to generate descent directions that drive convergence of the iterates and rapidly identify those variables that are active at the solution. The second-phase uses this prediction to further refine the active set and to accelerate convergence. The method for solving LCP combines matrix splitting iterations with a \"natural\" merit function. This combination allows one to prove convergence of the method and maintain excellent practical performance. Once again, a second subspace phase is used to accelerate convergence. I present numerical results for both algorithms on CUTEr test problems, randomly generated problems, and the pricing of American options. SUMMARY:Matrix splitting methods for bound-constrained quadratic programming and linear complementarity problems UID:1323 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110627T093000 DTEND;TZID=America/Chicago:20110627T103000 DTSTAMP:20130525T020110 LOCATION:TCS Conference Center Rms 1404 & 1405, Argonne National Laboratory DESCRIPTION:Moving large quantities of data across wide area networks can be a painful and laborious process. The Center for Enabling Distributed Petascale Science (CEDPS) project has developed a service called Globus Online that automates this process. Users hand off data transfer tasks (e.g., move a set of files, mirror a directory) to the hosted Globus Online service, which then does its best to ensure that the transfer completes successfully, monitoring performance and errors, correcting problems automatically whenever possible, and reporting back on status and errors. Thus, we eliminate the need to babysit transfers, freeing users to focus on domain-specific work. This lecture includes an overview of the motivation for Globus Online, a discussion of key design concepts, and a demonstration of current capabilities. SUMMARY:Introduction to Globus Online UID:1325 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110628T093000 DTEND;TZID=America/Chicago:20110628T103000 DTSTAMP:20130525T020110 LOCATION:TCS Conference Center Rms 1404 & 1405, Argonne National Laboratory DESCRIPTION:The presentation will provide an overview of how Cyber Security is implemented at an open science, highly collaborative DOE National Laboratory. Topics discussed will include: \"Worst to First\" - Evolution of the ANL Cyber Program , technical highlights, current collaborative projects and futures. SUMMARY:Cyber Security at a National Laboratory UID:1327 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110817T150000 DTEND;TZID=America/Chicago:20110817T160000 DTSTAMP:20130525T020110 LOCATION:Bldg 240 Conference Center 1404-1405, Argonne National Laboratory DESCRIPTION:A dimension reduction method called Discrete Empirical Interpolation (DEIM) is described and shown to dramatically reduce the computational complexity of the popular Proper Orthogonal Decomposition (POD) method for constructing reduced-order models for parametrized nonlinear partial differential equations (PDEs). DEIM is a technique for reducing the complexity of evaluating the reduced order nonlinear terms obtained with the standard POD-Galerkin. POD reduces dimension in the sense that far fewer variables are present, but the complexity of evaluating the nonlinear term remains that of the original problem. DEIM is a modification of POD that reduces complexity of the nonlinear term of the reduced model to a cost proportional to the number of reduced variables obtained by POD. The method applies to arbitrary systems of nonlinear ODEs, not just those arising from discretization of PDEs. \n\nIn this talk, the DEIM method will be developed along with a discussion of its approximation properties. Applications in Shape Optimization and PDE constrained optimization shall be emphasized. Additional applications from Chemically Reacting Flow and Neural Modeling will be presented to illustrate the effectiveness and wide applicability of the DEIM approach. SUMMARY:The Discrete Empirical Interpolation Method for Nonlinear Model Reduction UID:1329 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110707T120000 DTEND;TZID=America/Chicago:20110707T000008 DTSTAMP:20130525T020110 LOCATION:Cafeteria, Private Dining Rooms A&B, Argonne National Laboratory DESCRIPTION:See http://www.phy.anl.gov/theory/astrolunch.html SUMMARY:Cosmic Fireworks: An Introduction to Supernovae UID:1331 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110705T093000 DTEND;TZID=America/Chicago:20110705T103000 DTSTAMP:20130525T020110 LOCATION:TCS Conference Center Rms 1404 & 1405, Argonne National Laboratory DESCRIPTION:Basic concepts of parallel programming will be discussed, along with how to get access to Argonne\'s Laboratory Computing Resource Center. Presentation is largely meant for those new to parallel programming. The presentation covers common terminology, basic parallel hardware models, and parallel programming models. A few basic examples will also be discussed. SUMMARY:Introduction to Parallel Computing UID:1333 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110707T093000 DTEND;TZID=America/Chicago:20110707T103000 DTSTAMP:20130525T020110 LOCATION:TCS Conference Center Rms 1404 & 1405, Argonne National Laboratory DESCRIPTION:The spatial domain definition, usually in the form of a mesh, plays an important role in many codes solving systems of PDEs on massively parallel systems. In addition to defining the spatial domain, the mesh is also the basis for partitioning these problems for parallel solution, and is the vehicle for solution data during post-processing. I will discuss the MOAB mesh library and its use in these various contexts. Related libraries for CAD geometry (CGM) and mesh-geometry relations (Lasso) will also be described. SUMMARY:Geometry and Mesh Components for Scientific Computing UID:1335 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110708T093000 DTEND;TZID=America/Chicago:20110708T103000 DTSTAMP:20130525T020110 LOCATION:TCS Conference Center Rms 1404 & 1405, Argonne National Laboratory DESCRIPTION:I will summarize my experience porting the GPAW code on Blue Gene/P and emphasize high-level concepts in performance analysis. Time permitting; I will discuss the implications for extended these algorithms to Blue Gene/Q. SUMMARY:Performance Analysis: An Application Developer\'s Perspective UID:1339 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110711T093000 DTEND;TZID=America/Chicago:20110711T103000 DTSTAMP:20130525T020110 LOCATION:TCS Conference Center, Rms 1404 and 1405, Argonne National Laboratory DESCRIPTION:Our smartphones are providing increasingly sophisticated and useful social and commercial services. These rely on our willingness to reveal our location and to provide financial and identify information to a growing number of third parties, from Amazon to Google to Paypal to Starbucks. Reaction to news about smartphones tracking user movement, or GPS user data being disclosed to governments for optimization of speed traps has been mixed, but suggest that our comfort with loss of privacy may be tied to unexamined and in many cases naive expectations regarding its future availability and use. We will discuss the technology behind these types of services and experimental systems we are exploring with privacy as a design factor rather than an add-on, or completely absent, feature. SUMMARY:“Pervasive Technologies: Will an Intelligent City be Friendly?” UID:1341 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110712T093000 DTEND;TZID=America/Chicago:20110712T103000 DTSTAMP:20130525T020110 LOCATION:TCS Conference Center, Rms 1404 and 1405, Argonne National Laboratory DESCRIPTION:Climate models are the main source of information about how our climate might change with increases in greenhouse gases. Building a climate model has been called the most complex problem in computational science. Find out what makes up a climate model and how they are used at this talk. SUMMARY:Modeling the Climate System UID:1345 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110714T093000 DTEND;TZID=America/Chicago:20110714T103000 DTSTAMP:20130525T020110 LOCATION:TCS Conference Center, Rms 1404 and 1405, Argonne National Laboratory DESCRIPTION:The Argonne Leadership Computing Facility (ALCF) houses one of the largest installations of IBM\'s BlueGene/P, a massively parallel supercomputer. This talk will present an overview of the BlueGene/P architecture and why the system is attractive to application developers. SUMMARY:Introduction to Blue Gene/P Architecture UID:1347 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110715T103000 DTEND;TZID=America/Chicago:20110715T113000 DTSTAMP:20130525T020110 LOCATION:Building 240 / Room 4301, Argonne National Laboratory DESCRIPTION:This talk is concerned with the direct numerical simulation (DNS) of turbulent combustion and its application to the study of transverse reactive jet flows. Due to their interesting mixing properties, transverse jets are important to a variety of industrial applications (film cooling, primary or dilution jets in gas turbines). More recently, microjets have also been used to suppress combustion instabilities in lean premixed combustion systems. Understanding the detailed physics of the flow is important for optimizing the effectiveness of the microjets and the design of stable clean burning systems. To capture the jet complex structure, we developed a fast, accurate and parallel 3D code for direct simulation of turbulent reactive flows using mixed Eulerian-Lagrangian numerical methods. Lagrangian particle methods have been chosen for their natural adaptivity as they capture the multiscale nature of turbulence. We combine particle methods with Eulerian grid-based methods to solve the reactive transport equations.\n\nPlease click on the link to add this seminar to your calendar. SUMMARY:Adaptive Lagrangian/Eulerian Methods for High Reynolds Number Reactive Transverse Jet Simulations UID:1349 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110715T093000 DTEND;TZID=America/Chicago:20110715T103000 DTSTAMP:20130525T020110 LOCATION:TCS Conference Center, Rms 1404 and 1405, Argonne National Laboratory DESCRIPTION:The Swift parallel scripting language lets users apply parallel composition constructs to existing sequential or parallel programs to express highly parallel scripts. Swift scripts are flexible and portable, and can run efficiently on platforms ranging from multicore workstations to petascale supercomputers. For performing parameter sweeps and data analysis with exiting application programs, parallel scripting is typically easier and more productive than tightly-coupled parallel programming.\n\nThis talk will provide an overview of Swift and how its used to run scientific applications in parallel on clusters, grids, clouds, and petascale systems. The architectural challenges of scripting on large-scale systems will be covered, and case studies will be presented. Swift’s place in the taxonomy of parallel programming languages and environments will be discussed, and speculative ideas for hybrid models for multi-level programming to go beyond the petascale will be considered. \n SUMMARY:Swift: Parallel Scripting for Grids, Clouds, and Supercomputers UID:1351 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110718T093000 DTEND;TZID=America/Chicago:20110718T103000 DTSTAMP:20130525T020110 LOCATION:TCS Conference Center, Rms 1404 and 1405, Argonne National Laboratory DESCRIPTION:While the number of flops (floating point operations per second) of high performance computing systems has been steadily increasing, other performance aspects such as I/O speed have failed to keep up. As a result, I/O more and more dominates application execution time. This lecture will provide an introduction to parallel I/O, a technique aimed at increasing application I/O throughput. SUMMARY:Introduction to Parallel I/O UID:1353 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110802T103000 DTEND;TZID=America/Chicago:20110802T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, Conference Rooms 1404 and 1405, Argonne National Laboratory DESCRIPTION:The Parallel Log Structured File System (PLFS) was developed at the Los Alamos National Laboratory (LANL) to improve shared file write performance. Write performance is improved as PLFS transparently transforms the writes such that each process, while logically writing to a shared file, is physically writing to a unique file. By removing this concurrency, PLFS improved the write performance of many applications by multiple orders of magnitude. This was demonstrated on PanFS, Lustre, and GPFS but was not reproduced on PVFS. However, reconstructing the logical file from the multitude of physical files has proven difficult. To alleviate this issue we developed several collective techniques to aggregate information from multiple component pieces. This enables PLFS to maintain it\'s large write improvements without sacrificing read performance for many workloads. There are other workloads, however, which remain challenging. Currently, Los Alamos is developing a scalable HPC key-value store to address these remaining challenges. Additionally, the transformative properties of PLFS have recently also been leveraged to improve the metadata performance of a parallel file system. Finally, I will discuss some preliminary ideas about using PLFS to improve storage availability. \n\nShort Bio: \n\nDr. Adam Manzanares is currently a Nicholas C. Metropolis postdoctoral fellow at the Los Alamos National Laboratory (LANL). He was appointed this position in November 2010 after joining LANL in July 2010 as a postdoctoral researcher. Dr. Manzanares received his Ph.D. from Auburn University in May 2010 with a focus on energy efficient storage systems. Dr. Manzanares is currently focused on storage systems for high performance computing applications. Dr. Manzanares develops middleware layers to improve the performance of HPC storage systems. Dr. Manzanares is also currently researching compression techniques and data formatting libraries for scientific data sets. SUMMARY:Opportunities and Challenges of the Parallel Log Structured File System UID:1355 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110901T103000 DTEND;TZID=America/Chicago:20110901T113000 DTSTAMP:20130525T020110 LOCATION:Building 240 Seminar Room 4301, Argonne National Laboratory DESCRIPTION:With the advent of C-based programming environments like CUDA and OpenCL, the recent years saw a great deal of interest in developing high-performance general-purpose applications for GPUs. However, today’s performance tuning practice by GPU programmers still demands manual measurement and paper-and-pencil analysis. In this talk, we will present our work on GPU performance modeling. The model we developed is able to help GPU architects and programmers to identify performance bottlenecks, suggest them solutions, and quantitatively predict/evaluate the effectiveness of the proposed solutions. By dynamic program simulation based on the native GPU instruction set, our model is able to handle data-dependent applications and predict performance with a 5–15% error. We will also demonstrate our performance tool for a variety of numerical algorithms including dense/sparse matrix multiply and tridiagonal system solvers. Finally,we will discuss our future work on building a performance modeling infrastructure, and the opportunities and challenges in model-based performance autotuning. SUMMARY:A Quantitative Performance Model for the GPU UID:1357 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110802T140000 DTEND;TZID=America/Chicago:20110802T170000 DTSTAMP:20130525T020110 LOCATION:Building 240, Conference Center 1416, Argonne National Laboratory DESCRIPTION:Schedule\n2:00- 2:15 - Hayes Stripling\n Uncertainty Quantification for Nuclide Depletion Calculation\n2:15- 2:30 - Drew Wicke\n Identifying Active Variables to Improve the Performance of Operator Overloading Automatic Differentiation\n2:30- 2:45 - Karim Ahmed\n Phase-Field Modeling of Void Kinetics in Irradiated Metals\n2:45- 3:00 - Break\n3:00- 3:15 - Jayash Koshal\n Heuristics for Mixed Integer Non-Linear Programs (MINLPs)\n3:15- 3:30 - Alex Stovall\n Constructing the building blocks of a loop less code generator.\n3:30- 3:45 - Edward Nash\n The Cost of Using Loop Control\n3:45- 4:00 - Break\n4:00- 4:15 - Alexandru Cioaca\n Improving the accuracy of wind energy prediction\n4:15- 4:30 - Brett Robbins\n Parallel Newton\'s Method for Dynamic Simulation of Electrical Networks\n4:30- 4:45 - Zhu Wang\n Approximating the output of large models with uncertainty using model reduction and Kriging\n4:45- 5:00 - Yongjia Song\n Solving sample average approximation for stochastic programs\n\nAbstracts\n\nHayes Stripling\nTitle: Uncertainty Quantification for Nuclide Depletion Calculation\nAbstract:\nThis summer I worked to support CESAR, the recently-funded exascale co-design center focused on the simulation of nuclear reactors. One major task of the center is to develop efficient software/algorithms for quantification of error and uncertainty at scale, which is especially challenging in the high-dimensional, multi-scale calculations required for reactor simulations. We first developed an asymptotic model for the global time-discretization error in neutron/nuclide depletion calculations. We then developed and tested an adjoint framework for multi-physics systems governed by differential-algebraic equations. This framework was designed to be general and abstract in order to scale and be flexible in HPC environments.\n\nDrew Wicke\nTitle: Identifying Active Variables to Improve the Performance of Operator Overloading Automatic Differentiation\nAbstract:\nAutomatic Differentiation (AD) is a means of computing the derivative of a function within a computer program. AD can be performed by using the operator overloading approach which utilizes features of the programming language to alter the meaning of mathematical operators to compute the derivative. Operator overloading as a means of performing AD allows for maintainable code; however, speed of computation is sacrificed. One method to increase the speed is to perform derivative computation for only active variables. Active variables depend on the value of an input variable and are used in the computation of an output variable and therefore are necessary for the calculation of the derivative. All other variables are considered inactive and are not needed for the derivative calculation. Activity analysis is a technique used to identify active variables in the input source code. The goal of this research was to use activity analysis to improve the performance of the calculation of derivatives using Sacado -- an implementation in C++ of the operator overloading method of AD. The tool created to accomplish this combines the activity analysis of the source code analysis toolkit, OpenAnalysis, with the source-to-source transformation tool ROSE. The tool was tested to ensure proper identification of all active variables by creating test cases that used the C and C++ constructs.\n\nKarim Ahmed\nTitle: Phase-Field Modeling of Void Kinetics in Irradiated Metals\nAbstract:\nA phase field model was developed to simulate the void kinetics and interactions in irradiated metals. The model captures the growth and shrink of the void due to supersaturated/ sub-saturated vacancy content in the solid matrix because of radiation. The model is spatially resolved meaning that it takes into account the inhomogeneity of the domain which makes it more accurate than rate theory models since it can take into account the gradients in external forces such as temperature and stress.\n\nJayash Koshal\nTitle: Heuristics for Mixed Integer Non-Linear Programs (MINLPs)\nAbstract:\nIn this talk, we describe heuristics for global optimization and MINLP. These problems may have nonlinear nonconvex functions in objective and constraints. Branch-and-bound algorithm is usually used to solve these problems. While it finds bounds on optimal value quickly, finding a good feasible solution takes longer. A good feasible point from a heuristic helps in speeding up branch-and-bound algorithm by pruning the search tree. It can also be used as a \"good enough\" solution if the algorithm is terminated before completion.\n\nWe use a Multistart heuristic for global optimization of non-convex continuous NLPs. We iteratively call an NLP solver from different starting points selected on the basis of previous solutions and their objective values. A Diving heuristic works by changing bounds of some of the fractional variable and resolving the NLP relaxation. We implement different methods for selecting variables whose bounds are to be changed and backtracking if infeasibility is detected. The Feasibility Pump heuristic generates sequence of points by alternatively solving NLP or LP relaxation and rounding the fractional solution. Numerical results are presented along with future directions.\n\nAlex Stovall\nTitle: Constructing the building blocks of a loop less code generator.\nAbstract:\nSmall, dense, rectangular matrix-matrix multiplication is used extensively in the computation kernels of DOE simulation applications, such as MADNESS and NEK5000. A loop-less code generator has been developed that can be used to produce instruction sets that can compute these kernels at peak performance for a targeted computer platform. This research will show how to design and code macros that utilizing the fewest instructions, maximizing the use of computing resourses at the processor level (instruction scheduling, cache memory, and xmm registers).\n\nThe building blocks can be utilized to construct the loop less code generator so that no more tedious and time consuming assembly programming to deal with, and peak performance code can be achieved without much effort.\n\nEdward Nash\nTitle: The Cost of Using Loop Control\nAbstract:\nIn both high-level and low-level programming languages, the loop instruction is used to group instructions together and execute them continually. However, the use of loops incurs efficiency costs. This study reviews current research results on reducing the efficiency costs of a program by using loop-less codes. Comparisons are made between the performance of the similar code with and without loop control in computing small, dense rectangular matrix-matrix multiplication operations in targeted platforms (AMD 64 processor based systems). Information such as instruction counts, stalls, cycles, cache access, and conditional branching will be used as metrics to compare the efficiency of code without loops to that of code with loops. Based upon our research, we have observed that the compilers generate longer instruction code then that of loop-less code. Therefore, it increases the number of stalls which reduces the efficiency. We believe that this phenomenon is true to any well-defined computational tasks.\n\nAlexandru Cioaca\nTitle: Improving the accuracy of wind energy prediction\nAbstract:\nIntegrating wind farms in an intelligent electrical grid can be achieved by considering weather forecasts in the unit commitment and energy dispatch problems. These forecasts are performed through numerical models that need to be operated from accurate initial conditions, at high spatial resolutions. We are currently working on an adjoint sensitivity method that can indicate which areas to target for observation and forecast.\n\nBrett Robbins\nTitle: Parallel Newton\'s Method for Dynamic Simulation of Electrical Networks\nAbstract:\nTraditional approaches to perform dynamic simulations of an electrical network use an implicit method for the numerical integration and Newton\'s method to solve the system of nonlinear equations sequentially in time. Although effective, this strategy does not take advantage of computer systems with a large number of cores that are capable of parallel processing techniques to reduce the number of iterations required for the simulation. A proposed strategy that computes every time step simultaneously will be presented and demonstrated with a numerical example.\n\nZhu Wang\nTitle: Approximating the output of large models with uncertainty using model reduction and Kriging\nAbstract:\nIn this talk, we introduce model reduction technique in uncertainty quantification for large simulation model. Computationally cheap reduced-order model (ROM) is employed to replace expensive sampling of the full model. However, a limitation of ROM is that it is empirical. Since the quality of the ROM cannot be satisfactorily described a priori, we use Gaussian-processes based Kriging to correct the outputs of the reduced-order model. The method is supported by numerical experiments.\n\nYongjia Song\nTitle: Solving sample average approximation for stochastic programs\nAbstract:\nWe will talk about solution methods on sample average approximation(SAA) for stochastic programs. Two types of stochastic programs are considered in the talk: two-stage problem with recourse, and chance constrained problem. Decomposition algorithms are used to solve these problems within the SAA scheme.\n SUMMARY:Student Argonne Summer Symposium - SASSy 2011 UID:1363 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110819T133000 DTEND;TZID=America/Chicago:20110819T143000 DTSTAMP:20130525T020110 LOCATION:Building 240 / Conference Room 4301, Argonne National Laboratory DESCRIPTION:Progresses in two computational projects, computation of absolute binding free energy and implementation/application of Drude polarizable force field in NAMD will be reported. The heart of computation strategy for ligand binding free energy is a novel Distributed Replica (parallel/parallel) mode, which resolves the ‘embarrassingly parallel’ problems in free energy computation. Parallel performance on Blue Gene/P and CRAY XT5 with parallel/parallel mode will be shown. This strategy also facilitates a replica exchange scheme by point-to-point communications, forming a distributed Hamiltonian exchange molecular dynamics method. The novel methodology will be demonstrated with 3 difficult cases of ligand binding calculations, although the overall methodology is applicable to many other quantitative biological simulations. For the Drude oscillator model, the implementation strategy follows the seminal hybrid force/domain decomposition of NAMD and therefore realizes unprecedented large-scale molecular dynamics simulations with all-atom polarizable force field. Implementation details will be provided and parallel performance of general Drude model on Blue Gene/P will be shown. Application of NAMD of Drude model will be demonstrated with a novel large-scale simulation of salt solution at physiological concentration.\n\nClick on the link to add this event to your calendar. SUMMARY:Methodological Developments for Molecular Dynamics Simulations on Massively Distributed Computing Platforms UID:1365 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110809T133000 DTEND;TZID=America/Chicago:20110809T143000 DTSTAMP:20130525T020110 LOCATION:Building 240 / Conference Room 4301, Argonne National Laboratory DESCRIPTION:Progresses in two computational projects, computation of absolute binding free energy and implementation/application of Drude polarizable force field in NAMD will be reported. The heart of computation strategy for ligand binding free energy is a novel Distributed Replica (parallel/parallel) mode, which resolves the ‘embarrassingly parallel’ problems in free energy computation. Parallel performance on Blue Gene/P and CRAY XT5 with parallel/parallel mode will be shown. This strategy also facilitates a replica exchange scheme by point-to-point communications, forming a distributed Hamiltonian exchange molecular dynamics method. The novel methodology will be demonstrated with 3 difficult cases of ligand binding calculations, although the overall methodology is applicable to many other quantitative biological simulations. For the Drude oscillator model, the implementation strategy follows the seminal hybrid force/domain decomposition of NAMD and therefore realizes unprecedented large-scale molecular dynamics simulations with all-atom polarizable force field. Implementation details will be provided and parallel performance of general Drude model on Blue Gene/P will be shown. Application of NAMD of Drude model will be demonstrated with a novel large-scale simulation of salt solution at physiological concentration. SUMMARY:Methodological Developments for Molecular Dynamics Simulations on Massively Distributed Computing Platforms UID:1367 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110808T130000 DTEND;TZID=America/Chicago:20110808T140000 DTSTAMP:20130525T020110 LOCATION:Building 240 - Room 4301, Argonne National Laboratory DESCRIPTION:Van der Waals (vdW) interactions are ubiquitous in nature, playing a major role in defining the structure, stability, and function for a wide variety of molecules and materials.\nThus, the accurate description of vdW interactions is essential for improving our understanding of many biological, chemical, and (hard and soft) condensed matter.\nHere I present our recent developments of an efficient method to determine the full long-range many-body vdW energy for molecules and solids. This is achieved by combining the TS-vdW method (Phys. Rev. Lett. 102, 073005 (2009)) with the self-consistent screening equation of classical electrodynamics. This leads to a seamless description of anisotropy, polarization, and depolarization for the polarizability tensor of molecules and solids. The screened long-range many-body van der Waals (vdW) energy is obtained from the numerically exact solution of the Schroedinger equation for a system of coupled oscillators. We show that the screening and the many-body vdW energy play a significant role even for rather small molecules, becoming crucial for an accurate treatment of conformational energies in bio-molecules, and binding of molecular crystals. The computational cost of the developed theory is negligible compared to the underlying electronic structure calculation, enabling calculations for thousands of atoms. SUMMARY:Accurate and Efficient Method for Many-Body van der Waals Interactions UID:1369 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110811T093000 DTEND;TZID=America/Chicago:20110811T103000 DTSTAMP:20130525T020110 LOCATION:Building 240 - Room 4301, Argonne National Laboratory DESCRIPTION:Low dimensional mesh and torus topologies allow high performance networks to scale to hundreds of thousands of nodes. Many modern interconnects employ such topologies in Petascale-era supercomputers (BG/P, BG/Q, Cray XE6, K-computer). These torus networks operate at much higher efficiency when applications communicate in a structured and well-mapped manner. In this seminar, I will present\nnew topology-aware algorithms for dense matrix and tensor computations that run at much higher efficiency on torus networks.\n\nI will introduce a novel class of \'2.5D algorithms\' for distributed dense linear algebra computations. These algorithms have an optimal theoretical communication\ncomplexity and have a parameterized 3D logical topology. I will also present actual implementations of 2.5D matrix multiplication and LU factorization. These algorithms achieve speed-ups of up to ~3X on 16,384 nodes of BG/P and reduce communication time by up to ~90% over classical algorithms.\n\nThe topology-aware mapping ideas used in matrix multiplication generalize to higher-dimensional tensor contractions. Further, a cyclic decomposition allows for seamless mapping of tensors in symmetric packed layouts. Using all of the above ideas as building blocks, I will detail a library implementation that performs automatic topology aware mapping for contractions of tensors with arbitrary dimension and symmetry to arbitrary dimensional k-ary n-cube networks.\" SUMMARY:Communication-optimal and Topology-aware Algorithms For Dense Matrix and Tensor Computations UID:1371 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110816T160000 DTEND;TZID=America/Chicago:20110816T170000 DTSTAMP:20130525T020110 LOCATION:Building 240 - Room 1404, Argonne National Laboratory DESCRIPTION:The fragment molecular orbital (FMO) method is a well established method for massively parallel quantum-mechanical calculations of large systems (http://staff.aist.go.jp/d.g.fedorov/). By dividing the system into fragments and performing their ab initio calculations, one achieves a much reduced scaling in terms of size and very efficient parallelisation [1]. FMO as implemented in GAMESS is targeted for petascale computing in Japan (the K computer) and USA (Mira). The applications of FMO are mainly focused on biochemical systems and molecular clusters (explicit solvation), and some work is also done on inorganic systems such as zeolites and nanowires. An overview of the method development and applications will be presented, and\nthe challenges in the parallelisation strategies will be outlined. SUMMARY:The Fragment Molecular Orbital Method as a Massively Parallel Method for Chemical Applications UID:1373 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110809T103000 DTEND;TZID=America/Chicago:20110809T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, TCS Conference Center, 4301, Argonne National Laboratory DESCRIPTION:Graphics Processing Units (GPUs) are becoming more common in High Performance Computing (HPC) systems due to their unprecedented raw performances in a variety of domains, including molecular modeling, biological sequence analysis, linear solvers, etc. Moreover, three out of the current top five most powerful supercomputers in the world have GPUs as their fundamental compute accelerators. However, the efficiency of these heterogeneous systems has remained below par, mainly because of explicit and slow data transfers over the PCIe between their disjoint memory subsystems. The programmability of these systems has also remained a challenging goal, because it is hard and error-prone to efficiently manage the different memories within each node. As part of this internship, we extend one of the widely used Message Passing Interface (MPI) implementations, MPICH2, to natively support data transfers between any two memory regions, either host or GPU device, over the network. We efficiently perform internal pipelined data transfers between the GPU, the host and the network, so that the PCIe latency between the GPU and the host is hidden for improved performance. The GPU programming interface can be either CUDA or OpenCL, and our design is flexible to extend support to more programming models and device types if required. Our experiments over an infiniband network indicate that our pipelined design can achieve up to 22% improvement in two-sided (GPU-GPU) communication over the default blocking data communication at the user level, along with improved programmability. In the future, we will make our implementation available to the HPC community by releasing it with a future version of MPICH2.\n\nBio: Ashwin Aji is a PhD candidate (2nd year) from the Dept. of Computer Science at Virginia Tech. He is advised by Prof. Wu Feng, whose projects include the Green500 list, mpiBLAST, etc to name a few. Dr. Feng also guided him towards his Masters degree at Virginia Tech in 2008, for which he received the Outstanding M.S. Student award. Ashwin\'s research interests include high performance and parallel computing, emerging parallel architectures and performance modeling. He hopes to graduate with the doctoral degree by May 2013.\nWeb links:\nAshwin Aji: http://people.cs.vt.edu/~aaji\nDr. Wu Feng: http://people.cs.vt.edu/~feng\n SUMMARY:Efficient Inter-node Communication in High Performance GPU Systems UID:1377 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110804T120000 DTEND;TZID=America/Chicago:20110804T000008 DTSTAMP:20130525T020110 LOCATION:Cafeteria, Private Dining Rooms A&B, Argonne National Laboratory DESCRIPTION:Dark energy can differ from the cosmological constant, the simplest model for the cosmic acceleration, either by evolving with time or by coupling to known particles. Coupled theories typically invoke a screening mechanism, such as the chameleon mechanism, to explain tight constraints on fifth forces and the evolution of fundamental constants locally. I will discuss the GammeV-CHASE experiment, conducted last year at Fermilab, which was designed to produce, trap, and detect photon-coupled chameleon dark energy. GammeV-CHASE has improved constraints on photon-chameleon couplings by orders of magnitude and has ruled out a range of models at the dark energy scale. SUMMARY:How dark is dark energy? A laboratory search for couplings between photons and chameleon dark energy UID:1379 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110804T103000 DTEND;TZID=America/Chicago:20110804T113000 DTSTAMP:20130525T020110 LOCATION:Building 240 Conference Room 4301, Argonne National Laboratory DESCRIPTION:Nonequilibrium umbrella sampling (NEUS) addresses the\nchallenging problem of rare events in biology and chemistry, such as the transition between folded and unfolder states of a protein. This is done through the application of an umbrella sampling technique for non-equilibrium systems that divides the phase space into regions based on their distance from equilibrium and traces the dynamic trajectory of the system in these regions with enhanced sampling. \nCompared with traditional sampling methods, NEUS dramatically reduces the first passage time to observe extremely rare unfolding events by many orders of magnitude. The parallel NEUS algorithm must maintain a\nlarge amount of data about the trajectories in each phase space region and the boundary regions between them. The distribution of data, communication between processes, and load balance are critical to the performance of the algorithm. In this talk, I\'ll present our analysis, implementation, and performance results on the BG/P system.\nImproved data structures for one-sided access, the application of MPI one-sided communication and other optimizations to this application have yielded several-fold performance gains over the original application. The end result is a more portable, more robust and more scalable framework for performing enhanced sampling simulations in\nbiology and chemistry. SUMMARY:A One-Sided Approach to Data Management in a Nonequilibrium Umbrella Sampling Simulation UID:1381 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110811T140000 DTEND;TZID=America/Chicago:20110811T150000 DTSTAMP:20130525T020110 LOCATION:Building 221 Conference/Seminar Room 4301, Argonne National Laboratory DESCRIPTION:Multireference quantum chemical methods are crucial for the correct description of many chemical processes such as bond breaking, formation of biradicals, electronic excitation and nonadiabatic coupling of different electronic states. As prominent examples serve applications in the fields of photovoltaics, DNA photodynamics and graphene defects. The COLUMBUS program system is presented as an efficient and versatile tool for a wide range of applications within the multireference configuration interaction (MRCI) method. Accurate MRCI calculations require large amounts of computer time and usage of massively parallel program implementations are mandatory. In this talk the major strategies for parallelization of COLUMBUS are discussed and representative examples are presented. SUMMARY:Massively Parallel Calculations in Multireference Quantum Chemistry UID:1383 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110810T103000 DTEND;TZID=America/Chicago:20110810T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, TCS Conference Center, 4301, Argonne National Laboratory DESCRIPTION:GPU accelerators are gaining popularity in a high performance computing systems. Though they can bring significant gains in performance and power efficiency, GPUs introduce a distinct \"device\" memory that must be managed by the programmer. When the programmer wishes to communicate data to or from the GPU using MPI, they must currently perform explicit movement of data between host and device memory, leading to inefficient utilization of resources and additional data movement operations.\n\nIn this project, we address the efficiency of communicating data from GPU to GPU within the same node using MPI. We address this problem through two techniques: (1) using host-side shared memory to eliminate extra memory copies, and (2) enabling GPU context sharing across processes for direct device-to-device communication. Results indicate that the the use of host-side shared memory can yield up to 2.5x speedup for large messages. GPU context sharing is not currently supported by CUDA, however we have made significant progress toward enabling this highly efficient technique. I will report on the performance potential of context sharing as well as the technical hurdles we have overcome in this ongoing effort.\n\nIn addition, if time permits, I will discuss my ongoing work at NCSU: building a software distributed shared memory system for CPU-GPU heterogeneous systems. In this project, we seek to address the productivity and performance challenges of maintaining distinct CPU and GPU memories through the use of a software-based shared memory model.\n\nBIO:\n\nFeng Ji is starting his fourth year as a PhD candidate in the department of computer science of North Carolina State University, under the guidance of Dr. Xiaosong Ma. He is interested in systems research for parallel architectures, especially GPU-enabled heterogeneous systems. In the past, he got his bachelor and master degrees from Zhejiang University, Hangzhou, China. With an expected graduation in 2013, he is hoping to find a systems research position in the future. SUMMARY:Optimizing Intranode, GPU-to-GPU Communication in MPI UID:1385 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110811T103000 DTEND;TZID=America/Chicago:20110811T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, TCS Conference Center, 4301, Argonne National Laboratory DESCRIPTION:Recent advances in computer architecture have lead to significant changes in the degree of intra-node parallelism in supercomputing systems. Rapid increases in the number of cores per chip and multi-socket nodes have introduced significant non-uniformity into the latency of memory accesses and on-node communication operations. Hierarchical, node-aware collective communication have been incorporated into many MPI implementations, however these algorithms assume uniform intra-node communication latencies between all cores. In this work, we develop a NUMA-aware performance model for intra-node communication and use it to optimize the performance of MPI collective communication operations. We evaluate the performance of these NUMA-aware collective communication operations on multi-socket Intel and AMD node architectures and demonstrate that the choice of communication algorithm and topology can yield significant performance gains.\n\nBio: Li Rao is master student from Institute of Software, Chinese Academy of Sciences(ISCAS), he is supposed to graduate in July,2012. Li Rao\'s mentor is Dr.Yunquan Zhang, who is the main organizer of China TOP100 List of High Performance Computer. Li Rao will continue this work in Argonne as his Master Thesis. SUMMARY:Characterizing the Implications of Intra-­node Topologies on MPI Collective Communication UID:1387 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110819T103000 DTEND;TZID=America/Chicago:20110819T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, TCS Conference Center, 1405 & 1406, Argonne National Laboratory DESCRIPTION:Abstract: Collective communication is a key aspect of large scale parallel application performance, and is becoming increasingly important as the number of cores on a supercomputer is approaching a million. Although optimization of collectives is a well studied topic and implementations of collectives in major MPI implementations are mature, the recent trend toward supercomputers with accelerator devices presents a less familiar issue of collectives on data residing in memories of accelerator devices. This work examines the question of how to optimize MPI collectives on data distributed across system and accelerator memories. The work is done in the context of the MPICH implementation of MPI running on a cluster accelerated by CUDA GPUs. Our implementation of collectives integrates efficient data movement between system and accelerator memories and schedules MPI inter-process communication with awareness of processes which need to copy data from the accelerator device before participating in the collective. \n\nBio: Lukasz is a Ph.D. student in computer science at the University of Illinois at Urbana-Champaign, where he also received his B.S. and M.S. degrees in computer science. His areas of interest include optimization of communication patterns on supercomputers, tools and abstractions for parallel programming, general purpose graphics processing units, and large scale parallel applications. His adviser is Laxmikant Kale. SUMMARY:Optimization of MPI Collectives on GPU Clusters UID:1389 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110817T103000 DTEND;TZID=America/Chicago:20110817T113000 DTSTAMP:20130525T020110 LOCATION:Building 240 / Conference Room 4301, Argonne National Laboratory DESCRIPTION:Presented is computational astrophysics and cosmological research featuring various extremes. Brief mention is given to a numerical complication in the context of ultra-relativistic, hydrodynamic simulations of winds from dead stars involving matter densities on the order of 10^-19 times smaller than that of the diffuse Galactic medium. Results are further presented for a two-pronged application of the Argonne Blue Gene/P (BG/P) supercomputer to the study of dark energy covering spatial scales from a single star to the entire Universe. The stellar scale, and relevance to matter densities on the order of 10^6 times that of the Sun, is represented by radiative transfer calculations for thermonuclear supernova explosions. The largest spatial scale known to science is represented by simulations of the cosmic structure of the Universe. SEDONA code radiative transfer calculations are shown to scale well on the BG/P for a minimally parallel case. BG/P scaling of the FLASH code cosmology module is shown to be complex, and pursuit of BG/P simulation runs incorporating 1024^3 particles are demonstrated to be a significant challenge. Key improvements enabled by the 2012 arrival of the next-generation Argonne Blue Gene/Q supercomputer are discussed.\n\nClick on the link to add this event to your calendar. SUMMARY:Computational Astrophysics and Cosmology: A Tale of Vastly Different Scales UID:1391 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110810T093000 DTEND;TZID=America/Chicago:20110810T103000 DTSTAMP:20130525T020110 LOCATION:Building 240 Conference Room 4301, Argonne National Laboratory DESCRIPTION:Modeling the dynamics of rare event systems poses a significant challenge due to the disparity of timescales involved. Atomic vibrations occur on the femtosecond timescale while a chemical event of interest might occur on the millisecond timescale or longer. Thus to study these rare event systems approaches other than direct molecular dynamics must be developed.\n\nI will talk about one such long timescale simulation method: Adaptive kinetic Monte Carlo (AKMC). AKMC is a kinetic Monte Carlo simulation in which the event table is determined during the simulation. Transition state theory is used to calculate reaction rates and kinetic Monte Carlo is used to transition between the states. I will also discuss porting our software, Eon, to run on BGP. SUMMARY:Long Timescale Dynamics with Adaptive Kinetic Monte Carlo UID:1393 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110819T140000 DTEND;TZID=America/Chicago:20110819T150000 DTSTAMP:20130525T020110 LOCATION:Building 240, TCS Conference Center, 1406 & 1407, Argonne National Laboratory DESCRIPTION:Graphics Processing Units (GPUs) are becoming increasingly used at the high performance computing level, enabling the acceleration of large-scale scientific and engineering simulations. Currently, there is little meaningful integration between communication libraries such as MPI and the GPU, with the prevailing model being CPU-driven execution flow and data management of the GPU for use in MPI communication routines. This model is unlikely to change as long as GPUs use a discrete memory space, which introduces numerous challenges to allow efficient communication between the different memory spaces. One particular use-case is the communication of non-contiguous data, such as column vectors, 3D array slices, etc. As a step toward a more efficient integration of MPI with discrete GPU hardware, we discuss and provide an implementation of MPI datatype processing on the GPU. We provide a parallel extension to the current MPICH \"dataloops\" tree-based implementation, which allows for arbitrary point-wise packing and retrieval. We evaluate the data movement under a number of MPI datatypes, losing little next to \"close to metal\" CUDA API functions and also demonstrate the efficiency of 3D array slice packing for halo exchange, accelerating Y-Z face transfer to the CPU by up to an order of magnitude for larger messages. Finally, we examine the effect of various types of resource contention on the GPU, identifying under which scenarios contention causes performance degradation of the packing operation or resident computation.\n\nBio:\nJohn Jenkins is a second-year PhD student at North Carolina State University under Dr. Nagiza Samatova. He received his Bachelor\'s degree in Computer Science at Lafayette College in 2010. He is currently studying in the realm of HPC and large-scale data analytics, but hasn\'t yet finalized a main thrust of research.\n SUMMARY:Accelerating Movement of Non-contiguous Data on Hybrid MPI+GPU Environments UID:1397 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110812T103000 DTEND;TZID=America/Chicago:20110812T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, TCS Conference Center, 4301, Argonne National Laboratory DESCRIPTION:Abstract: Leadership-class supercomputers, such as the ALCF Intrepid IBM Blue Gene/P system, are a challenge to design and understand. These systems consist of many design points that influence application performance. While modeling and simulating these systems is useful for identifying successful system designs, these activities are a significant challenge because it is difficult to accurately and efficiently model the interactions between system components. Parallel discrete-event simulation (PDES) tools provide a convenient way to accurately model complex interactions of these systems components with sufficient fidelity, efficiency, and fast turnaround time. In this presentation, we present a PDES model of the Intrepid PVFS storage system using the Rensselaer Optimistic Simulation System (ROSS). We validated this model using data collected on the Intrepid storage system up to 128K application cores. We show that our initial simulation results closely track observed results for the Intrepid storage system for a variety of synthetic I/O workloads. SUMMARY:Modeling and Simulating a Leadership-Class Storage System UID:1399 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110818T103000 DTEND;TZID=America/Chicago:20110818T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, TCS Conference Center, 1404 & 1405, Argonne National Laboratory DESCRIPTION:The microprocessor chip technology revolution has moved rapidly to large-scale on-chip (hybrid) multiprocessing and parallel computing at an unprecedented scale and incorporation of different architecture types in the same system. Growing core-count and heterogeneity will produce increasingly fragmented resource pools in large-scale systems. Achieving efficiency, reliability and high performance will increasingly require flexible, dynamic, intelligent runtime resource management and corresponding activity scheduling for applications not present in existing software stacks. The needs far exceed what has existed in traditional runtimes support from current system vendors – and there is a clear gap in the industry-strength system software stack.\n\nIn this talk, we present our work toward a portable dynamic adaptive runtime model and software system. We demonstrate that the role of runtime systems, in contrast to the traditional OS, is to efficiently implement and support the target program execution and abstract machine models. In particular, our work is focused on Asynchronous, fine-grain, event-driven execution models with it deep root in dataflow-like models with major extensions. The work at Delaware, jointly performed between University Delaware and its spinoff (ET. International Inc., found in 2000), are based on practical experience of developing and deployment of end-to-end system software solutions (from bare-metal OS to parallel programming support) on four types of many-core chips: the IBM Cyclops chips/systems, the IBM CELL chip, the Intel SCC Chip and Adaptiva 16-core embedded low power (2W) many-core chip. The findings learned will be summarized through the initial experience with ETI SWARM runtime software under the Intel-led DARPA/UHPC extreme-scale architecture/system.\n\nProfessor Guang Gao is a ACM Fellow and IEEE Fellow, Endowed Distinguished Professor, Dept. of Electrical and Computer Engineering, University of Delaware SUMMARY:Toward A Portable Dynamic Runtime Model and Software Environment UID:1403 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110823T103000 DTEND;TZID=America/Chicago:20110823T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, TCS Conference Center, 4301, Argonne National Laboratory DESCRIPTION:Since Claude E. Shannon proposed information theory in 1940s, it has been successfully applied to different applications in communication since it provides the theoretic foundation to model the transmission of information. Besides communication, its applications include image processing, computer graphics and visualization, and currently we are working on demonstrating its utilities for in situ data analysis. In this talk, I will overview the entire roadmap of our project of integrating information theory with in situ analysis and visualization of scientific data. I will summarize my progress in this summer, which is mainly focusing on creating a software layer that interfaces scientific scientific simulation code to compute different information theoretic measurements. I will review related mathematical background, the design rationale, the challenges to create a general program interface for different scientific simulations, and the current integration with the NEK5000 and FLASH codes as examples. In the future, we plan to use information-theoretic metrics to identify salient data blocks and time steps, reduce data, and provide level-of-detail selection.\n\nTeng-Yok Lee is a PhD candidate in the Dept. of Computer Science & Engineering at The Ohio State University. His advisor is Prof. Han-Wei Shen. Teng-Yok\'s research areas cover computer graphic, scientific and information visualization, and GPGPU. He obtained his bachelor and master degrees in Computer Science & Information Engineering from the National Chiao-Tung Univertity, sin-Chu, Taiwan. He expects to receive his Ph.D. by December 2011.\n\nHis personal webpage: http://www.cse.ohio-state.edu/~leeten\n \n SUMMARY:Application of Information Theory to Scientific Data UID:1405 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110824T103000 DTEND;TZID=America/Chicago:20110824T103000 DTSTAMP:20130525T020110 LOCATION:Building 240, TCS Conference Center, 4301, Argonne National Laboratory DESCRIPTION:Information theory is widely studied and applied in various fields of computer science. Recently, it has been recognized for its potential to drive analysis and visualization of large scale scientific data. The theoretical results have motivated us to offer scientists the computing support necessary to leverage information theory for in situ analysis as well as for post-processing. Currently, we are in the process of developing a parallel C/C++ library called Information Theory Library (ITL) which provides kernels for computing various useful information theory metrics such as Shannon\'s entropy, conditional entropy and mutual information in a distributed environment. My talk will primarily focus on the key components of this library and its parallelization. In addition, our ongoing endeavor towards entropy-driven data organization for large vector and scalar field data will also be presented. \n\nBio:\nAbon is a Ph.D. student in the Computer Science & Engineering department at The Ohio State University since 2006. His research interests include flow visualization, visual analytics of scientific data at large scale. He is also interested in certain aspects of information visualization and geovisualization. He is advised by Dr.Han-Wei Shen. SUMMARY:Information Theory Library: A Tool for Information-driven Management, Analysis and Visualization of Scientific Data UID:1407 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110819T103000 DTEND;TZID=America/Chicago:20110819T113000 DTSTAMP:20130525T020110 LOCATION:Bldg 240 Conference Rm: 1406 & 1407, ANL DESCRIPTION:Data-Enabled life Scinces Alliance (DELSA) - Introduction and Invitation \n\n4th paradigm of science and DELSA\n•The research necessity of the life sciences community to work across diverse domains and with computer, cyberinfrastructure, and data experts to leverage opportunities in DES. \n•Scientific progress and accelerated rate of life sciences result in a pressing need for reproducibility.\n•A perceived gap between the needs of data-enabled life sciences and current funding initiatives.•A specific need to integrate data-enabled sciences with major international and national initiatives.\nDELSA includes: Biological sciences, ecology, environmental sciences, evolution, genomics, computer sciences, cyberinfrastructure, management, health sciences, policies SUMMARY:Data-Enabled life Scinces Alliance (DELSA) - Introduction and Invitation UID:1409 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110819T103000 DTEND;TZID=America/Chicago:20110819T113000 DTSTAMP:20130525T020110 LOCATION:Bldg 240 Conference Rm: 1406 & 1407, ANL DESCRIPTION:Data-Enabled life Scinces Alliance (DELSA) - Introduction and Invitation \n\n4th paradigm of science and DELSA\n•The research necessity of the life sciences community to work across diverse domains and with computer, cyberinfrastructure, and data experts to leverage opportunities in DES. \n•Scientific progress and accelerated rate of life sciences result in a pressing need for reproducibility.\n•A perceived gap between the needs of data-enabled life sciences and current funding initiatives.•A specific need to integrate data-enabled sciences with major international and national initiatives.\nDELSA includes: Biological sciences, ecology, environmental sciences, evolution, genomics, computer sciences, cyberinfrastructure, management, health sciences, policies SUMMARY:Data-Enabled life Scinces Alliance (DELSA) - Introduction and Invitation UID:1411 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110817T140000 DTEND;TZID=America/Chicago:20110817T150000 DTSTAMP:20130525T020110 LOCATION:Building 240 Conference Room 4301, Argonne National Laboratory DESCRIPTION:Differential solvation of the electronic states of a chromophore in a solvent, called solvatochromism, can be used as a measure of solute-solvent interactions. Since electronic states of organic chromophores in non-polar solvents and in a protein environment are typically well localized within the solute, it allows one to separate a system and use hybrid QM/MM (quantum mechanics / molecular mechanics) approach. We developed a series of QM/MM methods for description of the electronic excited states in the environment. The equation-of-motion coupled cluster with single and double excitations (EOM-CCSD) configuration interaction singles with perturbative doubles (CIS(D)) methods and time-dependent density functional theory (TD-DFT) are used to describe the QM region. The effective fragment potential (EFP) method describes the MM part. The EFP method overcomes the most significant limitations of QM/MM by replacing empirical MM interactions and QM-MM coupling by parameter-free first principles based ones, while retaining the computational efficiency of QM/MM. We will discuss accuracy of EFP and details of implementation of the QM/EFP schemes, as well as applications to solvatochromic shifts of organic chromophores in various solvents. SUMMARY:Effective Fragment Potential method for electronic excited states: Theory, applications, and benchmarks UID:1413 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110824T100000 DTEND;TZID=America/Chicago:20110824T110000 DTSTAMP:20130525T020110 LOCATION:Building 212/Conference room A157, Argonne National Laboratory DESCRIPTION:Cluster ion beam processes which employ ions comprised of a few hundred to several thousand atoms are being developed into a new field of ion beam technology. The processes are characterized by low energy surface interaction effects, lateral sputtering phenomena and high-rate chemical reaction effects. At the seminar, the current status of compact GCIB equipment development, studies of the fundamental cluster ion beam characteristics as they apply to nanoscale processing and present industrial applications will be reviewed. As new prospective applications, techniques are now being developed to employ cluster ions in surface analysis tools such as XPS and SIMS and to modify surfaces of bio-materials. Results related to these new projects will also be reviewed. SUMMARY:Cluster Ion Beam Technology: Review of Current and Prospective Applications UID:1415 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110831T133000 DTEND;TZID=America/Chicago:20110831T143000 DTSTAMP:20130525T020110 LOCATION:Building 240 Seminar Room 4301, Argonne National Laboratory DESCRIPTION:The existing simulation tools for studying the different electrical power system dynamics are specifically tailored to a particular range of time scale and are divided into two groups: Transient Stability Simulators (TS) and Electromagnetic Transients Simulators (EMT). A Transient Stability simulator, running at large time steps, is used for studying relatively slower dynamics e.g. electromechanical interactions among generators and can be used for simulating large-scale power systems. In contrast, an electromagnetic transient simulator models the same components in finer detail and uses a smaller time step for studying fast dynamics e.g. electromagnetic interactions among power electronics devices. Simulating large-scale power systems with an electromagnetic transient simulator is computationally inefficient due to the small time step size involved.\n \nBy modeling the bulk of the large-scale power system in a transient stability simulator and a small portion of the system in an electromagnetic transient simulator, the fast dynamics of the smaller area could be studied in detail, while providing a global picture of the slower dynamics for the rest of power system along with not sacrificing computational efficiency.\n \nThis talk presents a novel implicitly coupled solution approach for this combined transient stability and electromagnetic transient simulation approach. To combine the two sets of equations with their different time steps, and ensure that the TS and EMT solutions are consistent, the equations for TS and coupled-in-time EMT equations are solved simultaneously. While computing a single time step of the TS equations, a simultaneous calculation of several time steps of the EMT equations is proposed.\n \nAlong with the implicitly coupled solution approach, this research work also presents an improvement over existing TS simulator by modeling all three phases instead of using a positive-sequence balanced representation.\n \nFurthermore a parallel implementation of the developed three-phase transient stability simulator and the implicitly coupled multiple time-scale dynamics simulator, using PETSc, is discussed. Results of experimentation with different reordering strategies, linear solution schemes, and preconditioners are presented. \n SUMMARY:Implicitly Coupled Multiple Time-Scale Electrical Power Grid Dynamics Simulation UID:1417 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110928T150000 DTEND;TZID=America/Chicago:20110928T160000 DTSTAMP:20130525T020110 LOCATION:Building 240, Conference Center 1404-1405, Argonne National Laboratory DESCRIPTION:We model the problem of responding to threats detected in a computer network by shutting down compromised systems (or users) as a Mixed Integer Problem following the work of Altunay, Leyffer, Linderoth and Xie (2011). The objective is to maximize the utility of the network as a function of the communication links that remain open given the tolerable level of threat. We consider computational techniques such as specialized valid inequalities for solving larger instances of the problem. We also consider alternative models of the problem and how the thread may spread in a probabilistic setting. SUMMARY:The Network Intrusion Response Problem and Solution Techniques UID:1419 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110907T103000 DTEND;TZID=America/Chicago:20110907T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, TCS Conference Center, 1404 & 1405, Argonne National Laboratory DESCRIPTION:In this work, I will present an approach to accelerate task parallel computations using GPUs in the context of the Global Arrays parallel programming model. Task parallelism is an efficient technique for expressing parallelism in irregular programs. We extend the Scioto task parallel programming library to efficiently offload task execution to GPU accelerators. The execution of Scioto tasks on the GPU requires movement of data through three layers: the global address space, host memory, and device memory. We propose an automated, pipeline-based approach for handling the movement of data through these memory spaces. Data transfer is made transparent to the user, providing opportunities to hide overheads through optimizations like pipelining. On-device caching and task sequencing are also leveraged to exploit data locality. We evaluate our work using a block-sparse tensor contraction kernel. Tensor contractions, which are generalized multidimensional matrix multiplication, are widely used in quantum chemistry. Experiments show that the proposed techniques yield significant performance gains by hiding the cost of data movement.\n\nBio: Humayun Arafat is a PhD student from the Dept. of Computer Science and Engineering at The Ohio State University. He is advised by Prof. P. Sadayappan. He has received Bachelor degree from Bangladesh University of Engineering and Technology. Arafat\'s research interest is on high performance and parallel computing. His previous work includes improving load balancing of Dynamic Nucleation Theory Monte Carlo(DNTMC), which is used in NWCHEM. SUMMARY:GPU Accelerated Task Parallelism in a Global Address Space Model UID:1423 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110908T103000 DTEND;TZID=America/Chicago:20110908T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, Conference Room, 4301, Argonne National Laboratory DESCRIPTION:Many parallel quantum Monte Carlo (QMC) applications rely on the use of an ensemble data structure that represents the quantum state of the simulation. Ensemble data is typically represented using a large spline interpolation table that becomes read-only after initial data has been loaded. Although only a small fraction of the table may be accessed by each thread or process at any given time, the accesses are random. Hence, current implementations of these methods typically use replicated copies of the entire interpolation table at each node of a parallel computer. This limits scalability since increasing the number of processors does not enable larger systems to be run.\n\nIn this talk, I will present an automated data management approach that enables existing QMC codes to be adapted with minimal changes to enhance the range of problem sizes that can be run. We utilize the Global Arrays partitioned global address space model to provide efficient distributed, shared storage and enhance the performance of our system through data layout optimization, caching, and replication.\n\nBio: Qingpeng Niu is starting his third year as a PhD student in the Department of Computer Science and Engineering of the Ohio State University, under the guidance of Dr. Sadayappan. He is interested in systems research for parallel computing. In the past, he got his bachelor and master degrees from Northeastern University, Shenyang, China.\n SUMMARY:A Global Address Space Approach to Automated Data Management for Parallel Quantum Monte Carlo Applications UID:1425 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110914T090000 DTEND;TZID=America/Chicago:20110914T110000 DTSTAMP:20130525T020110 LOCATION:TCS Conference Room 1416, Argonne National Laboratory DESCRIPTION:Note: If you are interested in attending the following presentation, please read the \'How to RSVP\' instructions at the end of this message. \n\nThe first Introduction of Beagle presentation at Argonne National Laboratory will take place on Wednesday, September 14, from 9 to 11 a.m. in room 1416 at the TCS Conference Center.\n\nThe topics covered will be the same as those in the first presentation at Searle at the University of Chicago in July:\n\n* overview of Beagle\'s Cray XE6 system architecture\n* basic access and navigation operations\n* using compilers and applications\n* appropriate use of local and network filesystems\n* programming in the Swift workflow language\n* submitting jobs and scheduler operations\n\nInterested attendees are invited to see Beagle itself at the TCS Datacenter following the conclusion of the presentation at 11 a.m. The tour will last about 15 minutes. As a headcount is needed, please explicitly indicate that you wish to attend when sending a RSVP.\n\nThose already with Argonne site access are welcome to invite along interested guests who are either U.S. citizens or currently have the appropriate foreign national paperwork on file with the Laboratory. Gate passes can be requested using inside.anl.gov. If you need assistance, please contact the ANL visitors center at 630-252-5755.\n\nTechnical assistance will be available on-site for routine requests such as password resets and HPC project access. If you wish to get access to Beagle and do not already have an account, please submit a request in advance at https://accounts.ci.uchicago.edu/.\n\nHow to RSVP:\n\nIf you would like to attend, send a message to \nbeagle-support@ci.uchicago.edu, preferably with the following subject line:\n\nIntroduction to Beagle presentation, Sep 14 @ TCS [Beagle ticket #16060]\n\nPlease include (a) the name(s) of attendee(s) and (b) if you wish to attend the system tour.\n\nDetails about the training can be found at the URLs:\n\n http://beagle.ci.uchicago.edu/trainings-and-events/\n http://www.ci.uchicago.edu/wiki/bin/view/Beagle/BeagleTraining\n SUMMARY:Introduction to Beagle UID:1427 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20111005T130000 DTEND;TZID=America/Chicago:20111005T140000 DTSTAMP:20130525T020110 LOCATION:TCS Building 240, Room 4301, Argonne National Laboratory DESCRIPTION:Seminar: Coenzyme-Q (Q; ubiquinone) is widely distributed in living organisms and is a component of the mitochondrial respiratory chain. It was thought that in eukaryotes it is present only in the respiratory chain; however, recent developments have shown that it is distributed in all cell membranes. It undergoes oxidation-reduction in cell membranes of lysosomes, golgi and functions as an antioxidant, either by direct reaction with free radicals or by regeneration of vitamins E and C. Prokaryotes such as Escherichia coli synthesize p-hydroxybenzoate (PHB), the precursor of Q, directly from the shikimate pathway via chorismate. Higher animals including humans lack the shikimate pathway and hence synthesize PHB from the essential amino acid tyrosine. However, the steps involved in the formation of PHB from tyrosine are not known. In this study, we show that Klebsiella oxytoca, can synthesize PHB from chorismate similar to E.coli as well as from tyrosine similar to humans. We propose an alternate pathway for the conversion of tyrosine to PHB and provide supporting evidence. SUMMARY:Coenzyme-Q biosynthesis: A novel pathway for p-hydroxybenzoate in klebsiella oxytoca UID:1437 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110922T103000 DTEND;TZID=America/Chicago:20110922T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, Conference Room, 4301, Argonne National Laboratory DESCRIPTION:Reproducibility of execution times for parallel applications is being discussed as a huge problem these days. Is it really as bad as claimed? In this work we seek to prove/disprove the hypothesis. This talk will survey the available performance analysis tools with respect to I/O behavior and make some suggestions and observations.\n SUMMARY:A Road to I/O Variability UID:1439 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110923T103000 DTEND;TZID=America/Chicago:20110923T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, 4301, Argonne National Laboratory DESCRIPTION:Interior point methods rely heavily on sparse direct linear solvers for their performance and numerical robustness. For the most part, the sparse solvers are treated as black boxes; they are unaware that the linear systems come from an interior point method. This talk will consider how the needs of an interior point solver might influence the design and operation of the underlying sparse solver. We\'ll survey more integrated approaches that have been proposed in the literature, and discuss a few new ones. We will use empirical data from the Gurobi barrier solver to consider the potential scope for improvement. SUMMARY:Sparse linear algebra for interior point methods UID:1441 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20111005T100000 DTEND;TZID=America/Chicago:20111005T110000 DTSTAMP:20130525T020110 LOCATION:Building 240, Conference Room 4301, Argonne National Laboratory DESCRIPTION:Microbial ecology is an exciting and rapidly growing area of biology, with almost weekly publications in Science, Nature, PNAS, and even popular literature sources such as The New York Times. This field also exemplifies the increasingly data-intensive nature of modern Biology: a single study can easily generate greater than 80 gigabytes of raw sequence data and is therefore multidisciplinary by requirement. In this talk, Dr. Caporaso will present his recent work on increasing the scale on which microbial ecology is possible, both in terms of breadth (the types of communities that can be profiled in high-throughput) and depth (the amount of data that can feasibly be collected and analyzed). In particular he will talk about his work on the QIIME (Quantitative Insights Into Microbial Ecology; www.qiime.org) software package and on developing a community sequencing protocol for the Illumina sequencing technologies. These tools have made it possible to increase the scale of these studies by about 2000x in just two years without increasing the cost per sequence. Dr. Caporaso will conclude by presenting several projects that illustrate what is possible in ultra-high-throughput microbial ecology: for example, a timeseries analysis of the human microbiome profiling four body sites from two individuals with daily sampling for up to 18 months. SUMMARY:Ultra-high-Throughput Microbial Ecology: Software, Sequencing and Practice for Studying Tens of Thousands of Environments UID:1445 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20111004T083000 DTEND;TZID=America/Chicago:20111004T170000 DTSTAMP:20130525T020110 LOCATION:Building 240 Conference Room 1416, Argonne National Laboratory DESCRIPTION:The Getting Started workshop provides users with information on ALCF services and resources, and the techniques and knowledge they need to use our systems.\n\nPresentation topics are listed, and will include assisted hands-on exercises where applicable:\n\n Blue Gene/P architecture; ALCF infrastructure; Our software environment; Debugging; Visualization systems and services; RepastHPC and Globus Online SUMMARY:Argonne Leadership Computing Facility UID:1447 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20111006T133000 DTEND;TZID=America/Chicago:20111006T143000 DTSTAMP:20130525T020110 LOCATION:Building 240, Conference Room 4301, Argonne National Laboratory DESCRIPTION:LCF Seminar - Ivo Kabadshow, Juelich \n\nABSTRACT: The simulation of pairwise interactions in huge particle ensembles is a vital issue in scientific research. Especially the calculation of long-range interactions poses limitations to the system size, since these interactions scale quadratically with the number of particles. This talk is divided into two parts. First I will present the open source ScaFaCoS library. The purpose of the framework is to provide a unified parallel library for various methods including the fast multipole method, the Barnes-Hut tree method, particle-mesh Ewald methods or multigrid methods to solve electrostatic and gravitational problems in large particle simulations. In the second part I will highlight one specific method of the library, namely the FMM. We improved the fast summation algorithm by adding a tight error control scheme and on-the-fly runtime minimization. The current code also benefits from a reduced memory footprint and is therefore capable of computing all pairwise interactions for systems with open, 1D, 2D and 3D periodic boundaries on 300k BG/P cores with up to three trillion particles. To achieve strong parallel scaling, the code employs a one-sided, non-blocking parallelization scheme with small communication overhead. SUMMARY:FMM for petascale and beyond UID:1449 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20111006T120000 DTEND;TZID=America/Chicago:20111006T130000 DTSTAMP:20130525T020110 LOCATION:Cafeteria, Private Dining Rooms A&B, Argonne National Laboratory DESCRIPTION:Supernova dust in the Solar System SUMMARY:Supernova Dust in the Solar System UID:1451 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20111014T103000 DTEND;TZID=America/Chicago:20111014T113000 DTSTAMP:20130525T020110 LOCATION:Building 240 Conference Room 4301, Argonne National Laboratory DESCRIPTION:Abstract: \n\nThe production of biofuels from plant feedstocks that are not necessary for food production offers an attractive renewable solution both for the production of fuels as well as other ``specialty compounds\" that can improve the commercial viability of the overall process. However, the ability to use biomass for fuel production depends critically on our ability to extract the lignocellulosic components of the biomass in a cost- and energy-efficient manner. One such family of approaches involves the use of densely charged solutions as reaction media for the dissolution and transformation operations. However, the exact nature of the chemical structure of these media, and how they interact both with lignocellulosic biomass, as well as the derivatives of these materials that can be directly converted into fuels, is not well understood. Because of the inherently multiscale nature of these problems, there are myriad opportunities for computational materials science and engineering to offer insights into this problem. In addition to progress in understanding the fuel production process, we are also actively exploring connections between different computational paradigms (for instance, linking molecular dynamics simulations with quantum-based methods such as the COSMO-RS approach). In this talk, we will discuss some of the preliminary findings of our research group with respect to the interaction between solvent media and intermediary compounds, their interaction with the ``bound water\'\' trapped in the cellulosic matrix, and inroads in connecting the different simulation techniques. SUMMARY:Applying Computational Materials Science to Problems in Biofuels Production UID:1453 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20111013T140000 DTEND;TZID=America/Chicago:20111013T150000 DTSTAMP:20130525T020110 LOCATION:Building 240, Conference Rooms 1404 and 1405, Argonne National Laboratory DESCRIPTION:We study the propagation of turbulent premixed flames using a Navier-Stokes/front-capturing methodology within the context of hydrodynamic theory. The flame is treated as a thin layer separating burned and unburned gases, of vanishingly small thickness, smaller than the smallest fluid scales. The results are applicable to the wrinkled flamelet regime of turbulent combustion. In particular, we explore the individual effects of turbulence intensity, turbulence scale, thermal expansion, hydrodynamic strain and hydrodynamic instability on the propagation characteristics of the flame. Results are obtained assuming positive Markstein length, corresponding to lean hydrocarbon-air or rich hydrogen-air mixtures. For “stable planar flames” we find a quadratic dependence of turbulent speed on turbulence intensity, modulated by the effects of thermal expansion and integral scale. Upon onset of hydrodynamic instability, corrugated structures replace the planar conformation and we observe a greater resilience to turbulence, the quadratic scaling being replaced by scaling exponents less than one. Such resilience is also confirmed by the observation of a threshold turbulence intensity below which the propagation speed of corrugated flames is indistinguishable from the laminar speed. SUMMARY:The Propagation of Wrinkled Turbulent Flames UID:1455 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20111012T150000 DTEND;TZID=America/Chicago:20111012T160000 DTSTAMP:20130525T020110 LOCATION:Building 240, 1404-1405, Argonne National Laboratory DESCRIPTION:Stochastic parameterization of large scale models is becoming an important tool for evaluating climate predictability and for enabling accurate representation of the effects of micro-scale and short time physics on climate model statistics. Implementation of numerical schemes that properly handle these stochastic effects in large scale software is highly non-trivial. A systematic process is needed to apply stochastic parameterization techniques to representative climate models in an automated or semi-automated way. The source code transformation techniques pioneered by the automatic differentiation community are being investigated to address this problem.\n\nIn this talk I will review current research on stochastic parameterization, highlight some of the difficulties, and discuss the source code transformation approach. SUMMARY:Incorporation of Stochastic Effects in Climate Models UID:1469 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20111214T150000 DTEND;TZID=America/Chicago:20111214T160000 DTSTAMP:20130525T020110 LOCATION:Building 240, 1404-1405, Argonne National Laboratory DESCRIPTION:We present the Krylov-subspace solver MINRES-QLP and its FORTRAN 90 implementation for solving symmetric or Hermitian linear systems or least-squares problems. If the system is singular, MINRES-QLP computes the minimum-length solution. In all cases, it circumvents a potential instability in the original MINRES algorithm.\n\nA positive-de nite preconditioner may be supplied. Our FORTRAN 90 implementation illustrates a design pattern that allows users to make problem data known to the solver, but hidden and secure from other program units. Moreover, users are spared to program subroutines for reverse communication, which is widely used in scienti c computing with FORTRAN 77 but the resulting code usually appears formidable and sacrifices readability.\n\nWe also provide and maintain FORTRAN 77 and MATLAB 7.8 versions of MINRES and MINRES-QLP.\n\nThis is joint work with Christopher Paige and Michael Saunders. SUMMARY:MINRES-QLP for Singular Symmetric and Hermitian Linear Equations and Least-Squares Problems UID:1471 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20111021T103000 DTEND;TZID=America/Chicago:20111021T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, Conference Room, 4301, Argonne National Laboratory DESCRIPTION:Exascale is the next grand challenge for parallel file systems. \nConcurrency orders of magnitude higher than what is seen today will\ntest the scalability of the storage system as never before and an\nexplosion of data and, most importantly application metadata will\ntest current parallel filesystems models beyond breaking point.\n\nThis talk proposes a new model of I/O that replaces the single contiguous byte array that is a conventional file with object storage containers that guarantee scalability for distributed applications without polluting the filesystem namespace. Safe application-level access to object storage allows the creation of a common storage API that expresses concurrency explicitly so that many different high level I/O models tailored to different application domains can be layered above it cleanly without loss of scalability or performance. \n\nEric Barton has worked in HPC since the 1985, when he co-founded Meiko Scientific to build clustered supercomputers. He has worked on all aspects of HPC systems software development from communications and middleware libraries down to low-level drivers and has a special interest in parallel file systems. Since 2002, he has worked exclusively on the Lustre file system, initially taking responsibility for its networking layer and more recently at Sun, Oracle and now Whamcloud as the main Lustre architect. SUMMARY:DAOS containers - a storage abstraction for exascale UID:1477 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20111028T133000 DTEND;TZID=America/Chicago:20111028T143000 DTSTAMP:20130525T020110 LOCATION:Building 240 / Room 4301, Argonne National Laboratory DESCRIPTION:Aquaporin-1 (AQP1) is a specialized membrane protein that is crucial to facilitating rapid water exchange across biological membranes. As a representative example of the aquaporin family, the precise interpretation of its fingerprint structures is critical to understanding the general diversification of plants and vertebrates, as well as AQP1’s specific role in developing electrochemical gradient across the cell membranes. In this presentation, recent progress on clarifying the mechanism of proton solvation and transport in a series of AQP1 mutant channels will be addressed. Our multiscale modeling indicates the function of protonatable residues in determining channel proton conductance. The influence of the proton on alkali cation translation behaviors will also be addressed. Finally, the implications of this study on the evolution of the major intrinsic proteins will be presented. \n\nClick on the link to add this seminar to your calendar. SUMMARY:Multiscale Modeling of Proton Solvation and Transport in Aquaporin-1 Channels UID:1475 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20111031T140000 DTEND;TZID=America/Chicago:20111031T150000 DTSTAMP:20130525T020110 LOCATION:Building 240, TCS Conference Center, 1404 & 1405, Argonne National Laboratory DESCRIPTION:Exascale machines, and the kinds of applications that are expected to run on them, pose significant challenges for the programming models community. Significant increase in the degree of parallelism within a node, heterogeneity, the sheer number of computing elements available, as well as considerations of power and resilience create challenges and opportunities for the design of a new generation of programming models. I will outline a few broad imperatives, some of which require radical changes for such designs. \n\nTo begin with, I will argue that it would be beneficial to (almost) eliminate the notion of “processor” from the ontology of the programmer. The program should be expressed in terms of the work-units and data-units of the application without reference to which processor they are housed on. This can be supported by an automated resource management system which can adapt to variations in the application configurations/evolution, as well as to variations in the machine environment. At exascale, the programming models will have to depend on novel, highly powerful and intelligent adaptive runtime systems. These systems will (a) monitor multiple aspects of an ongoing execution, (b) take corrective actions in order to optimize multiple criteria such as performance, and power, and (c) effect recovery from component failures. Further, parallel composition of multiple, independently developed modules must be strongly supported, without requiring partitioning processors or sequencing modules. Message-driven execution is a powerful runtime mechanism that, in my opinion, is essential for supporting such modularity. \n\nSuch modularity is impossible without support for interoperability between programming models, including legacy models such as MPI and OpenMP. The runtime system can tie these modules together and support communication between disparate models, allowing the programmer to use whichever model is best suited to each portion of an application. Interoperability between models will also provide an opportunity to simplify parallel programming by introducing simple, incomplete programming models which may be incapable of expressing arbitrary parallel interactions or be limited to specific data structures but which provide increased safety and simplicity without sacrificing performance. I expect the exascale toolbox to eventually include a small set of interoperable incomplete models designed to support common patterns encountered in parallel applications. Further, we need to develop new data-structure specific frameworks, constructed from the point of view of interoperability. Compiler support will be required, not for automatic parallelization, but for supporting convenient syntax and basic analysis aimed at optimization. SUMMARY:Composable and modular Exascale Programming Models with intelligent runtime systems UID:1483 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20111103T150000 DTEND;TZID=America/Chicago:20111103T160000 DTSTAMP:20130525T020110 LOCATION:Building 240, 1404-1405, Argonne National Laboratory DESCRIPTION:This presentation will introduce what the Persepolis Fortification Archive (PFA) is (tens of thousands of clay tablets bearing texts in several ancient languages and impressions of thousands of seals, from about 500 BC), why it matters (rich, detailed and unique source of data on languages, art, institutions, society, and religion at the center of the Achaemenid Persian Empire), what the problem is (a lawsuit that threatens the integrity of the archive and future research access to the original documents), and what the PFA Project\'s (see http://oi.uchicago.edu/research/projects/pfa/, and http://persepolistablets.blogspot.com/) response is (enable continuing research by making and distributing usable records of the objects); and will then discuss some of the Project\'s imaging techniques (high-quality conventional scans and Reflectance Transformation Imaging) and one of its forms of dense presentation of images, editorial and analytical content (the On-Line Cultural Heritage Research Environment, see http://\nochre.lib.uchicago.edu/). SUMMARY:Emergency recording of persian antiquities: electronic imaging editing, and presentation of the persepolis fortification archive UID:1485 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:00000000T150000 DTEND;TZID=America/Chicago:00000000T160000 DTSTAMP:20130525T020110 LOCATION:Building 240, 1404-1405, Argonne National Laboratory DESCRIPTION:To be announced SUMMARY:Title to be announced UID:1487 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20111207T150000 DTEND;TZID=America/Chicago:20111207T160000 DTSTAMP:20130525T020110 LOCATION:Building 240, 1404-1405, Argonne National Laboratory DESCRIPTION:An algorithm for large-scale nonlinear nonconvex optimization is presented. Search directions are computed by an iterative linear solver, with termination criteria for the inexact solution ensuring global convergence under mild assumptions. The practical performance of the method is demonstrated on 3D PDE-constrained optimization examples. SUMMARY:Large-Scale Nonlinear Optimization with Inexact Step Computations UID:1489 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20111109T150000 DTEND;TZID=America/Chicago:20111109T160000 DTSTAMP:20130525T020110 LOCATION:Building 240, 4301, Argonne National Laboratory DESCRIPTION:Two talks for the price of one! I will present 2 topics on mixed-integer linear programming (MILP), one application (transmission switching), and one theoretical results (optimizing over extended formulations).\n\nTransmission switching provides a way to increase the efficiency in power systems operations by altering the topology of the transmission network. Altering the transmission topology can affect the reliability of the network. Incorporating reliability into the optimization problem increases the difficulty of an already complex optimization problem. We present an algorithm to deal with the islanding problem caused by transmission switching that can significantly decrease computation time.\n\nWe improve upon the LP relaxation bounds by computing the bound given by the Sherali-Adams extended formulation for highly symmetric MILP problems. Typically the extended formulations of Sherali-Adams relaxations can be very large. Symmetry can be used to generate an LP with significantly fewer variables that has an identical objective value. We demonstrate this by computing the bound associated with the level 1, 2, and 3 relaxations of several highly symmetric binary integer programming problems. SUMMARY:Anti-Islanding in Transmission Switching AND Using Symmetry to Optimize Over Extended Formulations UID:1491 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20111130T150000 DTEND;TZID=America/Chicago:20111130T160000 DTSTAMP:20130525T020110 LOCATION:Building 240, Conference Center 1404-1405, Argonne National Laboratory DESCRIPTION:In statistics we often have some a priori information or assumptions on the qualitative properties of the estimator, including constraints on the shape and support of the estimator. For instance, we may require that the estimator be increasing or decreasing, convex or concave, or unimodal over an interval, be bounded from above or below, or some combination of these. In this talk we shall discuss how to incorporate such information as hard constraints into the optimization model defining the estimator when the estimator is a polynomial spline. We show that in the univariate case these shape-constrained estimators can be computed in polynomial time. Examples from density estimation, monotone and convex regression, and arrival rate estimation will be presented. Time permitting, we shall discuss the theoretical complications that arise in the multivariate case, and present a few multivariate applications as well.\n SUMMARY:Statistical estimation with shape constraints UID:1492 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20111121T103000 DTEND;TZID=America/Chicago:20111121T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, 4301, Argonne National Laboratory DESCRIPTION:GPSI (pronounced \"gypsy\") provides computational scientists with a general purpose workbench for developing, testing, and using complex workflows for simulation and analysis. A key aspect that differentiates GPSI from other environments that support large parallel workflows is its integration with your desktop development practices. Individual process steps developed on the desktop using familiar tools and languages can be smoothly moved into GPSI to contribute to larger multi-part workflows. Piece by piece, large simulation and analysis pipelines can be built. GPSI promotes rapid prototyping through modularity, reuse mechanisms, and code sharing. The GPSI environment integrates support for job execution and management; data management and browsing; and application development and reuse. We are working with research groups in power grid simulation and analysis, automatic image analysis of APS cell data, computational neuroscience, and computational material science to improve productivity via GPSI. SUMMARY:Introducing GPSI: YOUR New Computing Environment UID:1494 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20111129T150000 DTEND;TZID=America/Chicago:20111129T160000 DTSTAMP:20130525T020110 LOCATION:Building 240, 1404-1405, Argonne National Laboratory DESCRIPTION:Our work is concerned with the development of a generic high-performance library for scientific computing. The library is targeted for assembling flexible-order finite-difference solvers for PDEs. Our goal is to enable fast solution of large PDE systems, fully exploiting the massively parallel architecture of Graphics Processing Units. We will detail a strategy for an iterative mixed-precision defect correction method, with p-multigrid preconditioning. We present a case study of the fully nonlinear potential flow equations, in which the bottleneck problem is finding the solution of a Laplace equation. Such large scale simulations can be of great value in coastal and offshore engineering application. SUMMARY:A Fast Mixed-Precision Strategy for Iterative GPU-Based Solution of the Laplace Equation UID:1496 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20111109T110000 DTEND;TZID=America/Chicago:20111109T120000 DTSTAMP:20130525T020110 LOCATION:Building 240, Conference Room 4301, Argonne National Laboratory DESCRIPTION:The societal impact of a weather event increases in proportion to its rarity yet our current ability to model extreme events is limited by not only rarity but also by the fidelity of current data and approaches and a lack of understanding of the way many of the underlying physical processes may vary in time. This challenging conflict is driving fresh approaches to assessing the way that high impact weather adjusts to climate variability and change. Here I discuss lessons learnt, lessons yet to be learnt, and promising avenues of exploration following several years’ experience with the NCAR Nested Regional Climate Modeling System. \n\nLessons learnt include: the importance of handling bias in the global driving model; the need for considerable care in selecting the domain size for high-resolution simulation; the substantial benefits from adopting a hybrid dynamical statistical approach; and the simple fact that the uncertainty level goes up with increased resolution. Lessons yet to be learnt (aka known unknowns) include: useable approaches to assessing uncertainty; the consequences of upscale interactions, where the local scale influences the global; and how to best integrate statistical-dynamical modeling with direct impacts assessment (societal, commercial and ecological). Each of these challenges opens up promising avenues of exploration. We are already exploring some of these, but there are many more opportunities, all of which hold the potential for opening up future research collaborations between ANL and NCAR.\n SUMMARY:Challenges in Next-Generation Regional Climate Modeling for Assessing High-Impact Weather UID:1498 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20111110T133000 DTEND;TZID=America/Chicago:20111110T143000 DTSTAMP:20130525T020110 LOCATION:Building 240 / Room 4301, Argonne National Laboratory DESCRIPTION:Quantum chemistry is extremely accurate and easily applicable in gas phase molecular systems. However, the same accuracy and ease is not always enjoyed when applying quantum chemistry to condensed phase systems, such as molecules dissolved in solvent. In this presentation, we discuss our efforts in modeling solution phase chemistry and the difficulties we have had to overcome in order to do so. For example, we analyze spurious charge transfer contamination in Time Dependent Density Functional Theory (TD-DFT) and how it can be ameliorated with long-range corrected functionals, such as LRC-wPBE. We then discuss the application of this functional to understanding excited electronic states of aqueous DNA, including its exciton and charge transfer states. In addition, we present several recent advances in the theory and computation of polarizable continuum models (PCMs) for implicit solvation, which are commonly used in quantum chemistry calculations. We discuss failures of traditional PCM methodology, our approach for rectifying these failures, our newly derived PCM for solving the linearized Poisson-Boltzmann equation, and our linear scaling parallel implementation of PCM for large solute molecules.\n\nClick on this link to add the event to your calendar. SUMMARY:Challenges of Quantum Chemistry in Solvent from Electronically Excited DNA to Polarizable Continuum Models UID:1504 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20111212T130000 DTEND;TZID=America/Chicago:20111212T140000 DTSTAMP:20130525T020110 LOCATION:Building 240, Conference Room 1404, Argonne National Laboratory DESCRIPTION:Dr. Sha will present a set of rigorously derived conservation equations of mass, momentum, and energy for multiphase systems with internal-stationary-solid structures. The starting point of the derivation is the conservation of mass, momentum, and energy equations and their interfacial balance equations. The local volume averaging is carried out first for the conservation equations and their interfacial balance equations, and followed by the time averaging of the local volume averaged conservation equations and their interfacial balance equations. A set of time averaging of local volume averaged conservation equations is in differential-integral form and is not a set of partial differential equations as currently “appear” in most literatures on multiphase flows. The integrals arise due to interfacial mass, momentum and energy transfer. The time-volume averaged conservation equations of mass, momentum, and energy and their interfacial balance equations serve as a reference point for modeling multiphase flow with simplified approximations and provide theoretical guidance and physical insight that may be useful to develop correlations for quantifying interfacial mass, momentum, and energy transfer between phases. SUMMARY:Novel Porous Media Formulation for Multiphase Flow Conservation Equations UID:1514 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120118T150000 DTEND;TZID=America/Chicago:20120118T160000 DTSTAMP:20130525T020110 LOCATION:Building 240, Conference Center 1404-1405, Argonne National Laboratory DESCRIPTION:Atomistic and first principles computational modeling has become indispensable in the understanding and design of materials. In this field, the optimization of physical properties as a function of atomic configurations is a common problem. Such optimization often requires stochastic and evolutionary approaches, as well as efficient transformations of the configuration -> property maps. In this talk, I will discuss a few examples in which we approach such optimization problems, in applications towards thermoelectric and energy storage materials. Specifically, I will discuss configurational determination in lithium battery materials and optimization of thermal conductivity in SiGe nanowires. Other materials-related problems for which novel algorithmic approaches may be useful will also be discussed. SUMMARY:Optimization problems in atomistic and first principles materials modeling -- a few examples UID:1516 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20111201T103000 DTEND;TZID=America/Chicago:20111201T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, 1404-1405, Argonne National Laboratory DESCRIPTION:Accelerating algebraic multigrid methods on massively parallel throughput-oriented processors, such as the GPU, demands algorithms with abundant fine-grained parallelism. Sparse matrix-vector multiplication operations dominate the performance of the cycling phase of algebraic multigrid and we use efficient GPU implementations to achieve notable speedup on a representative set of matrices. We also present novel sparse matrix operations required to construct the AMG hierarchy. The GPU sparse matrix-matrix and maximal independent operations avoid transfer operations and achieve an average of 2x speedup. Our algorithms are expressed as collections of data parallel primitives provided by the Thrust library and available as part of the Cusp sparse matrix library. SUMMARY:Accelerating Algebraic Multigrid on GPUs UID:1518 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20111122T150000 DTEND;TZID=America/Chicago:20111122T160000 DTSTAMP:20130525T020110 LOCATION:Building 240, Conference Center 1404-1405, Argonne National Laboratory DESCRIPTION:The hallmark of the cybernetic approach is its description of cellular regulation to provide for a dynamic framework of metabolic modeling. It views cells as optimal strategists frugally allocating internal resources among reactions in order to maximize a metabolic objective (e.g., carbon uptake rate or growth rate). Earlier development of cybernetic modeling had focused on the growth pattern on multiple substrates based on gross metabolic networks (e.g., Kompala et al. 1986). More recently, the hybrid cybernetic model (HCM) (Kim et al. 2008, Song et al. 2009, Song and Ramkrishna 2009) and lumped HCM (L-HCM) (Song and Ramkrishna 2010, 2011) has enabled consideration of more detailed networks through decomposition into elementary modes (EMs). Extension of this methodology to genome-scale networks, has, however, been challenged by difficulties associated with decomposition of large networks into EMs. So far, genome-scale networks have been handled only by flux balance analysis (FBA) and its derivatives (Orth et al. 2010). Application of those constraint-based approaches to metabolic modeling and metabolic engineering is subject to limitations because of lack of attention to regulatory dynamics. \n\nIn this work, a methodology is presented to incorporate genome-scale metabolic networks into the L-HCM framework. The L-HCMs describe cellular resources in terms of competition among lumped EMs (L-EMs) which are weighted averages of EMs in families classified according to similarity of metabolic function. As the weight takes a power-law form with an exponent of large constant, only a limited number of EMs play a role in computing L-EMs. Thus, instead of endeavoring to get the full set of EMs, only these \"dominant\" modes with an appreciable contribution to the averaging process are extracted. For this purpose, a MATLAB code is developed based on the mixed integer linear programming (MILP) algorithm originally proposed by Lee et al. (2000). In various test examples, it is shown that the MILP code successfully smokes out essential pathways including multiple optima and suboptima which are essential to compute L-EMs. This development enables the cybernetic modeling approach to address genome-scale networks. Also possible is direct comparison of L-HCMs with constraint-based approaches using the same size of genome-scale network. \n\nReferences:\nKim JI, Varner JD, Ramkrishna D. 2008. A Hybrid Model of Anaerobic E. coli GJT001: Combination of Elementary Flux Modes and Cybernetic Variables. Biotechnology Progress 24(5):993-1006.\nKompala DS, Ramkrishna D, Jansen NB, Tsao GT. 1986. Investigation of Bacterial-Growth on Mixed Substrates - Experimental Evaluation of Cybernetic Models. Biotechnology and Bioengineering 28(7):1044-1055.\nLee S, Phalakornkule C, Domach MM, Grossmann IE. 2000. Recursive MILP model for finding all the alternate optima in LP models for metabolic networks. Computers & Chemical Engineering 24(2-7):711-716.\nOrth JD, Thiele I, Palsson BO. 2010. What is flux balance analysis? Nature Biotechnology 28(3):245-248.\nSong HS, Morgan JA, Ramkrishna D. 2009. Systematic Development of Hybrid Cybernetic Models: Application to Recombinant Yeast Co-Consuming Glucose and Xylose. Biotechnology and Bioengineering 103(5):984-1002.\nSong HS, Ramkrishna D. 2009. Reduction of a Set of Elementary Modes Using Yield Analysis. Biotechnology and Bioengineering 102(2):554-568.\nSong HS, Ramkrishna D. 2010. Prediction of Metabolic Function From Limited Data: Lumped Hybrid Cybernetic Modeling (L-HCM). Biotechnology and Bioengineering 106(2):271-284.\nSong HS, Ramkrishna D. 2011. Cybernetic Models Based on Lumped Elementary Modes Accurately Predict Strain-Specific Metabolic Function. Biotechnology and Bioengineering 108(1):127-140. SUMMARY:Towards Genome-Scale Based Dynamic Modeling of Cellular Metabolism. The Cybernetic Approach UID:1520 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20111202T103000 DTEND;TZID=America/Chicago:20111202T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, 4301, Argonne National Laboratory DESCRIPTION:This talk will consider multigrid methods in the context of the scales of problems and computer architectures that are foreseen in the next 5-10 years. This includes issues of communication reducing data models and asynchronous algorithms. I will start by establishing what is probably a lower bound on work and memory complexity, and perhaps data movement complexity, for solving the algebraic equations that arise from discretized elliptic PDEs. Rigorous proofs will be omitted but an analysis is presented that suggests practical methods to extend this \"textbook multigrid efficiency\" to operators where rigorous theory is silent. An example of fast solvers for fully implicit eight field resistive compressible magnetohydrodynamics is presented. Algebraic multigrid (AMG) methods are introduced and application from science and industry are presented, including a Gordon Bell prize winning application from 2004. Parallel data models and asynchronous algorithms for Gauss-Seidel are presented that exploit special characteristics of discretized PDE graphs and are much faster than generic coloring approaches. Though Gauss-Seidel smoothers are largely eclipsed in practice by additive methods they are potentially useful for very unsymmetric operators and these algorithms lead to ideas that are useful in optimizing performance of coarse grid work in large scale multigrid solves. These ideas have been used in the previous work presented here and are used in recent work in developing unstructured geometric and AMG methods as native PETSc solvers using PETSc\'s common parallel primitives. Strong and weak scaling results are presented for a 3D elasticity model problem and a unstructured FEM 2D Poisson solver on an ITER mesh from the fusion gyrokinetic PIC code XGC1.\n SUMMARY:Communication Reducing Data Models and Asynchronous Algorithms UID:1526 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120117T150000 DTEND;TZID=America/Chicago:20120117T160000 DTSTAMP:20130525T020110 LOCATION:Building 240, 1404-1405, Argonne National Laboratory DESCRIPTION:Isotonic regression is a nonparametric approach for fitting monotonic models to data that has been widely studied from both theoretical and practical perspectives. However, this approach encounters computational and statistical overfitting issues in higher dimensions. To address both concerns we present an algorithm, which we term Isotonic Recursive Partitioning (IRP), for isotonic regression based on recursively partitioning the covariate space through solution of progressively smaller ``best cut\'\' subproblems. This creates a regularized sequence of isotonic models of increasing model complexity that converges to the global isotonic regression solution. Models along this sequence are often more accurate than the unregularized isotonic regression model because of the complexity control they offer. We quantify this complexity control through estimation of degrees of freedom along the path. Furthermore, we show that IRP for the classic l2 isotonic regression can be generalized to convex differentiable loss functions such as Huber\'s loss. In another direction, we use the Lasso framework to develop another isotonic path of solutions that is computationally more expensive but offers even better complexity control. Success of the regularized models in prediction and IRP\'s favorable computational properties are demonstrated through a series of simulated and real data experiments. SUMMARY:Efficient Regularized Isotonic Regression via Partitioning and L1 Penalization UID:1528 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20111212T103000 DTEND;TZID=America/Chicago:20111212T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, Conference Room, 4301, Argonne National Laboratory DESCRIPTION:The Square Kilometer Array (SKA) is an international project that aims to build a distributed radio telescope with a maximum baseline in the order of 3000 km, either in South Africa or Western Australia.\n\nAlthough the estimated processing requirements still vary wildly, these are clearly beyond the capabilities of even the most modern supercomputers. The first phase of the SKA, scheduled for first operations in 2018, requires anywhere from several to several hundred petaflop/s, while the full SKA, which we expect to build around 2020, is well on its way to require far in excess of an exaflop/s. Additionally, the I/O requirements run firmly into the scary range of several terabytes/s.\n\nIt is clear that the SKA project is critically dependent on improvements in high-performance computing, and the ability of the project to efficiently leverage these improvements. It is also obvious that the unique requirements of the SKA instrument mean that we cannot expect a tailored solution to just appear a few years from now.\n\nI will outline the SKA project and the experiences we\'ve had building some of the pathfinder instruments. I\'ll also look at the processing requirements demanded by the SKA, how these match onto the projected developments in high-performance computing and where we can expect them to fall short. SUMMARY:The Square Kilometer Array: exascale computing in a radio telescope UID:1530 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20111206T150000 DTEND;TZID=America/Chicago:20111206T160000 DTSTAMP:20130525T020110 LOCATION:Building 240, 1404-1405, Argonne National Laboratory DESCRIPTION:We present a scalable approach and implementation for solving stochastic programming (SP) problems, with application to the optimization of complex energy systems under uncertainty. Our methodology relies on approximating the underlying uncertainty of the SP problems via sampling, and solving the corresponding sample average approximation (SAA) problem using an interior-point method and parallel distributed-memory linear algebra. The scenario-based decomposition of the linear algebra is obtained by using a direct Schur complement technique (DSC) and allows most of the computations to be performed in parallel with the exception of the linear solves with the dense Schur complement matrix. In many applications the Schur systems are large and cause a computational bottleneck that adversely affects the large-scale computational performance and strong scaling of the DSC method.\n\nWe present two approaches that circumvent this bottleneck. The first approach uses a novel stochastic preconditioner and Krylov iterative methods to mitigate the expensive dense factorization of medium-sized Schur complements. The spectral analysis of the preconditioner shows that the eigenvalues of the preconditioner Schur complement clusters around unit exponentially with the size of the samples incorporated in the preconditioner. The second approach distributes and directly solves the large-sized Schur complement systems in parallel. We show numerical experiments and discuss the advantages and drawbacks of this approach in the case of a stochastic economic dispatch problem with wind power integration and transmission constraints which we solve on up to 131,072 cores of the ``Intrepid\'\' BG/P system.\n\nIn addition, we present a novel technique for the estimation of the uncertainty in the solution of SP problems. Traditional methods for the statistical inference of stochastic optimization problems require a large number of samples in order to obtain accurate uncertainty estimates. This requirement usually can not be satisfied in the case of energy models incorporating weather-related uncertainty since numerical weather prediction is extremely expensive and one cannot realistically afford a large number of samples. We propose and analyze an estimator that works with higher-order resampling methods such as bootstrapping, which provides reliable confidence regions despite of the availability of a small number of samples. Numerical evidence supporting this claim will be also presented.\n SUMMARY:Scalable stochastic programming UID:1532 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120111T150000 DTEND;TZID=America/Chicago:20120111T160000 DTSTAMP:20130525T020110 LOCATION:Building 240, 1404-1405, Argonne National Laboratory DESCRIPTION:Numerous physical phenomena and the devices which involve such phenomena are aptly modeled by Partial Differential Equations. For this reason, the stability and control of the systems modeled by PDEs is of significant importance. Unfortunately, current analytical methods for stability analysis and control of PDEs in in finite-dimensional spaces are only applicable to simple classes of PDEs. An alternative approach is to discretize the PDE at specific points in the space and represent it as a set of ODEs in finite-dimensional spaces. Although the resulting state-space system is much easier to analyze and control by using a wide range of available tools in fi nite-dimensional linear analysis, the large dimension of the state-space and the presence of uncertain parameters in the system makes the stability and control problem computationally intractable.\n\nIn this study, we propose a distributed computing approach for large-scale robust stability and control problems. First, we design a decentralized algorithm which uses Polya\'s algorithm to convert the robust stability and control conditions into a set of highly structured Linear Matrix Inequalities (LMIs). Then we show that a common implementation of a primal-dual interior-point method for solving these LMIs has a block-diagonal structure which is preserved at each iteration. By exploiting this property, we create a highly scalable cluster-computing implementation of our algorithm for stability analysis and control of large systems. The theoretical and experimental results for speed-up verify the scalability of the algorithms. Numerical examples demonstrate the ability to perform robust analysis of systems with more than one-hundred states and several uncertain parameters. SUMMARY:Distributed parallel algorithms for problems in polynomial optimization and robust control UID:1536 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20111209T103000 DTEND;TZID=America/Chicago:20111209T113000 DTSTAMP:20130525T020110 LOCATION:Bldg 240 Conference Rm: 1404 & 1405, Argonne National Laboratory DESCRIPTION:Seminar: Microbial communities are highly diverse with a richness of bacterial organisms that is often difficult to assess. Recent advances in DNA sequencing have uncovered more bacterial organisms than we had ever seen before. To augment classifications of bacterial communities based on the our limited ability to assign taxonomic names, a common technique is to sequence the ribosomal RNA gene, and cluster them, assuming that similar sequences represent organisms that are closely related and more divergent sequences represent organisms that are more distantly related. Unfortunately, different clustering algorithms can lead to different results, including sample-size dependencies. The appropriate choice of clustering technique is critical for meaningful comparisons of microbial communities. SUMMARY:Comparing Microbial Communities using DNA Clustering UID:1540 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20111216T103000 DTEND;TZID=America/Chicago:20111216T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, 1404-1405, Argonne National Laboratory DESCRIPTION:The presentation will cover the latest work performed in uncertainty quantification for the models that are too complex and computationally expensive for the standard, low-effort (non-intrusive, sampling-based) methods to be effective. Some previous work on derivative-augmented polynomial regression and Gaussian-Processes based Kriging will be explained, but the focus will be on training unknown surface response models on lower-fidelity data. SUMMARY:Intrusive Analysis and Uncertainty Quantification Tools for Simulation Codes UID:1542 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20111214T103000 DTEND;TZID=America/Chicago:20111214T120000 DTSTAMP:20130525T020110 LOCATION:Bldg 240 Seminar Room 4301, Argonne National Laboratory DESCRIPTION:The vast majority of microbes are unculturable and thus cannot be sequenced by means of traditional methods. High-throughput sequencing techniques like 454 or\nSolexa-Illumina make it possible to explore those microbes by studying whole natural microbial communities and analysing their biological diversity as well as the underlying metabolic pathways. Over the past few years, different methods have been developed for the taxonomic and functional characterization of metagenomic shotgun sequences. However, the taxonomic classification of metagenomic sequences from novel species without close homologue in the biological sequence databases poses a challenge due to the high number of wrong taxonomic predictions on lower taxonomic ranks. In this talk I will present CARMA3, a new method for the taxonomic classification of assembled and unassembled metagenomic sequences that has been adapted to work with both BLAST\nand HMMER3 homology searches. We show that our method makes fewer wrong taxonomic predictions (at the same sensitivity) than other BLAST-based methods. CARMA3 is freely accessible via the web application WebCARMA from \nhttp://webcarma.cebitec.uni-bielefeld.de. SUMMARY:Taxonomic Classification of Metagenomic Shotgun Sequences with CARMA3 UID:1544 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20111215T100000 DTEND;TZID=America/Chicago:20111215T110000 DTSTAMP:20130525T020110 LOCATION:Building 240, Conference Room, 4301, Argonne National Laboratory DESCRIPTION:This talk focuses on localizing storage-stack problems in parallel file systems by identifying, gathering and analyzing OS-level, black-box performance metrics on every server node in the cluster. Our peer-comparison diagnosis approach compares the statistical attributes of these metrics across file servers, to identify faulty disk arrays, storage controllers, and server nodes. We validate our approach by triggering real storage problems in a GPFS cluster running three different file-system benchmarks (dd, IOzone, and PostMark). We further demonstrate localization of storage problems through a preliminary analysis of our approach on the Intrepid storage cluster at Argonne National Laboratory. \n\nMike Kasick is a fifth-year PhD student in Electical & Computer Engineering at Carnegie Mellon University. His research focus is on methods for minimally-invasive, \"production-ready\" problem diagnsosis of parallel file systems and storage clusters. SUMMARY:Black-Box Localization of Storage Problems in Parallel File Systems UID:1548 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20111219T133000 DTEND;TZID=America/Chicago:20111219T143000 DTSTAMP:20130525T020110 LOCATION:Building 240, Seminar Room 4301, Argonne National Laboratory DESCRIPTION:In a growing number of information-processing applications, such as network-traffic monitoring, sensor networks, financial analysis, and data mining for e-commerce, data takes the form of continuous streams rather than traditional stored database tuples. These applications have some common features, such as, they require real time analysis, they possess huge volumes of data, and they suffer from unpredictable and bursty arrivals of data elements. In such applications, processing queries over data streams by first loading them into a traditional database management system (DBMS) or into main memory is infeasible. High speed data streams along with a large number of simultaneous continuous queries lead to resource limitations. \n\nIn this seminar, I outline the challenges of processing join queries over data streams, and present a few algorithms that generate exact results for the join queries incorporating secondary storages and non-dedicated computers. \nThe proposed techniques exploit the high bandwidth of a disk subsystem by rendering the data access pattern largely sequential, eliminating small, random disk accesses. I present an I/O-efficient algorithm to process hybrid join queries, that join a fast, time varying or bursty data stream and a persistent disk relation. Such a hybrid join is the crux of a number of common transformations in an active data warehouse. The algorithm reduces the response time in output results by exploiting spatio-temporal locality within the input stream, and minimizes disk overhead through disk-I/O amortization. \n\nLastly, I present a mechanism to distribute the loads of a stream join operator across a shared nothing system. The algorithm uses a fixed or predefined communication pattern, and dynamically maintains the degree of declustering to minimize communication and processing overheads. Experimental results show the efficacy of the proposed algorithms. SUMMARY:Processing Sliding Window Joins Over High Speed Data Streams UID:1550 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20111220T133000 DTEND;TZID=America/Chicago:20111220T143000 DTSTAMP:20130525T020110 LOCATION:Building 240, Seminar Room 4301, Argonne National Laboratory DESCRIPTION:Providing timeliness support for multihop wireless networks in a resource-efficient manner is an important and challenging problem, particularly at the medium access control layer. However, existing solutions are either over-coordinated (fixed-schedule-based schemes) or under coordinated (prioritized contention-based schemes), failing to address this problem efficiently. This paper introduces DRAMA, a new distributed, progressive, dynamic slot reservation mechanism, aiming to provide timeliness support at the medium access control layer. In DRAMA, each node progressively and dynamically makes short-term slot reservations according to the timeliness and bandwidth requirements of its outgoing traffic, thereby quickly adapting to traffic and link dynamics. Potentially interfering nodes reserve slots in a serialized and orthogonal manner, which ensures fast, negotiation-free, contention-free slot reservations with high bandwidth utilization and low bandwidth overhead. Similar to fixed schedule-based approaches, nodes in DRAMA can enter a low-power sleep mode when they do not transmit or receive data. Our experimental results show that DRAMA is able to meet end-to-end latencies under various traffic and link dynamics when other schemes may not be able to do so, while incurring little energy and bandwidth overhead. SUMMARY:DRAMA: Dynamic Reservation Medium Access for Multihop Wireless Real-Time Communications (Design and Implementation) UID:1554 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120111T103000 DTEND;TZID=America/Chicago:20120111T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, Conference Room, 4301, Argonne National Laboratory DESCRIPTION:Over the past several years, there has been increasing interest in injecting a layer of compute resources between a high-performance computing application and the end storage devices. For some projects, the objective is to present the parallel file system with a reduced set of clients, making it easier for file-system vendors to support extreme-scale systems. In other cases, the objective is to use these resources as \"staging areas\" to aggregate data or cache bursty I/O operations, thus improving the \"effective\" I/O throughput seen by the application. Still others want to use these staging areas for performing \"in-situ\" analysis on data in-transit between the application and the storage system. To simplify our discussion, we adopt the general term \"Integrated Data Services\" to represent these use-cases for HPC compute resources.\n\nAlthough there is great interest in providing integrated data services for HPC platforms, a number of issues exist that hinder these efforts on today\'s platforms:\n\n - There is no standard, portable, API to support data services across platforms.\n - There is no scheduler or runtime support for dynamic data services.\n - Security models sometimes hinder the use of data services.\n - Very little has been done to address resilience issues created by data services.\n\nIn this talk, I describe R&D efforts at Sandia to address some of these issues. In particular, I will introduce the Network Scalable Service Interface (Nessie), a parallel remote-procedure call API designed to enable the rapid development of data services on a variety of HPC platforms. I will also briefly describe a number of data services created using Nessie, including a PnetCDF staging service, an SQL proxy service, and an in-transit analysis service for the CTH shock-physics code. Finally, I will discuss new research directions, including an investigation of how to address resilience issues created by the use of data services.\n\n------- Bio ------\n\nRon A. Oldfield is a principal member of the technical staff at Sandia National Laboratories in Albuquerque, NM. He received the B.Sc. in computer science from the University of New Mexico in 1993. From 1993 to 1997, he worked in the computational sciences department of Sandia National Laboratories, where he specialized in seismic research and parallel I/O. He was the primary developer for the GONII-SSD (Gas and Oil National Information Infrastructure--Synthetic Seismic Dataset) project and a co-developer for the R&D 100 award winning project \"Salvo\", a project to develop a 3D finite-difference prestack-depth migration algorithm for massively parallel architectures. From 1997 to 2003 he attended graduate school at Dartmouth college and received his Ph.D. in June, 2003. In September of 2003, he returned to Sandia to work in the Scalable Computing Systems department. He has been the PI of a number of I/O, resilience, and systems simulation projects; and is currently project manager for the ASC/CSSE Scalable I/O Research at Sandia. His research interests include parallel and distributed computing, parallel I/O, resilience, and performance modeling. SUMMARY:Supporting Integrated Data Services: A New Challenge for High-Performance Computing UID:1556 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120105T110000 DTEND;TZID=America/Chicago:20120105T120000 DTSTAMP:20130525T020110 LOCATION:240 Conference Room 1404-05, Argonne National Laboratory DESCRIPTION:Our modern ability to acquire and generate huge amounts of data can potentially enable rapid progress in science and engineering, but we may not live up that promise if our ability to create data outstrips our ability to make sense of that data. In this talk I will present our work on visual computing in Connectomics, a new field in neuroscience that aims to apply biology and computer science to the grand challenge of determining the detailed neural circuitry of the brain. I will give an overview of the computational challenges and describe interactive visualization approaches that we developed to discover and analyze the brain\'s neural network. The key to our methods is to keep the user in the loop, either for providing input to our downstream reconstruction methods, or for validation and corrections of the reconstructed neural structures. The main challenges we face are how to visualize petabytes of image data in an efficient and scalable way, how to automatically reconstruct very large and dense neural circuits from the nanoscale-resolution electron micrographs, and how to analyze the brain\'s neural network once we have discovered it. SUMMARY:Visual Computing and Connectomics UID:1558 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120110T103000 DTEND;TZID=America/Chicago:20120110T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, Seminar Room 4301, Argonne National Laboratory DESCRIPTION:Over the past two decades, integral equation methods have become very successful at the numerical solution of many important differential equations. This has been due in large part to the development of algorithms to rapidly apply certain Greens function matrices, which, when coupled with Krylov subspace methods, have enabled fast iterative solvers with optimal or near-optimal complexities. Although such techniques have transformed many aspects of computational science and engineering, they remain somewhat ineffective in the face of ill-conditioning or the need to solve many related systems. Here, we discuss a fast direct solver for non-oscillatory integral equations that overcomes these deficiencies. The algorithm is based on a new form of multilevel matrix compression extending that underlying classical fast matrix-vector multiplication schemes, and has some connection with sparse direct solvers. Time permitting, we will also touch on similar investigations for oscillatory integral equations, for which much remains to be discovered. This is joint work with Leslie Greengard. SUMMARY:A Fast Direct Solver for Non-Oscillatory Integral Equations UID:1560 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120117T103000 DTEND;TZID=America/Chicago:20120117T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, Conference Room, 4301, Argonne National Laboratory DESCRIPTION:The massive amounts of data that are generated by large-scale simulations and supercomputers have exceeded our technological capacity to store it and our cognitive capacity to understand it. A solution to manage the data deluge is /in situ/ visualization and analysis, which stores smaller processed data by integrating analysis and visualization with simulations. Since most large-scale simulations are batch processes without a human- in-the-loop, /in situ/ analysis can miss discoveries achieved through interactive data exploration. To try to fill this discovery role, we extend /in situ/ analysis by adding feature and event detection algorithms to trigger visualization and analysis data pipelines. Triggering data processing pipelines for large-scale simulations is conceptually similar to high-energy physics analysis codes searching for events in accelerator experiments. We will discuss our previous work in feature and detection algorithms to support event-driven in situ analysis, various use-case scenarios, and our on-going /in situ/ analysis with DOE simulation codes.\n SUMMARY:Extending In Situ Analysis through Feature and Event Detectors UID:1562 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120201T150000 DTEND;TZID=America/Chicago:20120201T160000 DTSTAMP:20130525T020110 LOCATION:Building 240, 4301, Argonne National Laboratory DESCRIPTION:In statistics we often have some a priori information or assumptions on the qualitative properties of the estimator, including constraints on the shape and support of the estimator. For instance, we may require that the estimator be increasing or decreasing, convex or concave, or unimodal over an interval, be bounded from above or below, or some combination of these. In this talk we shall discuss how to incorporate such information as hard constraints into the optimization model defining the estimator when the estimator is a polynomial spline. We show that in the univariate case these shape-constrained estimators can be computed in polynomial time. Examples from density estimation, monotone and convex regression, and arrival rate estimation will be presented. Time permitting, we shall discuss the theoretical complications that arise in the multivariate case, and present a few multivariate applications as well. SUMMARY:Statistical estimation with shape constraints UID:1564 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120118T103000 DTEND;TZID=America/Chicago:20120118T113000 DTSTAMP:20130525T020110 LOCATION:TCS Conference Center, Rms 1404 and 1405, Argonne National Laboratory DESCRIPTION:Using derivative based numerical optimization routines to solve optimization problems governed by partial differential equations (PDE) with uncertain coefficients is computationally expensive due to the large number of PDE solves required at each iteration. I propose a framework for the adaptive solution of such optimization problems based on the retrospective trust region algorithm. I prove global convergence of the retrospective trust region algorithm under weak assumptions on gradient inexactness. If one can bound the error between actual and modeled gradients using reliable and efficient a posteriori error estimators, then the global convergence of the proposed algorithm follows. I present a stochastic collocation finite element method for the solution of the PDE constrained optimization problems. In the stochastic collocation framework, the state and adjoint equations can be solved in parallel. Initial numerical results for the adaptive solution of these optimization problems are presented. SUMMARY:An Approach for the Adaptive Solution of Optimization Problems Governed by PDEs with Uncertain Coefficients UID:1584 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120125T150000 DTEND;TZID=America/Chicago:20120125T160000 DTSTAMP:20130525T020110 LOCATION:Building 240, 4301, Argonne National Laboratory DESCRIPTION:A vast array of scientific problems are formulated on networks (i.e. graphs), which show the relationships between entities in areas such as cyber-security, material science, biology and social sciences. Frequently, obtaining the required scientific data about the networks is expensive or infeasible. Moreover, when developing network-based algorithms we often cannot know in advance the input network. In other words, the amount of available real data may be insufficient for simulations, verification, benchmarking, etc. In these situations, synthetic data should be generated from models but the existing network models are highly-stylized and poorly describe most of the real systems. We will present a novel strategy for network generation using multiscale network editing. Our tool, termed MUSKETEER uses this process to create high-fidelity artificial networks. Joint work with Ilya Safro. SUMMARY:Multiscale method for network generation UID:1568 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120123T013000 DTEND;TZID=America/Chicago:20120123T023000 DTSTAMP:20130525T020110 LOCATION:Building 240, Seminar Room 4301, Argonne National Laboratory DESCRIPTION:Electronic structure theory concerns the description of molecular properties according to the postulates of quantum mechanics. For practical purposes, this is realized entirely through numerical computation, the scope of which is constrained by computational costs that increases rapidly with the size of the system.\n\nThe signifi cant progress made in this field over the past decades have been facilitated in part by the willingness of chemists to forego some mathematical rigour in exchange for greater efficiency. While such compromises allow large systems to be computed feasibly, there are lingering concerns over the impact that these compromises have on the quality of the results that are produced. This research is motivated by two key issues that contribute to this loss of quality, namely i) the numerical errors accumulated due to the use of nite precision arithmetic and the application of numerical approximations, and ii) the reliance on iterative methods that are not guaranteed to converge to the correct solution. SUMMARY:Rigorous Numerical Approaches in Electronic Structure Theory UID:1588 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120126T103000 DTEND;TZID=America/Chicago:20120126T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, Seminar Room 4301, Argonne National Laboratory DESCRIPTION:At the cellular scale, many biological processes are driven by random interactions between molecules present in small populations. These systems are inherently stochastic and often display behavior that cannot be represented with traditional deterministic approaches. Gillespie\'s Stochastic Simulation Algorithm (SSA) provides a method for simulating spatially homogeneous biochemical systems in a way that captures this randomness. Since Gillespie\'s pioneering work, there have been several advances in efficient exact and approximate variants of the SSA. In this talk, I will provide an overview of the SSA and discuss StochKit2, a software package that provides efficient implementations of several of the most mature SSA variants. Then I will present my recent work on the development of a multiscale SSA implementation that exploits the stochastic partial equilibrium and quasi steady-state approximations. SUMMARY:Multiscale Discrete Stochastic Simulation of Biochemical Systems UID:1592 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120123T103000 DTEND;TZID=America/Chicago:20120123T113000 DTSTAMP:20130525T020110 LOCATION:Building 240 / Room 4301, Argonne National Laboratory DESCRIPTION:This seminar is in line with the enhancement of our capacities to perform massively parallel CFD simulation and to compress further the time-to-solution for challenging problems in science and engineering. In this context, the implementation of our fully implicit adaptive stabilized finite element flow solver called PHASTA will be first presented. Two unique assets characterize this flow solver. The first one is its ability to perform anisotropic adaptivity of the initial 3D unstructured finite element mesh, which is required for complex geometries. The second one is related to its strong scaling performance, as it has been shown to scale up to 95% on 288K cores of the Jugene BG/P supercomputer. In addition to the parallelism of our domain decomposition approach, the improvement of our I/O capabilities will also be emphasized. Subsequently, our recent progress in co-visualization will be summarized, whereby a live data analysis is able to provide continuous and reconfigurable insight into massively parallel simulations, paving the way for interactive simulation and simulation steering. Specifically, we demonstrated our co-visualization concept of either the full data or in situ data extracts on 160K cores of the ALCF Intrepid BG/P supercomputer tightly linked through a high-speed network to 100 visualization nodes of the Eureka system which share 800 cores and 200 GPUs.\n\nClick on the link below to add this seminar to your calendar. SUMMARY:Petascale Adaptive Computational Fluid Dynamics: Implementation and Key Ingredients for Time-To-Solution Compression UID:1594 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120123T080000 DTEND;TZID=America/Chicago:20120123T000008 DTSTAMP:20130525T020110 LOCATION:Building 240, Seminar Room 4301, Argonne National Laboratory DESCRIPTION:Electronic structure theory concerns the description of molecular properties according to the postulates of quantum mechanics. For practical purposes, this is realized entirely through numerical computation, the scope of which is constrained by computational costs that increases rapidly with the size of the system.\n\nThe signifi cant progress made in this field over the past decades have been facilitated in part by the willingness of chemists to forego some mathematical rigour in exchange for greater efficiency. While such compromises allow large systems to be computed feasibly, there are lingering concerns over the impact that these compromises have on the quality of the results that are produced. This research is motivated by two key issues that contribute to this loss of quality, namely i) the numerical errors accumulated due to the use of nite precision arithmetic and the application of numerical approximations, and ii) the reliance on iterative methods that are not guaranteed to converge to the correct solution.\n\nTaking the above issues in consideration, the aim of this thesis is to explore ways to perform electronic structure calculations with greater mathematical rigour, through the application of rigorous numerical methods. Of which, we focus in particular on methods based on interval analysis and deterministic global optimization. The Hartree-Fock electronic structure method will be used as the subject of this study due to its ubiquity within this domain.\n\nWe outline an approach for placing rigorous bounds on numerical error in Hartree-Fock computations. This is achieved through the application of interval analysis techniques, which are able to rigorously bound and propagate quanti-ties aff ected by numerical errors. Using this approach, we implement a program called Interval Hartree-Fock. Given a closed-shell system and the current elec-tronic state, this program is able to compute rigorous error bounds on quantities including i) the total energy, ii) molecular orbital energies, iii) molecular orbital coefficients, and iv) derived electronic properties.\n\nInterval Hartree-Fock is adapted as an error analysis tool for studying the impact of numerical error in Hartree-Fock computations. It is used to investi-gate the e ect of input related factors such as system size and basis set types on the numerical accuracy of the Hartree-Fock total energy. Consideration is also given to the impact of various algorithm design decisions. Examples include the application of diffi erent integral screening thresholds, the variation between sin- gle and double precision arithmetic in two-electron integral evaluation, and the adjustment of interpolation table granularity. These factors are relevant to both the usage of conventional Hartree-Fock code, and the development of Hartree-Fock code optimized for novel computing devices such as graphics processing units.\n\nWe then present an approach for solving the Hartree-Fock equations to within a guaranteed margin of error. This is achieved by treating the Hartree-Fock equations as a non-convex global optimization problem, which is then solved using deterministic global optimization. The main contribution of this work is the development of algorithms for handling quantum chemistry specifi c expressions such as the one and two-electron integrals within the deterministic global optimization framework. This approach was implemented as an extension to an existing open source solver.\n\nProof of concept calculations are performed for a variety of problems within Hartree-Fock theory, including those in i) point energy calculation, ii) geome-try optimization, iii) basis set optimization, and iv) excited state calculation. Performance analyses of these calculations are also presented and discussed.\n SUMMARY:Rigorous Numerical Approaches in Electronic Structure Theory UID:1590 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:00000000T103000 DTEND;TZID=America/Chicago:00000000T113000 DTSTAMP:20130525T020110 LOCATION:Building 240 / Room 4301, Argonne National Laboratory DESCRIPTION:This seminar is in line with the enhancement of our capacities to perform massively parallel CFD simulation and to compress further the time-to-solution for challenging problems in science and engineering. In this context, the implementation of our fully implicit adaptive stabilized finite element flow solver called PHASTA will be first presented. Two unique assets characterize this flow solver. The first one is its ability to perform anisotropic adaptivity of the initial 3D unstructured finite element mesh, which is required for complex geometries. The second one is related to its strong scaling performance, as it has been shown to scale up to 95% on 288K cores of the Jugene BG/P supercomputer. In addition to the parallelism of our domain decomposition approach, the improvement of our I/O capabilities will also be emphasized. Subsequently, our recent progress in co-visualization will be summarized, whereby a live data analysis is able to provide continuous and reconfigurable insight into massively parallel simulations, paving the way for interactive simulation and simulation steering. Specifically, we demonstrated our co-visualization concept of either the full data or in situ data extracts on 160K cores of the ALCF Intrepid BG/P supercomputer tightly linked through a high-speed network to 100 visualization nodes of the Eureka system which share 800 cores and 200 GPUs.\n\nClick on the link below to add this seminar to your calendar. SUMMARY:Petascale Adaptive Computational Fluid Dynamics: Implementation and Key Ingredients for Time-To-Solution Compression UID:1596 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120127T103000 DTEND;TZID=America/Chicago:20120127T113000 DTSTAMP:20130525T020110 LOCATION:TCS Conference Center, Rms 1404 and 1405, Argonne National Laboratory DESCRIPTION:A systematic approach is presented for constructing higher-order embedded boundary methods for solving partial differential equations (PDE) with dynamic boundary conditions in general, and fluid-structure interaction (FSI) problems in particular. Such methods are gaining popularity because they simplify a number of computational issues. These range from gridding the fluid domain, to designing Eulerian-based algorithms for challenging fluid structure applications characterized by large structural motions and deformations. However, because they typically operate on non body-fitted meshes, embedded boundary methods also complicated other issues such as treatment of wall boundary conditions in general, and fluid-structure transmission condition in particular. These methods also tend to be at best first-order space-accurate at the embedded boundaries; and in some cases they are provably inconsistent at these locations.\n\nTo address this issue, the proposed methodology leads to a departure from the current practice of populating ghost fluid values independently from the chosen spatial discretization scheme. Instead, it accounts for the pattern and properties of a preferred higher-order discretization scheme, and attributes ghost values as to preserve the formal order of spatial accuracy of this scheme. The methodology is described with assumption of prescribed moving wall boundary conditions, however its extension to flow-induced structural motions is straightforward. It is illustrated by its application to various finite difference and finite volume methods. Its impact is also demonstrated by multi-dimensional numerical experiments that confirm its theoretically proven ability to preserve higher-order spatial accuracy, including in the vicinity of the embedded interfaces. SUMMARY:A Systematic Approach for Constructing Higher-Order Embedded Boundary Methods and Its Application to Inviscid Compressible Flow in Fluid-Structure Int UID:1600 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120131T103000 DTEND;TZID=America/Chicago:20120131T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, Seminar Room 4301, Argonne National Laboratory DESCRIPTION:Reduced-order models are frequently used in the simulation of complex flows to overcome the high computational cost of direct numerical simulations, especially for three dimensional nonlinear problems. Proper orthogonal decomposition is one of the most commonly used methods to generate reduced-order models for turbulent flows dominated by coherent structures. To balance the low computational cost required by a reduced-order model and the complexity of the targeted turbulent flows, appropriate closure modeling strategies need to be employed. In this talk, we will present several new nonlinear closure methods for proper orthogonal decomposition reduced-order models. We will also present numerical results for the new models used in realistic applications such as uncertainty quantification in nuclear engineering, energy efficient building design and control, and climate modeling. SUMMARY:Reduced-Order Models of Complex Flows: Modeling, Analysis and Computations UID:1602 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120203T103000 DTEND;TZID=America/Chicago:20120203T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, Seminar Room 4301, Argonne National Laboratory DESCRIPTION:Accurate simulation of carrier transport in semiconductor devices requires the solution of the Boltzmann transport equation (BTE), for which a deterministic solution approach based on a spherical harmonics expansion (SHE) is considered. The SHE method leads to a system of partial dierential equations, which is solved numerically using a nite volume scheme. First, recent algorithmic improvements of the SHE method are discussed. Then, the components of the library-centric C++ simulation framework created for the simulation of state-of-the-art semiconductor devices using the SHE method are presented: ViennaGrid provides a exible grid management and allows for dimension-independent programming. ViennaData separates data storage from the grid data structure and allows for associating quantities of arbitrary type with grid elements. ViennaCL provides GPU-accelerated iterative solvers and preconditioners for the solution of the sparse linear systems of equations obtained from the SHE method. The interaction of these free open-source libraries in the device simulator Vienna SHE is outlined. Finally, a brief overview of current and future research activities is given. SUMMARY:Meeting the Many Challenges of Solving the Boltzmann Transport Equation UID:1606 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120130T130000 DTEND;TZID=America/Chicago:20120130T140000 DTSTAMP:20130525T020110 LOCATION:building 240/ Conference Room 4301, Argonne National Laboratory DESCRIPTION:Evolutionary relationships between a group of organisms are commonly summarized in a phylogenetic tree. The goal of phylogenetic inference is to infer the best tree structure that represents the relationships between a group of organisms, given a series of observations (e.g., molecular sequences). However, the most popular heuristics for inferring phylogenetic trees tend to output tens to hundreds of thousands of equally weighted trees. Biologists commonly encapsulate these trees into a single tree, which is referred to as the consensus. The assumption for building a consensus tree is that the information discarded is less important than the information retained. But, what if this assumption is not true?\n\nIn this talk, I will present compelling evidence that retaining and studying tree collections is valuable. Due to the burgeoning nature of phylogenetic analysis, high performance algorithms are needed to allow biologists to easily study and share tree collections. I will also describe two of my algorithms, MrsRF and TreeZip, which are the fastest and most efficient methods for comparing and storing trees. Lastly, I will briefly describe PVCS, a novel version control system that will make it easier for biologists to easily retain and share the results of their phylogenetic analyses. SUMMARY:\"Fast Algorightms for Exploring Large Collections of Phylogenetic Trees\" UID:1608 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120130T130000 DTEND;TZID=America/Chicago:20120130T140000 DTSTAMP:20130525T020110 LOCATION:building 240/ Conference Room 4301, Argonne National Laboratory DESCRIPTION:Evolutionary relationships between a group of organisms are commonly summarized in a phylogenetic tree. The goal of phylogenetic inference is to infer the best tree structure that represents the relationships between a group of organisms, given a series of observations (e.g., molecular sequences). However, the most popular heuristics for inferring phylogenetic trees tend to output tens to hundreds of thousands of equally weighted trees. Biologists commonly encapsulate these trees into a single tree, which is referred to as the consensus. The assumption for building a consensus tree is that the information discarded is less important than the information retained. But, what if this assumption is not true?\n\nIn this talk, I will present compelling evidence that retaining and studying tree collections is valuable. Due to the burgeoning nature of phylogenetic analysis, high performance algorithms are needed to allow biologists to easily study and share tree collections. I will also describe two of my algorithms, MrsRF and TreeZip, which are the fastest and most efficient methods for comparing and storing trees. Lastly, I will briefly describe PVCS, a novel version control system that will make it easier for biologists to easily retain and share the results of their phylogenetic analyses. SUMMARY:\"Fast Algorightms for Exploring Large Collections of Phylogenetic Trees\" UID:1610 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:00000000T103000 DTEND;TZID=America/Chicago:00000000T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, Seminar Room 4301, Argonne National Laboratory DESCRIPTION:This talk is concerned with preconditioning the system matrix resulting from finite element discretizations of time harmonic wave equations, such as Maxwell\'s equations in the frequency domain or the Helmholtz equation. The moving PML sweeping preconditioner, first introduced on a Cartesian finite difference grid, is generalized to an unstructured mesh with finite elements. The method dramatically reduces the number of GMRES iterations necessary for convergence, resulting in an almost linear complexity solver. Numerical examples including electromagnetic cloaking simulations are presented to demonstrate the efficiency of the proposed method. SUMMARY:Sweeping Preconditioners for Finite Element Methods in Time-Harmonic Wave Propagation UID:1612 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:00000000T103000 DTEND;TZID=America/Chicago:00000000T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, Seminar Room 4301, Argonne National Laboratory DESCRIPTION:Time-harmonic waves are observed in many physical settings, most notably in acoustics, electromagnetics, and elasticity. As the frequency increases, the numerical solution of such problems becomes increasingly difficult, due to the large number of degrees of freedom required to resolve each oscillation. The memory requirements and asymptotic complexity of direct solvers are too costly for these problems, so iterative solvers must be used. However, most iterative methods have their own numerical difficulties; for boundary element methods, the dense matrix-vector multiplication needed at each iteration is very costly, while for finite element methods, the indefiniteness and ill-conditioning of the matrix destroy convergence. In this talk, I will present fast algorithms and preconditioners which allow the convergence of Krylov subspace iterative solvers for high-frequency problems in a reasonable amount of time. SUMMARY:Fast Numerical Algorithms for Frequency-Domain Wave Propagation UID:1614 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120208T150000 DTEND;TZID=America/Chicago:20120208T160000 DTSTAMP:20130525T020110 LOCATION:Building 240, 1404-1405, Argonne National Laboratory DESCRIPTION:John T. Murphy, Computational Postdoctoral Fellow, Decision and Information Sciences Division (DIS) and ALCF, will present an introduction to his work on the Hohokam Water Management Simulation II, a simulation (in progress) of an archaeologically attested irrigation society that flourished and then declined in central and southern Arizona over roughly 1,000 years, from approximately 400 to 1450 AD. Technical aspects of large-scale archaeological simulation (especially in high-performance computing environments) will be presented against a background of the current field of computational social science and the use of agent-based models to understand complex social and coupled human-natural systems. SUMMARY:Pasts that Never Happened: Archaeological Simulation Models and Computational Social Science UID:1616 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120210T133000 DTEND;TZID=America/Chicago:20120210T143000 DTSTAMP:20130525T020110 LOCATION:Building 240 / TCS Conference Room 1404, Argonne National Laboratory DESCRIPTION:Fotis Sotiropoulos is the James L. Record professor of Civil Engineering, the director of the St. Anthony Falls Laboratory, and the director of the Department of Energy funded EOLOS wind energy research consortium at the University of Minnesota. His research centers on computational techniques for studying a broad range of interdisciplinary fluid mechanics problems in environmental hydrodynamics, wind and water power, cardiovascular hemodynamics, and aquatic swimming. His work is sponsored by NSF, NIH, DOE and the private industry. He is a fellow of the American Physical Society, recipient of a NSF Career award, and serves as an associate editor for the ASME Journal of Biomechanical Engineering and the ASCE Journal of Hydraulic Engineering. SUMMARY:Tackling Fluid Mechanics Challenges in Energy, Environment & Health via Numerical Simulation: From Cardiovascular Hemodynamics to Wind & Water Power UID:1618 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120206T100000 DTEND;TZID=America/Chicago:20120206T110000 DTSTAMP:20130525T020110 LOCATION:Building 240/Conference Room 1407, Argonne National Laboratory DESCRIPTION:Carbon monoxide binds tightly to many hemes and other porphyrins, proving toxic to humans and microbes. Nevertheless, CO is an attractive fuel for anaerobic chemolithotrophy for those microbes not impaired by COs toxicity. CO can be oxidized to CO2 anaerobically through the water-gas-shift reaction by hydrogenogenic carboxydotrophs. CO can also be used by certain homoacetogens in the production of acetate and by certain methanogens in the production of methane, acetate, and formate.\n\nCO is not only consumed by microbes, but a diverse range of evidence points towards biogenic production of CO. We present measurements of CO concentrations in dissolved and free phase gases in hot springs in Kamchatka, Russia, and Lassen Volcanic National Park, CA. These measurements implicate microbial CO production as a dominant source of CO in these anaerobic ecosystems. Biogenic CO production has important repercussions for the chemistry of the atmosphere and climate, especially on the ancient Earth.\n\nThere are few in situ measurements of dissolved CO in microbial mats and sediment pore waters. However, the concentration dependent biochemical and transcriptional responses of carboxydotrophs to CO offer strong evidence for the growth conditions to which these microbes have adapted. CooA is a CO-sensing transcriptional activator found in most anaerobic carboxydotrophs. We interpret the CO-binding properties of CooA in Carboxydothermus hydrogenoformans and Rhodospirillum rubrum as evidence for the ranges of environmental CO concentrations that are frequently encountered in anaerobic ecosystems. \n SUMMARY:\"Carbon monoxide based metabolism, CO cycling in anaerobic microbial ecosystems, and the composition of early Earths atmosphere\" UID:1624 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120321T150000 DTEND;TZID=America/Chicago:20120321T160000 DTSTAMP:20130525T020110 LOCATION:Building 240, 1404-1405, Argonne National Laboratory DESCRIPTION:We developed a parallel O(NlogN) adaptive treecode for microstructural computations in 2D. The code is tested first with randomly generated data for accuracy and time complexity, and then in a boundary integral method for evolution of microstructure in elastic media. In problems involving multiple precipitates, the number of computational markers N can be large, and an O(N^2) direct summation method becomes computationally prohibitive. It is shown that the serial version of the code is of time complexity O(NlogN), and at the same time fulfills stringent precision requirements prescribed by the spectrally accurate scheme in the boundary integral method. SUMMARY:An adaptive treecode for evolution of microstructure in elastic media UID:1622 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120217T103000 DTEND;TZID=America/Chicago:20120217T113000 DTSTAMP:20130525T020110 LOCATION:Building 240 / TCS Conference Room 1404-05, Argonne National Laboratory DESCRIPTION:Due to the infamous “memory wall” problem and a drastic increase in the number of data intensive applications, memory rather than processor has become the leading performance bottleneck of modern computing systems. Evaluating and understanding memory system performance is increasingly becoming the core of high-end computing. Conventional memory metrics, such as miss ratio, average miss latency, average memory access time, etc., are designed to measure a given memory performance parameter,\nand do not reflect the overall performance of a memory system. On the other hand, widely used system metrics, such as IPC and Flops are designed to measure CPU performance, and do not directly reflect memory performance. In this talk, a novel memory metric, Access Per Cycle (APC) is proposed to measure overall memory performance with consideration of the complexity of modern memory systems. A unique contribution of APC is its separation of memory evaluation from CPU evaluation; therefore, it provides a quantitative measurement of the “data-intensiveness” of an application. Simulation results show that APC is significantly more appropriate than the existing memory metrics in evaluating modern memory systems. After the introduction of APC, some previous works of the speaker’s about large-scale networks will be briefly presented. These works include large-scale network switch design considerations, computing network, and network fault tolerance. SUMMARY:A Performance Metric of Memory Systems UID:1620 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120210T103000 DTEND;TZID=America/Chicago:20120210T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, TCS Conference Center, 1406 & 1407, Argonne National Laboratory DESCRIPTION:Understanding applications\' behaviors and their interactions with system software and hardware is becoming increasingly difficult as the complexity of all three components increases. Thus, tools for understanding these in the contexts of both failure and performance are becoming more important. In the case of failure, early detection and attribution can increase productivity of both platform and user through the ability to quickly respond. In the context of performance, understanding how resources are being used can again drive increased productivity through more intelligent resource requests, allocations, and use. This talk will present work being done at Sandia on scalable lightweight tools for HPC monitoring and analysis of all three components as well as for feedback to drive application load balancing.\n\nJim Brandt is a Principal Member of Technical Staff at Sandia National Laboratories. His research interest is in strategies and enabling capabilities for intelligent dynamic resource management for improved HPC system and application performance. His work targets both failure and non-failure (e.g., memory contention) scenarios. Jim leads the OVIS project at Sandia for scalable, real-time analysis of very large datasets, targeting the analysis of HPC system data to characterize system health and application resource utilization and to determine and invoke beneficial response. SUMMARY:Scalable, flexible tools for understanding the HPC environment UID:1626 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120222T103000 DTEND;TZID=America/Chicago:20120222T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, 1404-1405, Argonne National Laboratory DESCRIPTION:We propose GROPHECY, a GPU performance projection framework that can estimate the performance benefit of GPU acceleration without actual GPU programming or hardware. Users need only skeletonize pieces of CPU code that are targets for GPU acceleration. Code skeletons are automatically transformed in various ways to mimic tuned GPU codes with characteristics resembling real implementations. The synthesized characteristics are used by an existing analytical model to project GPU performance. The cost and benefit of GPU development can then be estimated according to the transformed code skeleton that yields the best projected performance. With GROPHECY, users can leap toward GPU acceleration only when the cost-benefit makes sense. The framework is validated using kernel benchmarks and data-parallel codes in legacy scientific applications. The measured performance of manually tuned codes deviates from the projected performance by 17% in geometric mean. Extensions to GROPHECY include fusing multiple kernels to reduce data movement and modeling the data transfer overhead between CPU and GPU. SUMMARY:GROPHECY: GPU performance projection from CPU code skeletons UID:1628 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120229T150000 DTEND;TZID=America/Chicago:20120229T160000 DTSTAMP:20130525T020110 LOCATION:Building 240, 1404-1405, Argonne National Laboratory DESCRIPTION:Swift lets you write parallel scripts that run many copies of ordinary programs concurrently. In this talk we will discuss the Swift coasters framework and its usage to leverage the computational power of bag of workstations commonly found in universities and institutions. We will see a short demonstration on how to exploit the computational resources of MCS workstations with a simple application. SUMMARY:Parallel computation on bag of workstations using Swift coasters UID:1630 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120411T150000 DTEND;TZID=America/Chicago:20120411T160000 DTSTAMP:20130525T020110 LOCATION:Building 240, 4301, Argonne National Laboratory DESCRIPTION:In this talk, we will present our current work on a novel parallelization of the revised simplex method for large deterministic-equivalent forms of stochastic linear-programming (LP) problems. These problems have been considered too large to solve by using the simplex method, and instead, decomposition approaches based on row generation or, more recently, interior-point methods are generally used. However, these approaches do not provide optimal basic solutions, which allow for efficient hot-starts and can provide important sensitivity information. Our approach exploits the dual block-angular structure of these problems inside the linear algebra of the revised simplex method in a data-parallel manner suitable for high-performance distributed-memory clusters or supercomputers. While our focus is on stochastic LPs, the work is applicable to all problems with a dual block-angular structure. Our implementation is competitive in serial with highly-efficient sparsity-exploiting simplex codes and achieves significant relative speed-ups when run in parallel. Additionally, very large problems with hundreds of millions of variables have been successfully solved to optimal bases. SUMMARY:Parallel distributed-memory simplex for large-scale stochastic LP problems UID:1642 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120401T080000 DTEND;TZID=America/Chicago:20120401T170000 DTSTAMP:20130525T020110 LOCATION:Blood Research Institute, Milwaukee, WI DESCRIPTION:Ian Foster and Rachana Ananthakrishnan will be discussing GlobusOnline\'s available tools to simplify the sharing of distributed resources and data. \n\nThe research community\'s need for progressively granular results has driven the development of high-performance computing technologies. User-friendly interfaces and research gateways facilitate access by all domains of science, arts and humanities research, and education applications. \n\nThe amount of data produced can quickly overwhelm researchers and their service providers who, in many cases, are not funded to manage the unanticipated demand. Library informatics professionals, information technology service providers, and funding agents face unprecedented, and in some cases legislated, data lifecycle management challenges.\n\nThe institutions that comprise SeWHiP (the Southeast Wisconsin High Performance Cyberinfrastructure) invite you to participate in a one-day symposium to explore the challenges associated with storage, access, visualization, sharing, integration, and scaling of research data. Experts from U.S. agencies, national laboratories, and research institutes will present storage options for research communities, federated solutions and tools, visualization and data analysis highlights, and NSF/NIH funding agency considerations. SUMMARY:Discover Solutions to the Challenges of Data Intensive Research UID:1632 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120223T150000 DTEND;TZID=America/Chicago:20120223T160000 DTSTAMP:20130525T020110 LOCATION:Building 240, 4301, Argonne National Laboratory DESCRIPTION:In this talk, I will start by giving some background on my past research projects and how they connect to my current research. I will focus on the dynamic binary translation project that I worked on for my PhD thesis. The rest of the talk will be about my current research project, “10x10”. 10x10 is motivated by two current trends in computer architecture - a power wall limiting the energy consumption in processors and transistor scaling characteristics. With every generation, transistor area is decreasing by a factor of two, allowing more micro-architectural features to be packed into the same chip area and faster signal routing. However, transistor power consumption is decreasing at a slower rate, leading to an overall increase in power consumption. Therefore, all transistors on a chip cannot be switched at full frequency at the same time to avoid hitting the power wall. This fact has led to the adoption of multi-cores (operated at relatively low frequency) and many-cores (e.g. GPUs). However, even these architectures fall short of exascale-era computing requirements, because of insufficient energy scaling. Our solution for power and performance scaling stems from the insight that tailoring architectures to their target application set can lead to huge improvements. Homogeneous cores waste energy in powering logic and data movement that is not essential to the computation. Therefore, we are designing customized architectures for general-purpose workloads. We envision a chip that is an ensemble of heterogeneous cores, each tailored to certain application characteristics. These application characteristics may include parallelism type, memory access pattern, communication pattern, compute intensity, operation and datatype mix etc.\n\nThere are three fundamental challenges to this research. The first one is that general-purpose workloads span a large set of applications. We need to show that general-purpose workloads can be grouped based on architecturally exploitable features and that these groups are manageable in number, exhibit architecturally coherent features and are distinct. The second challenge is to design heterogeneous cores targeting application features. Chip area and data movement between cores are constraints in this challenge. The final challenge is programming for heterogeneous systems. While programming for multi-cores and many-cores is still viewed as a complex problem, heterogeneous cores take the problem one level further. Now it is necessary to first identify the core to target code for, before actually optimizing and generating code. I will present in-depth research on the first challenge and a discussion on the approaches that we are pursuing for the other challenges.\n\n SUMMARY:10x10: A Customized Approach to Energy-Efficient Execution UID:1640 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20110314T150000 DTEND;TZID=America/Chicago:20110314T160000 DTSTAMP:20130525T020110 LOCATION:Building 240, 1404-1405, Argonne National Laboratory DESCRIPTION:We present filling as a new type of spatial subdivision problem that is related to covering and packing. Filling addresses the optimal placement of overlapping objects lying entirely inside an arbitrary shape so as to cover the most interior volume. In $n$-dimensional space, if the objects are polydisperse $n$-balls, we show that solutions correspond to sets of maximal $n$-balls and the solutions space can reduced to the medial axis of a shape. We consider the structure of the solution space in two-dimensions. For polygons, we provide a detailed description of a heuristic and genetic algorithm for finding solutions of maximal discs. We also consider the properties of ideal distributions of $N$ discs in polygons as N goes to infinity. SUMMARY:Optimal Filling of Shapes UID:1644 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120314T150000 DTEND;TZID=America/Chicago:20120314T160000 DTSTAMP:20130525T020110 LOCATION:Building 240, 1404-1405, Argonne National Laboratory DESCRIPTION:We present filling as a new type of spatial subdivision problem that is related to covering and packing. Filling addresses the optimal placement of overlapping objects lying entirely inside an arbitrary shape so as to cover the most interior volume. In $n$-dimensional space, if the objects are polydisperse $n$-balls, we show that solutions correspond to sets of maximal $n$-balls and the solutions space can reduced to the medial axis of a shape. We consider the structure of the solution space in two-dimensions. For polygons, we provide a detailed description of a heuristic and genetic algorithm for finding solutions of maximal discs. We also consider the properties of ideal distributions of $N$ discs in polygons as $N$ goes to infinity.\n SUMMARY:Optimal Filling of Shapes UID:1646 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120301T103000 DTEND;TZID=America/Chicago:20120301T113000 DTSTAMP:20130525T020110 LOCATION:TCS Conference Center, Rms 1404 and 1405, Argonne National Laboratory DESCRIPTION:We propose a new adaptive mesh refinement (AMR) strategy, the goal is to reach a given error tolerance with the least amount of computational cost. This strategy is especially attractive in the setting of a first-order system least squares (FOSLS) finite element formulation in conjunction with algebraic multigrid (AMG) methods in the context of nested iteration (NI). To accomplish this, the refinement decisions are determined based on minimizing the predicted accuracy-per-computational-cost efficiency (ACE). The nested iteration approach produces a sequence of refinement levels in which the error is equally distributed across elements on a relatively coarse grid. Once the solution is numerically resolved, refinement becomes nearly uniform. \n\nThis talk will first describe the algorithm and demonstrate its efficiency on a simple test problem. Then, modifications that yield an efficient parallel algorithm will be discussed. These involve a geometric binning strategy to reduce communication cost. Load balancing begins at coarse levels using a parallel quad-tree and a space filling curve. We show that this automatically ameliorates load balancing issues at finer levels. \n\nNumerical results are presented for various examples including a 2D Poisson problem with steep gradients and a 2D reduced model of the incompressible, resistive magnetohydrodynamic equations (MHD). We show that, by using the NI-FOSL-AMG-ACE approach, we are able to resolve the physics features using a small number of work units. Using Frost, The NCAR/CU Blue Gene/L supercomputer, we demonstrate excellent weak and strong scalability up to 4,096 processor for problems with up to 15 million elements. SUMMARY:Parallel Efficiency-based Adaptive Mesh Refinement for FOSLS-AMG UID:1652 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120307T103000 DTEND;TZID=America/Chicago:20120307T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, Seminar Room 4301, Argonne National Laboratory DESCRIPTION:The application of adaptive finite elements for reliable numerical simulations requires they be executed in an automated manner with explicit control of the approximations made. In these applications, meshes are used to discretize the problem domains. Since there are no reliable a priori methods to control the approximation errors, adaptive methods must be applied. Adaptive meshing procedures provide a powerful tool for attacking problems such as fluid flows that can develop highly anisotropic solutions and can only be located and resolved through adaptivity. A mesh modification based procedure has been developed that accepts an anisotropic mesh size field defined using entity level correction indicators evaluated on the previous solution step. The mesh size field drives mesh refinement and coarsening operations on the mesh to yield a mesh that satisfies the requested mesh size field. SUMMARY:Parallel Anisotropic Mesh Adaptation with Boundary Layers to Support Large-Scale Scientific Applications UID:1654 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120319T110000 DTEND;TZID=America/Chicago:20120319T000008 DTSTAMP:20130525T020110 LOCATION:Building 240, Conference Room 1404 , Argonne National Laboratory DESCRIPTION:In present-day climate models, subgrid-scale clouds and turbulence in the boundary layer are parameterized using different schemes for different cloud regimes. A mass-flux scheme is used to parameterize shallow cumulus clouds, an eddy-diffusivity scheme is used to parameterize subgrid-scale turbulent fluxes, and a macrophysics scheme\nis used to parameterize stratiform clouds. Each of these three schemes has its own drawbacks, and the framework itself of using separate schemes for separate regimes is problematic.\n\nHere, we discuss an alternative framework for parameterizing clouds and turbulence. It is a unified approach in which all boundary layer clouds and turbulence are modeled using the same equation set. It involves estimating the probability density function (PDF) of subgrid\nvariability in moisture, heat content, and turbulence, and then using the PDF to diagnose relevant quantities such as cloud fraction and liquid water.\n\nWe have implemented and tested our PDF parameterization in two climate models, GFDL\'s AM3 and NCAR\'s CAM model. We will discuss the merits and deficiencies of our current climate simulations based on our PDF parameterization.\n SUMMARY:An alternative framework for parameterizing clouds and turbulence in climate models UID:1656 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120314T103000 DTEND;TZID=America/Chicago:20120314T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, Seminar Room 4301, Argonne National Laboratory DESCRIPTION:In the presentation, I will talk about different discretization methods for the diffusion equation on polyhedral meshes. I will start with a brief discussion of several well-known approaches, such as finite volumes and divconst. In the main part of the talk, I will speak about new recently developed approach called piece-wise constant approximation. This method has some interesting properties, such as low computational cost and good treatment of thin layers, which makes it suitable for applications arising, for instance, from oil industry. I will show the application of the method to polyhedral meshes consisting of hexahedrons and degenerated hexahedrons, such as prisms, and will demonstrate its competitiveness. SUMMARY:Discretizations of The Diffusion Equation on Polyhedral Meshes. UID:1658 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120319T150000 DTEND;TZID=America/Chicago:20120319T160000 DTSTAMP:20130525T020110 LOCATION:Building 240, 1404-1405, Argonne National Laboratory DESCRIPTION:Two multilevel frameworks for manifold learning algorithms are discussed which are based on an affinity graph whose goal is to sketch the neighborhood of each sample point. One framework is geometric and is suitable for methods aiming to find an isometric or a conformal mapping, such as isometric feature mapping (Isomap) and semidefinite embedding (SDE). The other is algebraic and can be incorporated into methods that preserve the closeness of neighboring points, such as Locally Linear Embedding (LLE) and Laplacian eigenmaps (LE). The multilevel coarsening technique presented in this paper can be applied to both directed and undirected graphs. It is based on the degree of dependency between vertices, a parameter that is introduced to control the speed of coarsening. In the algebraic framework, we can coarsen the problem by some form of restriction and then uncoarsen the solution by prolongation, as a standard procedure in multigrid methods. In the geometric framework, the uncoarsening method can improve the embedding quality in the sense of being isometric or conformal. The methods presented provide multiscale resolution of a manifold, and a coarse embedding is very inexpensive. An application to intrinsic dimension estimation is illustrated. Results of experiments on synthetic data and two image data sets are presented. This is joint work with Yousef Saad. SUMMARY:Enhanced Multilevel Manifold Learning UID:1662 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120321T103000 DTEND;TZID=America/Chicago:20120321T113000 DTSTAMP:20130525T020110 LOCATION:TCS Conference Center, Rms 1404 and 1405, Argonne National Laboratory DESCRIPTION:Nonlinear fluid-structure interaction (FSI) is a dominating feature in many important engineering applications. Examples include underwater implosions, pipeline explosions, flapping wings for micro aerial vehicles, and shock wave lithotripsy. Due to nonlinearity and system complexity, most nonlinear FSI problems have not been thoroughly analyzed, which greatly hinders the advance of related engineering fields.\n\nIn this talk, a high-fidelity computational framework will be presented for challenging FSI problems involving strong shock waves, two-phase flows, large structural deformations, and fluid-induced fracture. Key components of this computational framework include: (1) an embedded boundary method for compressible flows based on the exact solution of local, one-dimensional, fluid-structure Riemann problems; (2) an extended finite element method (XFEM) for nonlinear structures, possibly with strong discontinuities such as cracking; (3) an accurate and robust algorithm for tracking the wet surface of a structure undergoing large deformations and topological changes; (4) a level-set technique for capturing fluid-fluid material interfaces; and (5) a robust and second-order accurate fluid-structure coupled time-integrator. The performance of this computational framework will be illustrated with applications to underwater implosions, pipeline explosions, and highly flexible aeronautical systems. This is a joint work with Professor Ted Belytschko’s group at Northwestern University. The resulting software has been acquired by analysts and researchers at Naval Undersea Warfare Center, Naval Surface Warfare Center, Army Research Laboratory, and The University of Texas at Austin. SUMMARY:A High-Fidelity Computational Approach for Nonlinear Fluid-Structure Interactions UID:1664 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120405T100000 DTEND;TZID=America/Chicago:20120405T110000 DTSTAMP:20130525T020110 LOCATION:Building 240 Conference Room 4301, Argonne National Laboratory DESCRIPTION:In DFT based simulations each SCF cycle comprises dozens of large generalized eigenproblems. In a recent study, it has been proposed to consider simulations as made of dozens of sequences of eigenvalue problems, where each sequence groups together eigenproblems with equal k-vector and increasing iteration index i. It was then demonstrated that successive eigenproblems in a sequence are strongly correlated to one another. In particular, by tracking the evolution over iterations of the angle between eigenvectors of adjacent iterations, it was shown the angles decrease noticeably after the first few iterations and become close to collinear. This last result suggests we could use the eigenvectors solution of a problem in a sequence as an educated guess for the eigenvectors of the successive problem. In this talk we present preliminary results that would eventually open the way to a widespread use of iterative solvers in abinitio electronic structure codes. We provide numerical examples where opportunely selected iterative solvers benefit from the reuse of eigenvectors when applied to sequences of eigenproblems extracted from simulations of real-world materials. SUMMARY:Eigenproblems and Eigensolvers in Density Functional Theory UID:1670 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120328T103000 DTEND;TZID=America/Chicago:20120328T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, Conference Room, 4301, Argonne National Laboratory DESCRIPTION:Explicitly parallel languages and explicitly serial languages are each over-constrained, though in different ways. Concurrent Collections (CnC), on the other hand, maximizes the scheduling freedom for a given target (efficiency) and also among distinct targets (portability). The domain expert writing a CnC program focuses on the meaning of the application, not on how to schedule it. \n \nTo prepare an application for parallel execution, we first need to answer two questions: How should the data and computation be divided into chunks that are potential parallel? and What are the scheduling constraints among these chunks? A CnC program specifies exactly this information. The resulting program is \"ready for parallelism.\" CnC isolates the work of the domain expert (interested in finance, chemistry, gaming...) from the tuning expert (interested in load balance, locality, scalability,...) This isolation minimizes the need for the domain expert to think about all the complications of parallel systems. CnC is a coordination language that specifies the required orderings among potentially parallel chunks of application. As a coordination language it must be paired with a computation language. Intel Concurrent Collections for C++ supports C++ programs. \n \nThe talk will cover an introduction to the CnC domain specification and performance results for the Intel distributed CnC/C++ system. \n \nBio for Kath Knobe: \n\nKathleen Knobe worked at Compass (aka Massachusetts Computer Associates) from 1980 to 1991 designing compilers for a wide range of parallel platforms including Thinking Machines, MasPar, Alliant, Numerix, and several government projects. In 1991 she decided to finish her education. After graduating from MIT in 1997, she joined Digital Equipment\'s Cambridge Research Lab (CRL). She stayed through the DEC/Compaq/HP mergers and when CRL was acquired by Intel. She currently works in the Software Solutions Group/Developer Products Group) at Intel. \n \nIn addition to CnC, her major projects include the Subspace Model of computation (a compiler internal form for parallelism), Data Optimization (compiler transformations for locality), Array Static Single Assignment form (a method of achieving for array elements the advantages that SSA has for scalars), Weak Dynamic Single Assignment form (a global method for eliminating overwriting of data to maximize scheduling flexibility), Stampede (a programming model for streaming media applications). \n \nBio for Frank Schlimbach:\n\nFrank is a senior software engineer in DPD\'s Technology, Pathfinding and Innovation (TPI) group at Intel. His main interest is parallel/distributed computing, algorithms, and software development. He is currently responsible for advancing Intel Concurrent Collections for C++. Eight years ago he joined Intel to create new parallel/distributed performance analysis tools (Intel Trace Analyzer and Collector). Prior to this he earned his PhD at the University of Greenwich and he worked on MPI profiling tools at Pallas GmbH in Germany. SUMMARY:Concurrent Collections (CnC): Easy and effective distributed computing UID:1676 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120404T150000 DTEND;TZID=America/Chicago:20120404T160000 DTSTAMP:20130525T020110 LOCATION:Building 240, 1404-1405, Argonne National Laboratory DESCRIPTION:The talk will first present the basic context of the recompute vs. store tradeoff as it occurs in adjoint computations enabled by algorithmic differentiation. Then, it will cover the currently implemented options deciding the tradeoff, highlight their respective weaknesses, and motivate the need for a new approach. The main part of the talk explains the tradeoff as a problem on a set of directed acyclic graphs (DAGs) and a heuristic to decide between recomputing or storing of values. The heuristic approach will be motivated in the context of a source transformation implementation of adjoint computations and compared to other currently available options. The set of DAGs does not reflect the impact of the control flow on the actual cost. A second phase of the new approach addresses control flow impacts. The talk concludes with implementation concerns and an outlook on future research on this subject. SUMMARY:Automating Recompute vs. Store Decisions in Adjoint Computations UID:1678 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120410T103000 DTEND;TZID=America/Chicago:20120410T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, 4301, Argonne National Laboratory DESCRIPTION:We present an inexact trust-region sequential quadratic programming (SQP) method for the matrix-free solution of large-scale nonlinear programming problems. First, we discuss recent algorithmic advances in the handling of inequality constraints. Second, for optimization problems governed by partial differential equations (PDEs) we introduce a class of preconditioners for optimality systems that are easily integrated into our matrix-free trust-region framework and that efficiently reuse the available PDE solvers. We conclude the presentation with numerical examples in acoustic design, material inversion in elastodynamics, and fluid flow optimization. SUMMARY:A Matrix-Free Trust-Region SQP Algorithm for Large-Scale Optimization UID:1680 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120613T150000 DTEND;TZID=America/Chicago:20120613T160000 DTSTAMP:20130525T020110 LOCATION:Building 240, 4301, Argonne National Laboratory DESCRIPTION:Recently, attempts have been made to apply intelligent data analysis methods to quantum mechanical atomistic simulation [1]. I will discuss our newly introduced machine learning approach for the prediction of quantum chemical properties, based on nuclear charges and atomic positions only [2]. The problem of obtaining molecular properties across chemical compound space, aka. as solving Schroedinger\'s equation, is mapped onto a non-linear statistical regression problem of reduced complexity. We use the ``Coulomb\'\'-matrix to encode all Cartesian and atomic number variables of any molecular compound. Based on this representation, Kernel Ridge Regression models or Neural Networks are trained on, and compared to, various properties computed with hybrid density-functional theory for a sub-set of the GDB-13 database [3] consisting of more than seven thousand organic molecules. Cross-validation routinely yields mean absolute errors for out of sample predictions with single digit percentage errors, competitive to density functional theory accuracy. Investigated properties include atomization energies, HOMO/LUMO eigenvalues, static polarizability, excitation and absorption energies. Applicability and transferability is also demonstrated for the prediction of potential energy curves of unseen compounds. SUMMARY:Towards the Quantum Machine: Fast and accurate modeling of quantum chemical properties using machine learning UID:1682 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120502T150000 DTEND;TZID=America/Chicago:20120502T160000 DTSTAMP:20130525T020110 LOCATION:Building 240, 1404-1405, Argonne National Laboratory DESCRIPTION:Ant Colony Optimization (ACO) is a metaheuristic inspired by the foraging behavior of real ants, proposed to tackle NP-hard combinatorial optimization problems (COPs). An ACO algorithm can be defined by the interplay of two main procedures. The first is construction, which manages a colony of ants to incrementally build solutions. The second is the pheromone update, the process by which the pheromone information is modified in order to guide the ants towards better solutions. Over the years, ACO algorithms have been successfully improved by the application of different ways to update the pheromone. However, when ACO is applied to dynamic COPs, additional modifications are necessary. As an example of dynamic COP, we use the dynamic traveling salesman problem (DTSP), where customers are replaced by new ones during the execution of the algorithm. In such a case, the pheromone information may not make sense from one change to another. Thus, after every change, a restructuring procedure should be adopted to adapt the pheromone information. Few proposals for such procedures be found in the literature and has not been a systematic evaluation of their impact on the algorithms\' performance. In this work, we introduce different ways to restructure the pheromone information and compare them with those that have been already proposed in the literature. We also investigate the impact of using different local searches to enhance algorithms\' performance. Finally, we propose a simple and uniform way to evaluate the algorithm\'s solution quality. In our experiments, we analyzed the effects that different restructuring procedures and local searches have on different ACO algorithms. Furthermore, we defined an ACO algorithm that performs better than previous algorithms tested. SUMMARY:The improvement and analysis of ACO algorithms for the DTSP by the use of different pheromone restructuring procedures and local searches UID:1684 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120517T150000 DTEND;TZID=America/Chicago:20120517T160000 DTSTAMP:20130525T020110 LOCATION:Building 240, 1404-1405, Argonne National Laboratory DESCRIPTION:As supercomputers evolve toward the exascale, the storage subsystems are facing more imminent challenges. On the one hand, application scientists face continued pressure to minimize their interactions with the I/O system, and this situation is likely to result in missed discoveries. On the other hand, storage system architects are striving to seek a balance point between a cost-effective, green storage system and a reliable, resilient, and powerful system capable of dealing with skyrocketing parallelism and extreme bursts of I/O activities. \n\nIn this talk, I am going to discuss several aspects of the Co-Design of next generation storage system. First, I will present an exascale communication network model and its simulation. This model is validated using Little\'s Law and a series of P2P communication tests on BG/L platform. We scaled up the experiments to using 128K cores on Intrepid, the BG/P system. Then I will discuss the Co-Design of Exascale Storage System (CODES) framework for evaluating exascale storage system design points. We compared and verified the experimental results from both our storage system simulator and Intrepid storage system. I will further describe enhancements to the storage system simulator to enable burst buffer simulations. We show that burst buffers can accelerate the application perceived throughput to the external storage system and can reduce the amount of external storage bandwidth required to meet a desired application perceived throughput goal. The models and simulations are based on a parallel discrete-event simulation platform: Rensselaer Optimistic Simulation System (ROSS). SUMMARY:Designing and Understanding Future Storage System UID:1686 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120516T150000 DTEND;TZID=America/Chicago:20120516T160000 DTSTAMP:20130525T020110 LOCATION:Building 240, 1404-1405, Argonne National Laboratory DESCRIPTION:This presentation shares early results from a research program that attempts to quantify legal complexity and explain its variance across time. The talk will focus on formulating legal system measurements that are endogenous to the citation network system and subsequently using those measures to predict precedent citation network growth. Yearly court influence is modeled as a function of system stability, complexity, precedent age, and judicial tenure. SUMMARY:Legal Complexity: Measuring complexity within the Supreme Court precedent network and predicting network growth UID:1688 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120523T150000 DTEND;TZID=America/Chicago:20120523T160000 DTSTAMP:20130525T020110 LOCATION:Building 240, 4301, Argonne National Laboratory DESCRIPTION:In order to obtain more accurate solutions of polynomial systems with numerical continuation methods we use multiprecision arithmetic. Our goal is to offset the overhead of double double arithmetic by accelerating the path trackers and in particular Newton\'s method with a general purpose graphics processing unit. In this talk we present our algorithms for the massively parallel evaluation and differentiation of sparse polynomials in several variables. The implementation of the automatic differentiation of products of variables on the NVIDIA Tesla C2050 Computing Processor using the NVIDIA CUDA framework is described. SUMMARY:Evaluating polynomials in several variables and their derivatives on a GPU computing processor UID:1694 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:00000000T150000 DTEND;TZID=America/Chicago:00000000T160000 DTSTAMP:20130525T020110 LOCATION:Building 240, 1404-1405, Argonne National Laboratory DESCRIPTION:To solve polynomial systems more accurately with numerical continuation methods we use multiprecision arithmetic. Our goal is to offset the overhead of double double arithmetic accelerating the path trackers and in particular Newton\'s method with a general purpose graphics processing unit. Multivariate polynomial evaluation and differentiation often bear the biggest computational cost during path tracking. We present our parallel implementation of an automatic differentiation reverse mode like algorithm for evaluating a polynomial system and its Jacobian matrix on the NVIDIA Tesla C2050 using the CUDA computing architecture. SUMMARY:Evaluating polynomials in several variables and their derivatives on a GPU computing processor UID:1690 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120418T150000 DTEND;TZID=America/Chicago:20120418T160000 DTSTAMP:20130525T020110 LOCATION:Building 240, 1404-1405, Argonne National Laboratory DESCRIPTION:For three decades we have had the capability to take a molecular peak at microbial life on our planet. However, it is only within the last few years that it has become possible to explore the microbial earth with any semblance of depth. With these exciting new technologies comes great responsibility; not least of these is the necessity to appropriately record information about each study. Here we present the Earth Microbiome Project (www.earthmicrobiome.org ), an international effort to redefine our exploration of this frontier. Here we will outline the project aims and goals, and explain why and how the paradigm shift must occur. There are more bacteria in the oceans alone than stars in the known universe, and hence we need a systematic and defined approach to survey this biosphere. Importantly, a survey is only as good as the analysis of the data. Here we will employ revolutionary new algorithms to explore the synergy between existing data and new data to explore overlap and identify gaps in knowledge. Additionally, through metagenomic sequencing we will create a metabolic map of microbial life on this planet, which will shape and define our perspective of how life responds and adapts to changes in the environment. The preliminary EMP projects will be discussed in terms of their immediately findings and how much they tell us about the task ahead. O’Dors Law of Biology states that “Physics and chemistry have laws, while biology only has lawyers looking for loopholes”. It is essential that we better define these biological laws and help to reduce the knowledge gap that prevents us from managing the resources of this planet effectively. SUMMARY:The Earth Microbiome Project: A new paradigm in geospatial and temporal studies of microbial ecology UID:1692 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120606T150000 DTEND;TZID=America/Chicago:20120606T160000 DTSTAMP:20130525T020110 LOCATION:Building 240, 1404-1405, Argonne National Laboratory DESCRIPTION:Scripts, workflows, and related techniques are widely used in scientific computing. These tools allow developers to compose complex applications out of existing simulators, analysis codes, or library functions. However, this model is difficult to apply on high-performance systems at large scale. Our recent work has produced a new implementation of Swift that allows users to compose applications from library function calls. By translating the Swift program into an MPI program via the ADLB load balancer, we have been able to run applications on up to 128K cores of the BG/P. In this talk, we will present the new distributed dataflow-based runtime that enables highly scalable script processing, as well as the compiler techniques used. SUMMARY:Swift/T: Scalable dataflow programming for composite applications UID:1696 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120405T150000 DTEND;TZID=America/Chicago:20120405T160000 DTSTAMP:20130525T020110 LOCATION:Building 240, 4301, Argonne National Laboratory DESCRIPTION:We present a goal-oriented finite element method to solve the compressible Euler/Navier-Stokes equations using continuous Galerkin finite elements. A posteriori error estimations of the quantity of interest are derived in terms of a dual problem for the linearized Euler equations. The primal and dual equations are stabilized by using a residual based artificial viscosity method and solved using explicit Runge-Kutta methods in time. Implementation details of the entropy viscosity [Guermond et al, 2008-2011], stability of the dual problems and applications for turbulent flows will be discussed. SUMMARY:Adaptive finite element methods for compressible flows using high-order stabilization UID:1698 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120402T110000 DTEND;TZID=America/Chicago:20120402T120000 DTSTAMP:20130525T020110 LOCATION:Building 240, TCS Conference rooms 1404 & 1405, Argonne National Laboratory DESCRIPTION:Parallelizing existing sequential programs to run efficiently on multicores is hard. Despite the continuous advances on auto-parallelizing compilers, most of the code is parallelized manually. Interactive tools take a completely different approach: sometimes, less automation is better! They let the programmer be in the driver\'s seat. The programmer is the expert on the problem domain, and so understands the domain concepts amenable to parallelism.\n\nThis talk describes our interactive toolset that enables programmers to refactor sequential code into parallel code that uses the Java concurrent libraries. Our growing toolset supports data parallelism, task parallelism, and concurrent collections. These automated refactorings do not require any program annotations, yet the transformations span multiple, non-adjacent, program statements and require control- and data-flow analysis. Empirical evaluation shows\nthat our toolset is useful: it reduces the burden of analyzing and rewriting code, it is fast enough to be used interac tively, and it correctly identifies and applies transformations that open-source developers overlooked.\n SUMMARY:Interactive Refactoring for Parallelism UID:1702 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120410T133000 DTEND;TZID=America/Chicago:20120410T143000 DTSTAMP:20130525T020110 LOCATION:Bldg 240 Seminar Room 4301, Argonne National Laboratory DESCRIPTION:Stimulation of subsurface microbial communities to induce reductive immobilization of metals is a common approach for bioremediation. However, the overall community response is typically poorly understood, hindering optimization. Here we used community proteogenomics to show that excess input of acetate activates syntrophic interactions involving organisms that use the forward tricarboxylic acid (TCA) cycle for energy generation and those that use the reverse TCA cycle for CO2 fixation. A flow-through sediment column was incubated in a groundwater well in an acetate-amended aquifer and recovered during sulfate reduction. Genomic DNA extracted from the community was sequenced, and the sequences were used to reconstruct, de novo, near-complete genomes for relatives of Desulfobacter (Deltaproteobacteria), Sulfurovum and Sulfurimonas (Epsilonproteobacteria), Bacteroidetes, and Clostridiales (Firmicutes). Fragmentary genomic datasets were also obtained for a Desulfuromonadales-like Deltaproteobacterium and other bacteria. Some are related to known metal-reducing bacteria. The majority of proteins identified by mass spectrometry were derived from Desulfobacter-like species, and indicate the role of this organism in sulfate reduction (Dsr and APS), nitrogen-fixation (Nif) and acetate oxidation to CO2. Proteogenomic data indicate that less abundant community members, Desulfuromonadales bacteria and Bacteroidetes, also contribute to the TCA cycle. Interestingly, proteomic data suggest that sulfide was partially re-oxidized by Epsilonproteobacteria through nitrate-dependent sulfur oxidation (using Nap, Nir, Nos, SQR and Sox) and CO2 fixed using the reverse TCA cycle. Thus, high-levels of carbon amendment aimed to stimulate anaerobic heterotrophy led to carbon fixation in co-dependent chemoautotrophs. These results have implications for understanding complex ecosystem behavior, and show that high levels of organic carbon supplementation can expand the range of microbial functionalities accessible for ecosystem manipulation. SUMMARY:A proteogenomics approach to understanding syntrophic interactions in subsurface microbial communities during carbon amendment UID:1704 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120413T133000 DTEND;TZID=America/Chicago:20120413T143000 DTSTAMP:20130525T020110 LOCATION:Building 240, 1406 and 1407, Argonne National Laboratory DESCRIPTION:The test harness TH is a tool developed by Numerica 21 to facilitate the testing and evaluation of scientific software during the development and maintenance phases of such software. This presentation describes how the tool can be used to measure uncertainty in scientific computations. It confirms that the actual behavior of the code when subjected to changes, typically small, in the code input data reflects formal analysis of the problem\'s sensitivity to its input. Although motivated by studying small changes in the input data, the test harness can measure the impact of any change, including those that go beyond the formal analysis. SUMMARY:A Tool TH to Test and Evaluate Scientific Computations UID:1708 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120411T100000 DTEND;TZID=America/Chicago:20120411T110000 DTSTAMP:20130525T020110 LOCATION:Building 240, 4301, Argonne National Laboratory DESCRIPTION:TBA SUMMARY:TBA UID:1710 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120409T103000 DTEND;TZID=America/Chicago:20120409T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, Seminar Room 4301, Argonne National Laboratory DESCRIPTION:As supercomputers evolve toward the exascale, the storage subsystems are facing more imminent challenges. On the one hand, application scientists face continued pressure to minimize their interactions with the I/O system, and this situation is likely to result in missed discoveries. On the other hand, storage system architects are striving to seek a balance point between a cost-effective, green storage system and a reliable, resilient, and powerful system capable of dealing with skyrocketing parallelism and extreme bursts of I/O activities. \n\nIn this talk, I am going to discuss several aspects of the Co-Design of next generation storage system. First, I will present an exascale communication network model and its simulation. This model is validated using Little\'s Law and a series of P2P communication tests on BG/L platform. We scaled up the experiments to using 128K cores on Intrepid, the BG/P system. Then I will discuss the Co-Design of Exascale Storage System (CODES) framework for evaluating exascale storage system design points. We compared and verified the experimental results from both our storage system simulator and Intrepid storage system. I will further describe enhancements to the storage system simulator to enable burst buffer simulations. We show that burst buffers can accelerate the application perceived throughput to the external storage system and can reduce the amount of external storage bandwidth required to meet a desired application perceived throughput goal. The models and simulations are based on a parallel discrete-event simulation platform: Rensselaer Optimistic Simulation System (ROSS).\n SUMMARY: Co-Design of Next Generation Storage System UID:1712 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120418T123000 DTEND;TZID=America/Chicago:20120418T133000 DTSTAMP:20130525T020110 LOCATION:Building 240/TCS Conference Center Room 1416, Argonne National Laboratory DESCRIPTION:As supercomputers continue to grow in size and complexity, system mean-time-between-failure (SMTBF) decreases dramatically, resulting in more frequent system-wide interrupts. Studies have shown that SMTBF for production systems are only in the order of 10-100 hours, even for systems based on ultra-reliable components. Extrapolating from the current performance and scale, the SMTBF may fall to only 1.25 hours of useful computation on exascale systems. In this talk, I will present our on-going research work to address the resilience problem from three aspects: (1) online failure prediction, (2) automatic failure diagnosis, and (3) online data filtering. Fundamentally, our approach explores advanced data analytics technologies from information fusion, data mining/machine learning, and statistical modeling that utilize HPC-specific domain knowledge. I will also present case studies of applying our work to real system logs collected from various production HPC systems including the Blue Gene/P system at Argonne. This work is conducted in collaboration with several national laboratories including Argonne, Sandia, and ORNL.\n\nBio:\nDr. Zhiling Lan is an associate professor of Computer Science at Illinois Institute of Technology. She received her PhD degree in Computer Engineering from Northwestern University in 2002. Her research interests are in the area of high performance computing, in particular, fault tolerance, resource management and scheduling, and performance analysis and modeling. She has authored/coauthored over 50 papers in leading referred journals and conferences. One of her recent papers on job scheduling received a Best Paper Aware at IPDPS’10. Her work on online failure prediction for supercomputers has been selected as the Top-10 Data Mining Case Studies at the 10th ICDM conference. More information about her and her research can be found at http://www.cs.iit.edu/~lan. SUMMARY:System Monitoring, Diagnosing, and Predicting for Extreme-Scale Computing UID:1714 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120426T103000 DTEND;TZID=America/Chicago:20120426T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, 4301, Argonne National Laboratory DESCRIPTION:In many real-world problems, a big scale gap can be observed between micro- and macroscopic scales of the problem because of the difference in mathematical (engineering, social, biological, physical, etc.) models and/or laws at different scales. The main objective of the multiscale algorithms is to create hierarchies of coarse problems, each representing the original problem at different scales with fewer degrees of freedom. We will discuss different strategies of creating these hierarchies and other components of multiscale methods for several problems related to the network science: compression, partitioning, response to infection spread and cyber attacks, and generation of artificial networks. SUMMARY:Multiscale Methods in the \"Big Data\" World of Networks UID:1716 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120427T103000 DTEND;TZID=America/Chicago:20120427T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, 4301, Argonne National Laboratory DESCRIPTION:We present a new stochastic approximation (SA) type algorithm, namely the randomized stochastic gradient (RSG) method, for solving a class of nonlinear (possibly nonconvex) stochastic programming problems. We establish the rate of convergence of this method for computing an approximate stationary point (resp., optimal solution) for nonconvex (resp., convex) nonlinear programming problems. We also show that this method can handle stochastic programming problems with endogenous uncertainty where the distribution and/or support of the random variables depend on the decision variables. A variant, which consists of applying a post-optimization phase to evaluate a short list of solutions generated by a few independent runs of the RSG method, is proposed to significantly improve the large-deviation properties of the aforementioned convergence rates. These methods are then specialized for solving a class of simulation-based optimization problems, where only stochastic zero-order information (function values) is available. To the best of our knowledge, this is the first-time that the convergence rates of SA type algorithms have been established for solving nonconvex nonlinear programming problems (with possibly endogenous uncertainty). SUMMARY:Stochastic First- and Zero-order Methods for Nonconvex Stochastic Programming UID:1718 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120424T103000 DTEND;TZID=America/Chicago:20120424T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, Seminar Room 4301, Argonne National Laboratory DESCRIPTION:In this talk I will present a scalable and efficient resiliency scheme based on the concept of Containment Domains. Containment domains are programming and system constructs that encapsulate and express application resiliency needs and interact with the system to tune and specialize error detection, state preservation and restoration, and recovery schemes. Containment domains have weak transactional semantics and are nested to take advantage of the machine hierarchy and to enable distributed and hierarchical state preservation, restoration, and recovery as an alternative to non-scalable and inefficient checkpoint-restart (and variants). One of the key motivations behind this work is the idea of proportionality, where the resources devoted to a feature are proportional to the application and scenario needs. Proportionality is critical to continued scaling and performance under the increasing constraints of bandwidth, power, and energy. Essentially, one-size-fits-all and worst-case design approaches are no longer sufficient to building reliable and efficient systems. Time permitting, I will describe additional projects in my group that enable proportional resilience and bandwidth usage in the memory system. SUMMARY:Containment Domains for Scalable and Efficient Resilience UID:1724 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120508T103000 DTEND;TZID=America/Chicago:20120508T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, 1404-1405, Argonne National Laboratory DESCRIPTION:Extrapolating the current trends in Supercomputer development one can expect that there will be at least one Exascale machine (i,e, capable of $10^{18}$ flops peak performance by 2020. Indeed there is already a 10 Petaflop machine (The \"K-Computer\" in Kobe, Japan) with more in the pipeline.\n\nIf one simply multiplies current Petaflop machines by a factor of 1000, an Exascale machine can be expected to have some scary characteristics: $10^9$ cores, Power consumption of ~1GW, mean time to failure significantly below 1 day and memory accesses so expensive that flops will be almost for free by comparison.\n\nThere are some concrete suggestions (on the hardware side) of how these limitations can be overcome, however they will not make programming such a machine any easier. In particular the question remains of what (if any) numerical algorithms will actually scale to $10^9$ cores. Nevertheless past experience tells us that a machine that is bleeding edge in 2020 will probably be standard fare in 2030. The challenges posed by these machine characteristics will remain, and overcoming them is likely to dominate algorithm development in the decades to come.\n\n\nThis talk will present some of the challenges posed by Exascale computing with a particular emphasis on their impact on Optimization Algorithms.\n\nI have been involved over the last two years with the European Exascale Software Initiative (EESI) - which is attempting identify where current software (i.e. middleware, operating systems, compilers/tools, algorithms) is inadequate for use on an Exascale machine and what effort is needed to bridge those gaps by 2020. I will present some of the findings and recommendations coming out of the working group on Numerical Algorithms and Libraries (which includes Optimization).\n\nIn particular I will discuss in what form optimization is likely to be used on such a system, what barriers there currently are to achieve performance and what likely future research avenues are to overcome those barriers.\n\nI will by no means have definite answers to any of these questions, but hope that my talk leads to some interesting and stimulating discussions.\n SUMMARY:Optimization and Exascale Computing UID:1726 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120510T103000 DTEND;TZID=America/Chicago:20120510T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, 4301, Argonne National Laboratory DESCRIPTION:We present progress on an Interior Point based multi-step solution approach for stochastic programming problems. Our approach works with a series of scenario trees that can be seen as successively more accurate discretizations of an underlying probability distribution and employs IPM warmstarts to \"lift\" approximate solutions from one tree to the next larger tree. SUMMARY:Interior Point warmstarts and stochastic programming UID:1730 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120508T103000 DTEND;TZID=America/Chicago:20120508T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, conferenc room 1404-1405, Argonne National Laboratory DESCRIPTION:Extrapolating the current trends in Supercomputer development one can expect that there will be at least one Exascale machine (i,e, capable of $10^{18}$ flops peak performance by 2020. Indeed there is already a 10 Petaflop machine (The \"K-Computer\" in Kobe, Japan) with more in the pipeline.\n\nIf one simply multiplies current Petaflop machines by a factor of 1000, an Exascale machine can be expected to have some scary characteristics: $10^9$ cores, Power consumption of ~1GW, mean time to failure significantly below 1 day and memory accesses so expensive that flops will be almost for free by comparison.\n\nThere are some concrete suggestions (on the hardware side) of how these limitations can be overcome, however they will not make programming such a machine any easier. In particular the question remains of what (if any) numerical algorithms will actually scale to $10^9$ cores. Nevertheless past experience tells us that a machine that is bleeding edge in 2020 will probably be standard fare in 2030. The challenges posed by these machine characteristics will remain, and overcoming them is likely to dominate algorithm development in the decades to come.\n\n\nThis talk will present some of the challenges posed by Exascale computing with a particular emphasis on their impact on Optimization Algorithms.\n\nI have been involved over the last two years with the European Exascale Software Initiative (EESI) - which is attempting identify where current software (i.e. middleware, operating systems, compilers/tools, algorithms) is inadequate for use on an Exascale machine and what effort is needed to bridge those gaps by 2020. I will present some of the findings and recommendations coming out of the working group on Numerical Algorithms and Libraries (which includes Optimization).\n\nIn particular I will discuss in what form optimization is likely to be used on such a system, what barriers there currently are to achieve performance and what likely future research avenues are to overcome those barriers.\n\nI will by no means have definite answers to any of these questions, but hope that my talk leads to some interesting and stimulating discussions. SUMMARY:Optimization and Exascale Computing UID:1732 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120521T140000 DTEND;TZID=America/Chicago:20120521T150000 DTSTAMP:20130525T020110 LOCATION:Building 240 Conference Room 4301, Argonne National Laboratory DESCRIPTION:In this presentation I will give an overview of our project to improve the measurement science of fresh cement and concrete through computer simulation. Both the science we are studying and the computational algorithms we use will be described.\n\nConcrete is a dense suspension of aggregates in a non-Newtonian fluid matrix with aggregates that span many decades in size. Additionally, there is also a large shape variation that cannot be accounted for by modeling the aggregates as idealized spheres. As you mix a batch of concrete, the torque required to mix the ingredients changes as the aggregates form and destroy structures within the material. The result is that it is nearly impossible to create a concrete rheometer that has a simple geometry such that analytical solutions relating torque and rotational velocity to actual rheological parameters are possible. Therefore, it is necessary to model the flow in these rheometer geometries in order to correctly interpret measurements in terms of fundamental units.\n\nOur simulations model the flow of dense suspensions in candidate rheometers (e.g. vane) in order to link measurements (torque and angular velocity) from the concrete rheometer with fundamental rheological parameters (viscosity and yield stress). Further, analysis and visualization of the simulated flows will enable us to develop a fundamental framework to understand important physical mechanisms that control the flow of such complex fluids systems. Results from this study will advance the science of granular fluids and improve measurement science for rheometer design for granular fluid systems.\n\nBio:\n\nWilliam George is a Computer Scientist at NIST, working in the Applied and Computational Mathematics Division of the Information Technology Laboratory. His research interests include parallel algorithms for scientific applications as well as the design of distributed computing environments. In 2009 he was awarded a Department of Commerce Silver Medal in for his work in the simulation of dense suspensions with applications to cement and concrete. SUMMARY:Modeling the Flow of Cement and Concrete in a Rheometer UID:1736 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120524T100000 DTEND;TZID=America/Chicago:20120524T110000 DTSTAMP:20130525T020110 LOCATION:Bldg. 240 Rms 1406 & 1407, Argonne National Laboratory DESCRIPTION:Computational materials scientists have been among the earliest and heaviest users of leadership-class supercomputers. The codes and algorithms which have been developed span a wide range of physical scales, and have been useful not only for gaining scientific insight, but also as testbeds for exploring new approaches for tacking evolving challenges, including massive (nearly million-way) concurrency, an increased need for fault and power management, and data bottlenecks. As examples, I will describe our classical molecular dynamics simulations as early users on the LLNL BG/L and LANL Roadrunner platforms, including in situ analysis and visualization of trillion-atom simulations. Multiscale, or scale-bridging, techniques are attractive from both materials science and computational perspectives, particularly as we look ahead from the current petascale era towards the exascale platforms expected to be deployed by the end of this decade. In particular, the increasingly heterogeneous and hierarchical nature of computer architectures demands that algorithms, programming models, and tools must mirror these characteristics if they are to thrive in this environment. Given the increasing complexity of such high-performance computing ecosystems (architectures, software stack, and application codes), computational “co-design” is recognized to be critical as we move from current petascale (10^15 operations/second) to exascale (10^18 operations/second) supercomputers over the next 5-10 years. The Exascale Co-design Center for Materials in Extreme Environments (ExMatEx) is an effort to do this by initiating an early and extensive collaboration between computational materials scientists, computer scientists, and hardware manufacturers. Our goal is to develop the algorithms for modeling materials subjected to extreme mechanical and radiation environments, and the necessary programming models and runtime systems (middleware) to enable their execution; and also influence potential architecture design choices for future exascale systems. SUMMARY:Exascale Co-design for Materials in Extreme Environments: Heterogeneous Algorithms for Heterogeneous Architectures UID:1740 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120604T100000 DTEND;TZID=America/Chicago:20120604T110000 DTSTAMP:20130525T020110 LOCATION:Building 240, TCS Conference Room 1416, Argonne National Laboratory DESCRIPTION:Numerical simulation is considered the third pillar of science, complementing theory and experiments. Simulation is also of utmost importance in industry. Indeed many science domains, like earth science, life science, physics, chemistry, rely largely on computationally intensive simulations. Scientists have defined their computational needs for the end of this decade at the Exascale level; 10^18 operations per second. Since 2009, many workshops of leaders and top researchers at national and international scale produced roadmaps in hardware and software to meet this objective. Participants expressed a major concern: Exascale systems will experience many more faults than current systems. Fault resilience is not an option; it is essential in the design of Exascale platforms. However, the current approach used in production relies on concepts defined 30 years ago for generic distributed systems. Several recent studies question its applicability at Exascale and advocate for the exploration of novel, potentially disruptive fault resilience techniques. In this talk, we will present promising techniques toward fault resilience at Exascale. We will cover optimizations of fault tolerance and more disruptive fault avoidance approaches. In particular, we will show how we can conceive more efficient fault resilience based on the fundamental characteristics of the numerical simulation codes and the dynamic behaviors of the production platforms. SUMMARY:Advanced Fault Resilience Techniques for Exascale Numerical Simulations UID:1742 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120524T103000 DTEND;TZID=America/Chicago:20120524T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, 1404-1405, Argonne National Laboratory DESCRIPTION:Coherent vortices, being durable for some time, are often observed nature. The vorticity vector is defined by the curl of the velocity vector of a fluid. In the absence of viscosity, the vorticity is frozen into the fluid (the Helmholtz\'s law). Preservation of the link and the knot type of vortex lines, Kelvin\'s circulation theorem and invariance in time of the helicity (=the Hopf invariant) are all consequences of the Helmholtz\'s law. An exposition is given to the significance of these topological invariance, based on the Euler-Poincar`e framework. \n\nA steady incompressible Euler flow is characterized as an extremal of the total kinetic energy with respect to perturbations constrained to an isovortical sheet. An isovortical perturbation preserves vortex-line topology and is expressible most efficiently by the Lagrangian variables. I will show how topological ideas work in the variational formulation for deriving steady solutions of the Euler equations, and in analyses of their linear and nonlinear stability. SUMMARY:Kinematic variational principle for vortical structure of Euler flows UID:1744 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120530T123000 DTEND;TZID=America/Chicago:20120530T133000 DTSTAMP:20130525T020110 LOCATION:KPTC 206 (57th and Ellis), University of Chicago DESCRIPTION:We report the first experimental measurements of nonlinear rheological material properties of hagfish gel, a volume-expanding self-defense material composed of a hydrated biopolymer/biofiber gel network. To explain the observed nonlinear viscoelastic behavior, we develop a microstructural constitutive model that has also proven useful for other biopolymer physical gels with non-covalent crosslinks. The linear elastic modulus of the network is observed to be G\' ~ 2 Pa for timescales of 0.1s to 10s, making it one of the softest elastic biomaterials known. Nonlinear rheology is examined via simple shear deformation, and we observe a secant elastic modulus which strain-softens at large input strain while the local tangent elastic modulus strain-stiffens simultaneously. This juxtaposition of simultaneous softening and stiffening suggests a general network structure composed of nonlinear elastic strain-stiffening elements, here modeled as Finite Extensible Nonlinear Elastic (FENE) springs, in which network connections are destroyed as elements are stretched. We simulate the network model in oscillatory shear and creep, including instrument effects which cause inertio-elastic creep ringing. The network model captures the simultaneous softening of the secant modulus and stiffening of tangent modulus as the model enters the nonlinear viscoelastic regime. SUMMARY:Hagfish Self-Defense: Non-linear Rheology of a Biopolymer Physical Gel UID:1748 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120529T130000 DTEND;TZID=America/Chicago:20120529T140000 DTSTAMP:20130525T020110 LOCATION:Building 240 Conference Room 6178, Argonne National Laboratory DESCRIPTION:Computer software exhibits complex interactions with underlying hardware, making understanding performance difficult. Using a memory reference trace, a list of memory transactions performed by software at runtime, enables deep exploration into the structure of software execution. In my talk I will overview 3 systems. First, Waxlamp uses abstract visual encodings, animating the event streams so that the user can directly observe what happened at runtime. Next, a topologically-based approach is developed that finds and visualizes cyclical patterns in the normally linear reference trace as spiral structures expanding out into the time dimension. Finally, an ensemble-based method visualizes side-by-side several reference traces, or a single trace simulated through different cache configurations, bundled into a \'cache simulation ensemble\'. Several case studies illustrate the approach. SUMMARY:Visual Exploration of Software Memory Behavior UID:1750 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120618T140000 DTEND;TZID=America/Chicago:20120618T150000 DTSTAMP:20130525T020110 LOCATION:Building 240/Conference Room 1406-1407, Argonne National Laboratory DESCRIPTION:Next-generation sequencing technologies have led to recognition of a so-called “rare biosphere”. These microbial species are defined by low relative abundance and may be specifically adapted to maintaining low population sizes. We hypothesized that mining of low-abundance next-generation 16S rRNA gene data would lead to the discovery of novel phylogenetic diversity, reflecting microorganisms not yet discovered by previous sampling efforts. Here we test this hypothesis by combining molecular and bioinformatic approaches for targeted retrieval of phylogenetic novelty within rare biosphere species. Using our Illumina 16S rRNA gene sequencing approach and two in-house applications (AutoQIIME and SSUnique), we combined BLASTN network analysis, phylogenetics and targeted primer design to amplify 16S rRNA gene sequences from unique potential bacterial lineages, comprising part of the rare biosphere from a multi-million sequence dataset from an Arctic tundra soil sample. Demonstrating the feasibility of the protocol developed here, three of seven recovered phylogenetic lineages represented extremely divergent taxonomic entities. These divergent target sequences correspond to a) a previously unknown lineage within the BRC1 candidate phylum, b) a sister group to the early diverging and currently recognized monospecific Cyanobacteria Gloeobacter, a genus containing multiple plesiomorphic traits and c) a highly divergent lineage, phylogenetically resolved within mitochondria. A comparison to twelve next-generation datasets from additional soils suggested persistent low-abundance distributions of these novel 16S rRNA genes. The results demonstrate this sequence analysis and retrieval pipeline as applicable for exploring phylogenetic novelty and recovering taxa that may represent significant steps in bacterial evolution. SUMMARY:Combining molecular and computational tools for targeted explorations of taxonomic novelty. UID:1753 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120622T100000 DTEND;TZID=America/Chicago:20120622T110000 DTSTAMP:20130525T020110 LOCATION:Building 240 Conference Room 1406-1407, Argonne National Laboratory DESCRIPTION:This seminar will deal with the work accomplished during my PhD at CERFACS sponsored by Renault.\n\nCycle-to-cycle variations (CCV) in Internal Combustion (IC) engines need to be controlled since they can cause an unwanted increase of fuel consumption, significant increase in pollutants emissions and reduce vehicle driveability.\nLarge Eddy Simulation (LES) seems to be an appropriate tool for studying and predicting this unstationary phenomenon due to its capability of solving the time-resolved Navier-Stokes equations. However, LES applied to piston engines are not numerous in the literature due to the lack of experimental data to validate the numerical results. To tackle this issue, the SGEMAC project (funded by the French research agency) has been setup to acquire experimental data on a 4-valves mono-cylinder IC engine fueled by gaseous propane tailored for LES validation. The experimental database includes numerous operating points and diagnostics (such as Particle Image Velocimetry (PIV), pressure transducers, chemiluminescence and OH planar laser-induced fluorescence (PLIF) for flame visualization).\n\nLES have been conducted with the AVBP solver on more than 100 consecutive cycles and accordingly the seminar will be focus on:\n\n• Presentation of the LES methodology which has been setup to simulate the SGEMAC bench: simulation domain, meshing strategy with full tetrahedron elements, moving mesh techniques and modeling of flame arrestors.\n• Validation of the LES methodology on a motored operating point:comparison of flow field to PIV measurements.\n• Extension of the methodology for fired operating points and comparison with the experiments on a stable operating point (low CCV) and an unstable one with high CCV.\n• Analysis of the physical phenomena responsible for high CCV.\n SUMMARY:Large Eddy Simulation of Cycle-to-Cycle Variations in a Mono-Cylinder Piston Engine UID:1755 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120620T130000 DTEND;TZID=America/Chicago:20120620T140000 DTSTAMP:20130525T020110 LOCATION:Building 240, TCS Conference Room 1404, Argonne National Laboratory DESCRIPTION:“Nothing in biology makes sense except in the light of evolution” is a famous statement of the evolutionary biologist Theodosius Dobzhansky almost four decades ago. Unquestionably, this statement has had a profound impact on generations of biologists and biochemists and continues to do so. However, the relevance of evolution as the theoretical foundation of biology to the applied aspects of the biological sciences (for instance, in the development of bioinformatics pipelines and biotechnological applications) has largely been neglected. In this seminar, I will argue that evolutionary principles can be exploited to generate novel bioinformatic approaches that can guide the discovery of novel natural products, as well as highly constrained novel antibiotic targets that are less prone to evolve antibiotic resistance. These approaches find their metabolic foundations on GSMR, which in turn can be exploited to understand bacterial speciation. Moreover, since the occurrence of enzyme promiscuity, and its relationship with protein conformational diversity, plays a central role in the approaches I will be discussing, this biological phenomenon will also be discussed in both theoretical and experimental grounds. In particular, an unpublished episode of enzyme sub-functionalization through conformational diversity after differential gene gain-and-loss in bacteria will be presented. SUMMARY:GSMRs and enzyme promiscuity: from antibiotic targets to genome (evo) mining of natural products UID:1757 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120619T140000 DTEND;TZID=America/Chicago:20120619T150000 DTSTAMP:20130525T020110 LOCATION:Building 240, TCS Conference Room 1404, Argonne National Laboratory DESCRIPTION:From an evolutionary point of view, Natural selection shapes the sequence, structure and biophysical properties of proteins to fit their environment. DNA contains the complete genetic information that defines the structure and function of an organism. Proteins are formed using the genetic code of the DNA (or RNA). Three different processes are responsible for the inheritance of genetic information and for its conversion from one form to another: Replication, (Reverse) Transcription and Translation. The later is true for prokaryotes and eukaryotes, however, viruses do not have the molecular machinery to accomplish them and have to hijack higher organisms through infection to replicate themselves. After host recognition, viruses insert into a cell and liberate their genome, which comes encapsidated inside a protein shell in most cases. After replication occurs, all different parts of a new virus have to come together inside the cell in a particular way. The process of viral capsid self-assembly involves multi-specific recognition interactions which are not yet well understood. A commonly accepted hypothesis states that there is a small set of residues in the protein-protein interfaces of a capsid (hot-spots) mainly responsible for the molecular mechanism by which capsid proteins can find the thermodynamically correct path to spontaneously form symmetric macrostructures. Using evolutionary and structural criteria, we have developed novel analysis and visualization bioinformatic tools for the prediction of viral hot-spots. Our theoretical results show the presence of specific network patterns, or fingerprints, and their implication in the self-assembly process is being validated in vitro. Whether a universal recognition molecular mechanism common to all virus families exists is an open question. In this seminar I will present current HPC strategies followed in order to address this. SUMMARY:Viral capsid self-assembly: A multidisciplinary approach UID:1759 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120710T080000 DTEND;TZID=America/Chicago:20120710T180000 DTSTAMP:20130525T020110 LOCATION:Building 240 Conference Room 1416, Argonne National Laboratory DESCRIPTION:What do GPUs, FPGAs, vector processors and other exotic special-purpose chips have in common? They are advanced processor architectures that the scientific community is using to accelerate computationally demanding applications. While high-performance computing systems that use application accelerators are still rare, they will be the norm rather than the exception in the near future. The Symposium on Application Accelerators in High-Performance Computing brings together developers of computing accelerators and end-users of the technology to exchange ideas and learn about the latest developments in the field.\n\nThe symposium focuses on the use of application accelerators in high-performance and scientific computing and issues that surround it. Topics of interest include:\n\n - novel accelerator processors, systems, and architectures\n - integration of accelerators with high-performance computing systems\n - programming models for accelerator-based computing\n - languages and compilers for accelerator-based computing\n - run-time environments, profiling and debugging tools for accelerator-based computing\n - scientific and engineering applications that use application accelerators\n SUMMARY:2012 Symposium on Application Accelerators in High-Performance Computing (SAAHPC12) UID:1761 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120621T103000 DTEND;TZID=America/Chicago:20120621T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, Seminar Room 4301, Argonne National Laboratory DESCRIPTION:Today is so-called data deluge era. A huge amount of data is flooded in many domains of modern society based on the advancements of technologies and social networks. Dimension reduction is a useful tool for data visualization of such high-dimensional data and abstract data to make data analysis feasible for such large-scale high-dimensional or abstract scientific data. Among the known dimension reduction algorithms, multidimensional scaling (MDS) is investigated in my research due to its theoretical robustness and high applicability. For the purpose of large-scale multidimensional scaling, we need to figure out two main challenges. One problem is that large-scale multidimensional scaling requires huge amounts of computation and memory resources, because it requires O(N2) memory and computation. Another problem is that multidimensional scaling is known as a non-linear optimization problem so that it is easy to be trapped in local optima if EM-like hill-climbing approach is used to solve it.\n\nTo tackle two challenges mentioned above, we have applied three methodologies to multidimensional scaling: i) parallelization, ii) interpolation, and iii) deterministic annealing (DA) optimization. Parallelization is applied to provide required huge amounts of computation and memory resources by utilizing large-scale distributed-memory systems, such as multicore cluster systems. In addition, we have investigated an interpolation method which utilizes the known mappings of a subset of the given data, named in-sample data, to generate mappings of the remaining out-of-sample data. This approach dramatically reduces computational complexity and memory requirement. DA optimization method has been applied to multidimensional scaling problem in order to avoid local optima. Experimental results illustrate the proposed methodologies are effective to scale up the mapping capacity of multidimensional scaling algorithm and to improve the mapping quality of multidimensional scaling via avoiding local optima. SUMMARY:SCALABLE HIGH PERFORMANCE MULTIDIMENSIONAL SCALING UID:1765 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120621T090000 DTEND;TZID=America/Chicago:20120621T100000 DTSTAMP:20130525T020110 LOCATION:Building 240, Room 4301, Argonne National Laboratory DESCRIPTION:Job scheduling is essential on large-scale computing systems. While research on scheduling has been conducted for decades, our work is motivated by emerging practical issues observed in current production supercomputers, caused by reasons such as human behaviors, new workload characteristics, and increasing system complexity. Further, several challenges are identified in building extreme scale supercomputers, such as reliability, I/O performance, and energy efficiency, which, from our perspective, can be mitigated by appropriate job scheduling strategies. In this talk, Dr. Tang will introduce an integrated resource management and scheduling framework, aiming at addressing the issues and challenges in resource management for large-scale production supercomputers. In this work, he has designed a set of new schemes, implemented them in a production resource manager named Cobalt, and evaluated them with real job traces from production the Blue Gene/P system at Argonne National Laboratory. Experimental results show schemes can effectively improve job scheduling regarding both user satisfaction and system utilization.\n SUMMARY:An Integrated Resource Management and Scheduling Framework for Production Supercomputers UID:1767 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120627T103000 DTEND;TZID=America/Chicago:20120627T113000 DTSTAMP:20130525T020110 LOCATION:Building 240 / Room 4301, Argonne National Laboratory DESCRIPTION:The velocity-vorticity formulation for the solution of unsteady three-dimensional incompressible flows by Kim, Moin and Moser (JFM 1987) is extended for low Mach number combustion. A key advantage of the method is the elimination of the pressure from the momentum equations and, as a result, the errors and complications of pressure boundary conditions inherent in pressure-splitting algorithms common in primitive variable-based CFD solvers. Another merit of the proposed formulation is its efficiency for horizontally homogeneous flows where Fourier expansions can be used for discretization of horizontal derivatives. The resulting elliptic system comprises two evolution equations for two dependent variables and two Poisson type equations. Spatial discretization in the vertical is performed using sixth-order accurate compact finite difference and time stepping is carried out using third order implicit-explicit Runge-Kutta scheme. Open boundary conditions are used at the top boundary and free-slip, no-flux conditions are employed at the bottom boundary. Test cases involving the simulation of compressible Taylor Green vortex flows using a parallel MPI based FORTRAN code are briefly discussed and future research directions are summarized. SUMMARY:Extension of a Velocity-Vorticity Formulation to Low Mach Number Combustion UID:1769 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120702T100000 DTEND;TZID=America/Chicago:20120702T110000 DTSTAMP:20130525T020110 LOCATION:1404/1405 TCS Conference Center, Argonne National Laboratory DESCRIPTION:We examine some esoteric and fundamental issues that are central to efficient simulation of transport-based problems (fluid flow, heat transfer, MHD, electromagnetics, etc.) at petascale and beyond. These issues include effective choices of discretization and solvers, as well as understanding the interaction of computational and communication complexity with hardware parameters for future extreme-scale architectures. SUMMARY:Simulations Beyond Petascale UID:1771 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120627T150000 DTEND;TZID=America/Chicago:20120627T160000 DTSTAMP:20130525T020110 LOCATION:Building 240, 4301, Argonne National Laboratory DESCRIPTION:Modern materials and energy research is increasingly driven by computational chemistry. Ab initio computations produce atom geometry and scalar fields from which the wavefunction, or electron density clouds, can be directly visualized. Larger-scale phenomena are computed using classical molecular dynamics, consisting of atom geometry alone. Visualizing molecules as ball and stick is problematic for models with thousands of atoms or more. The biomolecular community has conventionally used abstractions such as molecular surfaces and ribbons to portray surface and structure, but these representations bear no direct relationship to electronic structure. In Nanovol, we propose a wholly volumetric means of visualizing and modeling material structures at Angstrom to micron scales, using either volume data computed in first-principles calculations or modeled from radial density distributions in bulk. We employ a physically-driven uncertainty classification technique for both volume rendering and analysis, allowing for estimation of material interfaces and boundaries while taking electron structure into account. Finally, we discuss problems with volumetric representation of molecular data, and how topological analyses may provide solutions. SUMMARY:Nanovol: A Volumetric Visualization and Analysis Framework for Chemical Materials Science UID:1773 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120727T103000 DTEND;TZID=America/Chicago:20120727T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, TCS Conference Center, 1406 & 1407, Argonne National Laboratory DESCRIPTION:In this talk I will present some recent work on gene classification and assembly. In the first part of the talk, I\'ll present a new alignment-free algorithm for gene classification. This method transforms DNA sequences into feature vectors which contain the occurrence, location and order relation of k-tuples in the DNA sequence, and uses a hierarchical procedure to cluster DNA sequences based on the feature vectors. In the second part of the talk, I will present a de bruijn graph based parallel gene assembler, called SWAP (small world asynchronous parallel model), that my group has developed. Unlike existing approaches, SWAP uses algorithmic techniques to guarantee localized communication where each process communicates with at most 8 other processes irrespective of the total number of processes in the system. This makes our approach scalable to large supercomputers and large assembly datasets. Nevertheless, like with most applications doing fine-grained data accesses, there is still room for improvement. In the last part of the talk, I\'ll describe some possible future directions we are investigating to utilize MPI-3 RMA and active messages to improve the efficiency of our computation.\n\nProfessor Yanjie Wei received his BS in Applied Physics from Sichuan University, China in Jul. 2004 and PhD in Computational Biophysics from Michigan Technological University in Dec. 2007. He is currently an associate professor in the Center for High Performance Computing at Shenzhen Institutes of Advanced Technology (SIAT), Chinese Academy of Sciences. Before joining SIAT, he worked as a postdoctoral research associate in the Department of Chemical and Biological Engineering at Princeton University. Specifically, his research interests include protein folding, protein structure prediction, gene assembly and classification. By developing advanced computational algorithms and optimization techniques, he aims at solving biologically important problems in the above mentioned areas. Dr. Wei has published more than 10 papers in peer-reviewed journals, such as Proteins, Journal of Chemical Physics, Journal of Physical Chemistry B, Chemical Engineering Sci., etc. He has presented his works at various conferences and workshops, including an invited talk at ACS 2010 annual conference. He also serves as a reviewer for several journals and conferences, including Journal of Chem. Phy., ICIC 2012, 3PGCIC, and ICPADS 2012. SUMMARY:Developing Computational Tools for Gene Classification and Assembly UID:1775 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120713T100000 DTEND;TZID=America/Chicago:20120713T110000 DTSTAMP:20130525T020110 LOCATION:Building 240, TCS Conf Center, Rm 1407, Argonne National Laboratory DESCRIPTION:Abstract:\n\nChipMulti-Processor (CMP) architectures have become mainstream for designing processors. With a large number of cores, Network-On-Chip (NOC) provides a scalable communication method for CMP architectures, where wires become abundant resources available inside the chip. NOC must be carefully designed to meet constraints of power and area, and provide ultra low latencies. In this paper, we propose an Adaptive Physical Channel Regulator (APCR) for NOC routers to exploit huge wiring resources. The flit size in an APCR router is less than the physical channel width (phit size) to provide finer granularity flow control. An APCR router allows flits from different packets or flows to share the same physical channel in a single cycle. The three regulation schemes (Monopolizing, Fair-sharing and Channel-stealing) intelligently allocate the output channel resources considering not only the availability of physical channels but the occupancy of input buffers. In an APCR router, each Virtual Channel can forward a dynamic number of flits every cycle depending on the run-time network status. Our simulation results using a detailed cycle-accurate simulator show that an APCR router improves the network throughput by over 100% in synthetic workloads, compared with a traditional design with the same buffer size. An APCR router can outperform a traditional router even if the buffer size is halved.\n\nShort bio:\n\nLei Wang is a phd candidate in Computer Science and Engineering department at Texas A&M University. His research areas are computer architecture and Networks-on-chip. Lei got his MS from Peking University and BS from Beijing Normal University. Both are from computer science department. SUMMARY:An Adaptive Physical Channel Regulator for On-Chip Interconnects UID:1777 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120709T133000 DTEND;TZID=America/Chicago:20120709T143000 DTSTAMP:20130525T020110 LOCATION:Building 240, 1404-1405, Argonne National Laboratory DESCRIPTION:During my Master thesis, I developed an hybrid algorithm combining a global search method, more precisely a genetic algorithm (GA), with a local search method of trust-region type. The power of this hybrid algorithm is to take advantage of the ability of the GA to explore the whole admissible set and to speed up the convergence toward the optimum by applying the local search at the end of the optimization. Such hybrid algorithms are designed to be applied on expensive black box functions. We also used surrogates models of these functions to improve the efficiency of the methods. This work was carried out in partnership with the research center Cenaero (Belgium). We are now currently working on a more challenging problem involving integer and categorical variables. The increasing difficulty comes from the non continuous variables and the fact that we are still working with expensive and black box functions. SUMMARY:Simulation-based optimization in engineering: from hybrid derivative-free methods to MINLP and MVP. UID:1779 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120724T103000 DTEND;TZID=America/Chicago:20120724T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, Seminar Room 4301, Argonne National Laboratory DESCRIPTION:Particle advection, the tracing of massless particles through a vector field, is a core technique for visualizing fluid flows. The resulting field lines, also called streamlines and pathlines, reveal the direction of the flow as well as features such as vortices. When dealing with large-scale flow fields, parallel methods must be used for particle advection. Field lines are difficult to scale in parallel, though, because of high load-imbalance and demanding communication and I/O overhead. In this talk, I will discuss two methods for advecting particles in large-scale flow fields. The first technique is designed to create a load-balanced data partitioning for parallel streamline computation. A graphical representation of the flow field, called a flow graph, is created to help estimate the workload of each data block. The second method is concerned with computing a series of Finite-Time Lyapunov Exponent Fields (FTLE) in parallel, which requires advecting massive numbers of pathlines in a time-varying flow field. By separating all available processes into different groups and forming a pipeline of process groups, faster computation times and less I/O overhead is achieved. SUMMARY:Parallel Particle Advection of Large-Scale Flow Fields UID:1781 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120723T100000 DTEND;TZID=America/Chicago:20120723T110000 DTSTAMP:20130525T020110 LOCATION:1404/1405 TCS Conference Center, Argonne National Laboratory DESCRIPTION:Today 50% of the global population lives in cities, and this will grow to 70% by 2030. The rate at which cities and megalopolises are growing has outpaced the ability of city designers to anticipate their impact on the local and regional environment and populations. Consequently today\'s cities are becoming increasingly costly and unhealthy. Mature areas of computational science, ranging from microbiology to climate, have much to contribute to the design of new cities and the evolution and growth of existing cities. SUMMARY:The Role of Computation in Designing Future Cities UID:1785 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120730T100000 DTEND;TZID=America/Chicago:20120730T110000 DTSTAMP:20130525T020110 LOCATION:1404/1405 TCS Conference Center, Argonne National Laboratory DESCRIPTION:This talk will provide a high level overview of Globus Online and its impact on scientific research enterprise. SUMMARY:Using Software as a Service for Science: Experiences with Globus Online UID:1787 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120717T150000 DTEND;TZID=America/Chicago:20120717T160000 DTSTAMP:20130525T020110 LOCATION:Building 240, 1404-1405, Argonne National Laboratory DESCRIPTION:During the last decade, stable high order finite difference methods as well as finite volume methods applied to initial-boundary-value-problems have been developed. The stability is due to the use of so-called summation-by-parts operators (SBP), penalty techniques for implementing boundary and interface conditions, and the energy method for proving stability. In this talk we discuss some aspects of this technique including the relation to the initial-boundary-value-problem. By reusing the main ideas behind the recent development, new coupling precodures for multi-physics applications have been developed. We will present the theory by analyzing simple examples and apply to complex multi-physics problems such as fluid flow problems, elastic and electromagnetic wave propagation, fluid-structure interaction and conjugate heat transfer. SUMMARY:Initial Boundary Value Problems, Summation-By-Parts Operators and Weak Boundary Conditions with Multi-Physics Applications UID:1789 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120815T103000 DTEND;TZID=America/Chicago:20120815T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, Conference Room 1405, Argonne National Laboratory DESCRIPTION:We evaluate the climate and the computational performance of the Community Earth System Model, version 1, running with the new spectral element atmospheric dynamical core option. The spectral element method is configured to use a cubed-sphere grid, providing quasi-uniform resolution over the sphere, increased parallel scalability and removing the need for polar filters. It uses a fourth order accurate spatial discretization which locally conserves mass and total energy. Using the Atmosphere Model Intercomparison project protocol, we compare the results from the spectral-element dynamical core with those produced by the default finite-volume dynamical core and with observations, for both CAM4 and CAM5 physics.\n\nThe spectral element method also obtains unprecedented scalability on DOE Petascale computers, resulting in record setting performance at 1/4 and 1/8 degree configurations and the ability to effectively use O(200,000) cores on both the ORNL Cray XT6 and ANL IBM BG/P system.\n SUMMARY:The Spectral Element Dynamical Core in the Community Earth System Model Version 1 UID:1791 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120725T150000 DTEND;TZID=America/Chicago:20120725T160000 DTSTAMP:20130525T020110 LOCATION:Building 240, 1404-1405, Argonne National Laboratory DESCRIPTION:Extreme scale computing provides very cheap computations, which allow us to simulate very complex systems. These simulations depend on building blocks that compute approximate answers to a specified tolerance. Given an error tolerance, eps, the algorithm should automatically determine the number and type of numerical data needed to provide the approximation in a reasonable amount of time. While automatic numerical algorithms exist, e.g., quad in MATLAB, we do not understand well under what conditions the can be fooled. This talk addresses this dilemma, in particular for integration and function recovery.\n\nThe error analysis justifying a numerical algorithm typically assumes input functions to lie a ball in a Banach space of some radius, r. To determine the number of function values needed to guarantee that the error is small enough, one must know r, which means that one needs to know the semi-norm of the input function, e.g., || f\' ||. An automatic algorithm approximates this semi-norm, but typically there is no rigorous justification for the estimate. The approach here assumes that f lies in a cone, not a ball. This allows a rigorous justification of the semi-norm of the function and a guarantee that the automatic algorithm returns the correct answer. SUMMARY: Automatic Numerical Algorithms with Performance Guarantees UID:1793 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120809T143000 DTEND;TZID=America/Chicago:20120809T153000 DTSTAMP:20130525T020110 LOCATION:Building 240/Conference Room 1407, Argonne National Laboratory DESCRIPTION:Pyrosequencing studies on scleractinian corals have primarily focused on surveying healthy specimens or those exposed to laboratory treatments of various abiotic stressors (i.e. temp, pH, DOC, and other nutrients). Few studies have applied these next generation techniques in corals, as it pertains to disease. \n\nBiological agents that affect multiple coral species are of particular interest to understand compositional and functional changes in the coral holobiont. White Plague Disease (WPD) is described as a bacterial disease, reported to infect over forty coral species, and responsible for epidemics that have caused significant reef decline in the Caribbean. Although a pathogen was confirmed by Koch’s postulates for WPD, recent 16S barcoding efforts failed to detect and verify this pathogen in diseased samples of the same and other species. However, drastic shifts in the microbial community were observed. \n\nSo far, a holistic understanding of coral disease as it affects the coral holobiont is missing. Here, we apply comparative metatranscriptomics to the ‘dirty’ holobiont (i.e. the coral, algal, and bacterial community) to better understand WPD infection. We collected tissue samples from healthy and WPD-compromised Montastraea sp. samples that were collected at reef sites off Puerto Rico (USA) to generate expression profiles. Metagenomic analysis of the entire microbial community will also be conducted to validate the bacterial transcriptome findings in each sample type. This will be the first study to apply dual metagenomics and -transcriptomics pyrosequencing to obtain a global assessment of a coral disease in situ. \n\n SUMMARY:Transcriptomic landscape of the coral holobiont in White Plague Disease UID:1795 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120803T103000 DTEND;TZID=America/Chicago:20120803T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, Conference Room, 4301, Argonne National Laboratory DESCRIPTION:The study of the universe has been accompanied by ever larger cosmological simulations. These simulations would easily encompass tens of billions of particles. The study of the universe requires efficient analysis tools for these simulations. My talk would mainly focus on our study from how to identify voids through parallel voronoi tessellation to the further step of post processing and analysis using ParaView, a parallel, open-source visualization toolkit.\n\nBio:\nI am a Ph.D. student from University of Tennessee, Knoxville and my research interest focuses on large scale scientific data visualization and analysis. In the summer I am working with Dr. Thomas Peterka on analysis and visualization of cosmological simulation. SUMMARY:Meshing, Visualizing and Analyzing the Universe UID:1797 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120802T130000 DTEND;TZID=America/Chicago:20120802T173000 DTSTAMP:20130525T020110 LOCATION:Building 240, Room 1416, Argonne National Laboratory DESCRIPTION:\n_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/\nTalk: 01\nTime: 13:10 to 13:25\nName: Xuan Zhou\nTitle: PETSc Interface for Elemental Package\nAbstract: Elemental is a library written by Jack Poulson in C++\nfor distributed-memory dense linear algebra that strives to be both\nfast and convenient. However Elemental\'s own interface is quite\ndifferent from PETSc\'s. The PETSc interface for Elemental is meant\nto provide PETSc users with an interface they are familiar with to\naccess Elemental\'s functionality. This presentation introduces the\ndesign of the interface and some intuitive examples.\n_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/\nTalk: 02\nTime: 13:25 to 13:40\nName: Tom (Lihua) Li\nTitle: Parallel Solver for Toeplitz Linear System\nAbstract: Solving a large Toeplitz linear system is a critical\nstep in various data analysis tasks, for a slow solver could\nseverely limit the overall performance. This work demonstrates\nan novel approach to parallelize the solving process, resulting\nan an efficient and scalable solver. Techniques such as global\ncommunication reduction and grid data partitioning are implemented\nto further enhance solver speed.\n_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/\nTalk: 03\nTime: 13:40 to 13:55\nName: Hong Zhang\nTitle: Implementation of extrapolated IMEX methods for ordinary\ndifferential equation in PETSc\nAbstract: Implicit-explicit methods are intended for problems with\nwell-separated time scales. IMEX methods using extrapolation can\nattain very high orders of accuracy and have great parallelization\npotential. I developed an efficient implementation of extrapolated\nIMEX ODE solver in PETSc. The solver provides various control\noptions and can be order adapative. Numerical results show that\nthe solver can achieve theoretical order for nonstiff problems and\norder reduction is observed for stiff problems.\n_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/\nTalk: 04\nTime: 13:55 to 14:10\nName: Michael Burkhart\nTitle: Gaussian Process Modeling\nAbstract: Gaussian Processes provide a powerful tool for data\nmodeling and prediction. They prove more versatile and require fewer\nassumptions than the standard linear model. Discussion will center\non the construction and use of Gaussian Processes. Examples will\nillustrate the power and utility of this fascinating theoretical\nframework.\n_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/\n\n\nBreak: 14:10 to 14:25 \n\n_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/\nTalk: 05\nTime: 14:25 to 14:40\nName: Jay(Jie) Liu\nTitle: Brief introduction to LHS design and GP implementation GP in R\nAbstract: Latin Hypercube Sampling(LHS) is a statistical method which\ncan be used to generate reasonable design for computation. It\'s\noften applied in uncertainty analysis. I am going to make a\nbrief introduction about the LHS design and show some results for\nGP(Gaussian Process) implementation in R which coincides with Mike\'s.\n_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/\nTalk: 06\nTime: 14:40 to 14:55\nName: Ying He\nTitle: High-order solvers for Schrodinger and acoustic wave equations\nAbstract: Targeting accurate prediction of charge carrier dynamics\nin solar cells, to utilize the available energy from the sun, we have\ndeveloped Schrodinger solver, based on spectral element discontinuous\nGalerkin approach, into the petascale solver NekCEM. I\'ll demonstrate\nthe numerical schemes, algorithms, and computational results. I\'ll\nalso discuss about non-reflecting boundary treatment, called\nDrichlet to Neumann technique, that can be widely applied for\nwave-dominated problems.\n_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/\nTalk: 07\nTime: 14:55 to 15:10\nName: Shengyong Cai\nTitle: Mesh sweeping using a cage-based method for interior point\nlocation\nAbstract: The mesh sweeping is an algorithm for all-hexahedral mesh\ngeneration by sweeping the all-quad mesh from the source surfaces\nto the target surfaces. One of the difficult issues is to locate\nthe interior nodes inside the volume after all the source surfaces,\ntarget surfaces and linking surfaces are meshed. In this project,\nthe cage-based method is used to locate the interior nodes. First,\na topological model is created and the interior nodes could be\nplaced by simple translation. Then the interior nodes are bound\nwith the cage, that is, the interior nodes can be computed as an\nexplicit function with the input argument by the cage. The binding\nprocess is done on the topological model. Because all the surfaces\nare meshed, the deformed cage can be obtained directly from the\nreal model. Finally, the interior nodes can be placed in real model\nwith the input from the deformed cage. The new interior nodes\nlocating method is able to guarantee that the interior nodes are\nrelocated accordingly when the bounding cage is deformed. Therefore,\nany inverted element could be avoided during sweeping.\n_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/\nTalk: 08\nTime: 15:10 to 15:25\nName: Drew Wicke\nTitle: Automatic Identification of Partial Separability\nAbstract: Many optimization problems have a partially separable\n(PS) formulation. ADIC2 exposed the sparsity in the Jacobian of\nPS computations by expanding the elementals. The sparsity was then\nexploited by computing the derivative of the smaller elementals. We\nimplemented a tool that automatically identifies PS computations\nusing activity and linearity analysis. Activity analysis is used\nto identify statements that are possibly partially separable and\nlinearity analysis confirms partial separability and determines\nthe elemental variables.\n_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/\n\nBreak 15:25 to 15:40\n\n_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/\nTalk: 09\nTime: 15:40 to 15:55\nName: Xu Zhang\nTitle: Uncertainty Quantification of Fluid Flow Simulation using\nModel Reduction and Kriging\nAbstract: We use Galerkin POD model reduction technique\nin uncertainty quantification of the fluid flow\nsimulation. Gaussian-processes based Kriging is applied to\ncalibrate the outputs of the reduced model. Our experiments\nindicate that statistical quantities such as mean, variance, and\ncumulative distribution can be well predicted. We explore several\napproaches to improve the prediction of the full model outputs\nin our experiments. Numerical results are presented to indicate\nfeatures of these approaches.\n_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/\nTalk: 10\nTime: 15:55 to 16:10\nName: Carmeline Dsilva\nTitle: Molecular dynamics simulations to extract phase field model\nparameters\nAbstract: Phase field models allow us to simulate material systems on\nmuch longer time and length scales than is feasible with traditional\natomistic simulation methods. However, such models require various\nthermodynamic and kinetic parameters. I will discuss using molecular\ndynamics simulations to extract the relevant parameters for phase\nfield simulations of uranium dioxide grains.\n_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/\nTalk: 11\nTime: 16:10 to 16:25\nName: Lulu Liu\nTitle: The quarter-five spot problem with ASPIN\nAbstract: The quarter-five spot problem is a well-known model in\nreservoir simulation. As we all know, the advantage of implicit\ntime integration methods is to take a larger time step, however,\nwe have to solve the nonlinear system. It is shown that the Additive\nSchwartz preconditioned inexact Newton(ASPIN) methods work well for\nseveral computational fluid dynamics problems, but few applications\nof ASPIN to porous media problems have been reported. Hence, we\ntry to apply ASPIN to our model equations obtained from a fully\nimplicit discretization of immiscible two-phase flow in heterogeneous\nporous media.\n_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/\nTalk: 12\nTime: 16:25 to 16:40\nName: Yunfei Song\nTitle: Solving Mixed Integer Nonlinear Programming by Diving\nHeuristic\nAbstract: We present the performance comparisons of 40\ndiving-heuristics in solving mixed-integer nonlinear programming\n(MINLP). These diving-heuristics combine fractional, vector-length,\nlexicographic and reduced-cost heuristics. Moreover, we present a\nnovel heuristic by incorporating Special-Order-Set (SOS) with the\ntraditional diving-heuristics. Rather than only considering variables\nin the traditional diving-heuristics, our approach concerns both\nvariables and SOS constraints simultaneously. The numerical result\nshows the new approach can obtain better feasible solutions than\nthe traditional diving-heuristics for some instances.\n_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/\nTalk: 13\nTime: 16:40 to 16:55\nName: Jing Fu\nTitle: Parallel I/O Optimizations for a Massively Parallel\nElectromagnetic System\nAbstract: In this talk, we examine a few parallel I/O approaches\nfor the checkpointing of a massively parallel electromagnetics\nsolver system called NekCEM. We discuss a MPI-IO collective\napproach (coIO), an application-level I/O staging approach,\ncalled reduced-blocking I/O (rbIO) and threaded version of\nrbIO. We demonstrate their respective performance advantages over\nthe traditional 1 POSIX file per processor approach and also\nshow their impact on production performance improvement on up to\n32K processors on BG/P and Cray.\n_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/\nTalk: 14\nTime: 16:55 to 17:10\nName: Ryan Lewis\nTitle: Improved Point Location for Solving Weakly Coupled PDEs\nAbstract: We present the point location problem and briefly survey a \nfew algorithms for solving it.\n_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/\nTalk: 15 \nTime: 17:10 to 17:25 \nName: Tom Hayes \nTitle: Efficient and accurate climate data representation in MOAB \nAbstract: Managing climate data in an efficient and scalable manner\nposes an interesting challenge to computational scientists. The\nperformance bottleneck in some simulations comes from reading large,\nhigh-resolution input data files. This study develops a parallel file\nreader for the unstructured High Order Multi-scale Modeling Enviroment\n(HOMME) data format in the Mesh Oriented datABase (MOAB) software.\nMOAB’s capabilities for general mesh manipulation combined with an\nefficient HOMME data reader would enable scientists to perform more\ntargeted climate data analysis. The scalability of the parallel reader\nis demonstrated.\n_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/\n SUMMARY:SASSy 2012 UID:1799 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120816T103000 DTEND;TZID=America/Chicago:20120816T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, Conference Room 4301, Argonne National Laboratory DESCRIPTION:Distribution based analysis is becoming increasingly relevant in the context of large data. Distributions (represented in many forms such as histograms and kernel density estimations) provide a concise yet meaningful summary, which can be efficiently stored, searched and analyzed to derive insight about the original data. Moreover, distributions provide the basis of information theoretic and many other types of analysis. \n\nThis talk will describe our effort towards a parallel framework which supports scalable computation and accuracy analysis of data distributions at different levels of detail and resolution. It will also focus on our ongoing study on efficient compression and fast indexing of distributions. Case studies of performing distribution based in situ analysis and post-processing using NEK5000 and HACC simulation data will also be presented.\n\nBio:\nAbon is a Ph.D. candidate in the Computer Science & Engineering department at The Ohio State University. He is broadly interested in analysis and visualization of large scale scientific data. His specific interests include flow visualization, fractal and information theoretic data analysis, query-driven visualization and geovisualization. He is advised by Dr. Han-Wei Shen.\n\n SUMMARY:Distribution Based Analysis and Visualization of Large Scale Scientific Data UID:1801 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120808T150000 DTEND;TZID=America/Chicago:20120808T160000 DTSTAMP:20130525T020110 LOCATION:Building 240, 1404-1405, Argonne National Laboratory DESCRIPTION:We design and implement a framework for parallel computation of homology of cellular spaces over field coefficients, by decomposing the space. Theoretically, we show that optimal decomposition into local pieces is NP-Hard. In practice, we achieve roughly an 8x speedup of homology computation on a 3-dimensional complex with about 10 million simplices using 11 cores. SUMMARY:Multicore Homology UID:1803 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120810T103000 DTEND;TZID=America/Chicago:20120810T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, Conference Room 4301, Argonne National Laboratory DESCRIPTION:Understanding extreme events, such as hurricanes or forest fires, is of paramount importance because of their adverse impacts on human beings. Such events often propagate in space and time. Predicting—even a few days in advance—what locations will get affected by the event tracks and/or at what seasonal intensity these events are anticipated at those regions could benefit our society in many ways. Arguably, simulations from first principles, where underlying physics-based models are described by a system of equations, provide least reliable predictions for variables characterizing the dynamics of these extreme events. Data-driven model building has been recently emerging as a complementary approach that could learn the relationships between historically observed or simulated multiple, spatio-temporal ancillary variables and the dynamic behavior of extreme events of interest. While promising, the methodology for predictive learning from such complex data is still in its infancy. I will present a suit of dynamic graph-based methodologies for in-advance prediction of the dynamic tracks of emerging extreme events, for quantifying seasonal hurricane activity in North America, and for assessing rainfall activity in the Western Africa. These methods offer a superior predictive skill compared to any other methodology currently available in literature. \n\nBio:\nDr. Nagiza F. Samatova is an Associate Professor in Computer Science Department of North Carolina State University and a Senior Research Scientist in Computer Science and Mathematics Division of Oak Ridge National Laboratory. She received the B.S. degree in applied mathematics from Tashkent State University, Uzbekistan, in 1991 and her Ph.D. degree in mathematics from the Computing Center of Russian Academy of Sciences (CCAS), Moscow, in 1993. She also obtained an M.S. degree in Computer Science in 1998 from the University of Tennessee, Knoxville, USA. Dr. Samatova specializes in Graph Theory and Algorithms, High Performance Data Analytics, Bioinformatics, Systems Biology, Data Management, Scientific and High Performance Computing, and Machine Learning. She is the author of over 150 publications in peer-reviewed journals and conference proceedings. SUMMARY:Graph-based Forecasting of Adverse Spatio-Temporal Climate Extremes UID:1805 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120814T140000 DTEND;TZID=America/Chicago:20120814T150000 DTSTAMP:20130525T020110 LOCATION:Building 240/Conference Room 4301, Argonne National Laboratory DESCRIPTION:Pyrosequencing studies on scleractinian corals have primarily focused on surveying healthy specimens or those exposed to laboratory treatments of various abiotic stressors (i.e. temp, pH, DOC, and other nutrients). Few studies have applied these next generation techniques in corals, as it pertains to disease. \n\nBiological agents that affect multiple coral species are of particular interest to understand compositional and functional changes in the coral holobiont. White Plague Disease (WPD) is described as a bacterial disease, reported to infect over forty coral species, and responsible for epidemics that have caused significant reef decline in the Caribbean. Although a pathogen was confirmed by Koch’s postulates for WPD, recent 16S barcoding efforts failed to detect and verify this pathogen in diseased samples of the same and other species. However, drastic shifts in the microbial community were observed. \n\nSo far, a holistic understanding of coral disease as it affects the coral holobiont is missing. Here, we apply comparative metatranscriptomics to the ‘dirty’ holobiont (i.e. the coral, algal, and bacterial community) to better understand WPD infection. We collected tissue samples from healthy and WPD-compromised Montastraea sp. samples that were collected at reef sites off Puerto Rico (USA) to generate expression profiles. Metagenomic analysis of the entire microbial community will also be conducted to validate the bacterial transcriptome findings in each sample type. This will be the first study to apply dual metagenomics and -transcriptomics pyrosequencing to obtain a global assessment of a coral disease in situ. \n SUMMARY:Transcriptomic landscape of the coral holobiont in White Plague Disease UID:1807 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120817T103000 DTEND;TZID=America/Chicago:20120817T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, Conference Room 4301, Argonne National Laboratory DESCRIPTION:The Message Passing Interface (MPI) is widely considered to be the de-facto standard for portable parallel programming on high-end computing systems. However, the MPI standard only provides functional portability, and not performance portability, across platforms. That is, while an MPI application can be executed anywhere, it is hard to predict the relative performance of different MPI operations on a given platform. For example, how much data does the MPI implementation need to coalesce before pushing it to the network for optimal performance? Is a bulk-synchronous communication model better or worse than an asynchronous PUT/GET based model on a given platform? Can RMA PUT/GET operations safely be replaced with local load/store options for local buffers on all platforms? The answers to all these questions are specific to the MPI implementation and to the hardware architecture. The goal of our work is to design a compiler-assisted framework that would use user annotations, hardware architecture information, and program analysis to transform the input MPI code into a better optimized architecture-specific output MPI code. I will describe the design of our framework and several aspects that the framework needs to work around for optimal performance while maintaining correct behavior. I\'ll also show some preliminary performance numbers demonstrating the improved performance that our framework enables.\n\nShort Bio:\nMd. Ziaul Haque is a PhD student at Dept. of Computer Science, University of Texas at San Antonio. His PhD advisor is Dr. Qing Yi, UTSA. Currently he is working on MPI-Refactoring under the supervision of Dr. Pavan Balaji, Argonne National Laboratory. His research interest includes High Performance Computing, Compiler optimization, Automatic source to source transformation. SUMMARY:Achieving MPI Performance Portability through Annotation-based Compiler Transformations UID:1809 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120824T103000 DTEND;TZID=America/Chicago:20120824T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, TCS Conference Center, 1404 & 1405, Argonne National Laboratory DESCRIPTION:Many new large-scale applications have emerged recently and become increasingly important in areas of bioinformatics, social network analysis, etc. Unlike traditional HPC problems, these applications are data-intensive and often require large amount of small messages sent to random nodes. The traditional parallelization approaches for scientific applications are no longer well-suited to them. Active messages have been proven to be an effective way to parallelize such nontraditional HPC applications. However, most existing active messages libraries are low-level and system-specific, which are hard to be directly used in applications. In this talk I will focus on our study on adding active messages support in MPI based on MPI RMA interface to provide programmability and portability, and show how active messages works with multithreading to improve the performance of applications.\n\nBio:\n\nI am a second-year PhD student from University of Illinois at Urbana-Champaign. My research interest focuses on parallel programming models and runtime systems for high performance computing. In this summer, I work with Dr. Pavan Balaji on exploring the compatibility of active messages with MPI RMA interface. SUMMARY:Supporting Active Messages Functionality in MPI UID:1811 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120820T103000 DTEND;TZID=America/Chicago:20120820T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, Conference Room 4301, Argonne National Laboratory DESCRIPTION:Object-Based Storage Systems are becoming increasingly popular in the storage community. These systems store and access the data as objects, an ordered set of bytes with a unique numerical identifier. Despite their popularity, Object-Based Storage Systems do not have explicit support for namespaces, such as traditional directories. It is possible to support namespace operations in these systems by mapping namespace primitives to the underlying storage. Implementing the namespace operations on an existing storage model is challenging, since these operations have to be atomic in order to guarantee consistency in the namespace.\n\nThis presentation will introduce a number of methods to support atomic and consistent namespace operations on top of a novel object-based storage system, the ASG storage system being developed at Argonne. Advantages and disadvantages of each implementation method will be discussed. Extensions to the storage model in order to better support namespace operations will be explained. Comparison between new namespace operations and the existing standards will also be presented.\n\nBio:\n\nCengiz is a Ph.D. candidate in the Electrical and Computer Engineering Department of the University of Connecticut. His research interests include Parallel Computing, Storage Systems, Distributed File Systems and Computer Architecture. He is currently working on Optimizing PVFS Implementation on top of OSD and Techniques for Energy Conservation in Parallel File Systems at his institution. His advisor is Professor John Chandy. SUMMARY:Namespace Implementation on top of an Object-Based Storage System UID:1813 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120823T103000 DTEND;TZID=America/Chicago:20120823T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, Conference Room 4301, Argonne National Laboratory DESCRIPTION:The storage subsystem of supercomputers is growing more slowly than the compute part. For instance, the Blue Gene/P generation of Argonne supercomputers has one I/O node for 64 compute nodes; and the new Blue Gene/Q Mira has one I/O node for 128 compute nodes. This compute to I/O node ratio is expected to get worse with the shift towards the exascale era. To better cope with the potentially overwhelming number of client requests, the exascale-directed Triton storage project puts the servers in control of I/O communications. The I/O servers decide when and how to service requests while the clients remain passive. This communication model, known as one-sided, matches the semantic of the now ubiquitous Remote Direct Memory Access (RDMA) feature which offloads data transfer from the CPU.\n\nIn this talk, I present the one-sided communication protocol that was adopted for the Triton storage project. I briefly explain the rationale behind the protocol and show how it fits in a typical triton I/O request lifetime. Then, the focus is put on the design decisions made to provide an implementation over two one-sided libraries.\n\nBio: Judicael Zounmevo is a Ph.D. candidate in the Electrical and Computer Engineering department of Queen\'s University in Kingston, ON, Canada. He is interested in optimizing the MPI one-sided communication model, the MPI progress-engine in general and its message queues at large scale. He is supervised by Dr. Ahmad Afsahi. This summer, he worked in the storage group with Dr. Dries Kimpe.\n SUMMARY:A One-sided Communication Protocol for Exascale Storage Systems UID:1815 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120822T103000 DTEND;TZID=America/Chicago:20120822T113000 DTSTAMP:20130525T020110 LOCATION:Building 240 Conference Room 4301, Argonne National Laboratory DESCRIPTION:In this talk, we will discuss the Blue Waters Project Visualization Team, both how it fits within the organization as well as the responsibilities of the team. We will outline the Blue Waters Project: the Blue Waters Project began in 4 B.C. (before Cray), but we will focus on the development of the project beginning at 0, with the installation of the Blue Waters Test and Development System, JYC (Dec. 2011). We will discuss both this system as well as the Early Science System, H2O. Each system presented challenges in administration, support, outreach, research and development and we will discuss these in detail. We will also describe various visualization undertakings and areas in which we see excellent opportunities for collaboration. Finally, we will conclude with some info on the final Blue Waters Machine. SUMMARY:Visualization in the Blue Waters Project UID:1817 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120830T103000 DTEND;TZID=America/Chicago:20120830T113000 DTSTAMP:20130525T020110 LOCATION:Building 240; Room 4301, Argonne National Laboratory DESCRIPTION:Multi-scale, multi-physics scientific and engineering simulation codes take years to develop and optimize. At the same time effective utilization of high performance computing (HPC) resources has always been a balancing act between portability and performance. Expected heterogeneity and ongoing deep architecture changes in the HPC platforms from one generation to the next further complicate this balancing act. Deep infrastructure abstractions within the applications codes that could utilize an ecosystem of tools such as augmented and/or domain specific languages, code transformations and runtime systems provide one possible approach that could deliver portability without significant performance loss. This presentation will outline one such approach which is applicable to a class of scientific codes that solve partial differential equations with predominantly explicit methods. The presentation will also describe FLASH, a community code with a wide user base, which is the primary testbed for our approach. SUMMARY:Architecting Codes for Portability on Future Platforms UID:1819 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120910T140000 DTEND;TZID=America/Chicago:20120910T150000 DTSTAMP:20130525T020110 LOCATION:Building 240 Conference Room 1404-1405, Argonne National Laboratory DESCRIPTION:In this seminar I will present some recent efforts in characterizing \"simple chemical systems\" by using computational approaches - primarily atomistic simulations. The processes of interest include energy transfer, simple reactions and ligand binding. The discussion will also focus on the question for what types of problems HPC is a requirement and how the dramatically increased computational resources may affect the scientific questions we ask. SUMMARY:Atomistic Simulations in the Era of High Performance Computing UID:1821 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120918T133000 DTEND;TZID=America/Chicago:20120918T143000 DTSTAMP:20130525T020110 LOCATION:Building 240 Conference Room 4301, Argonne National Laboratory DESCRIPTION:Cellulase enzymes from fungi and bacteria form the basis of most industrial enzyme cocktails for deconstructing biomass to fermentable sugars for biofuels production. Cellulases are typically multi-modular proteins with carbohydrate binding modules and catalytic domains connected by flexible, glycosylated linkers. Understanding how these enzymes act at solid-liquid interfaces to break down biomass and subsequently engineering them for higher performance is of significant importance for the development of economically-viable biofuels. To that end, our group uses various molecular simulation methods to understand both cellulases and cellulose, with the overall aim of developing a comprehensive, molecular-level picture of how these enzymes function. This talk will review several of the recent discoveries made in our group, including new functions for cellulase sub-domains, a detailed thermodynamic examination of the various crystal forms of cellulose, and identification of new strategies for improving cellulase performance. SUMMARY:Identifying New Routes to Improve Cellulase Enzymes for Biofuels Production with Simulation UID:1823 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120919T103000 DTEND;TZID=America/Chicago:20120919T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, Seminar Room 4301, Argonne National Laboratory DESCRIPTION:One dimensional (graph-like) spaces arise naturally in many fields,both in modeling natural phenomena, and in understanding abstract procedures and simulations. They are the underlying structures for modeling road and river networks, blood vessels, tree roots\' systems, and particle trajectories, to name a few. \n\nLearning these one dimensional spaces becomes an important problem under the more general rubric of space inference. However, there has been only limited work on obtaining a general-purpose algorithm to automatically extract skeleton graph structures. \n\nWe present an algorithm that reconstructs one dimensional spaces from a potentially noisy point cloud, by bringing in a topological concept called the Reeb graph to extract skeleton graphs. Our algorithm is simple, efficient, and easy to use. We demonstrate the generality and effectiveness of our algorithm via several applications in both low and high dimensions. \n\nWe focus on one application in particular: features reconstruction for a special class of singular surfaces. We discuss in some detail how the above algorithm, combined with recent results from spectral theory, allows for a novel algorithm to recover and reconstruct sharp features. SUMMARY:Reconstructing One Dimensional Spaces UID:1831 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20121009T103000 DTEND;TZID=America/Chicago:20121009T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, Conference Room 1406 & 1407, Argonne National Laboratory DESCRIPTION:Michela Taufer research group at the University of Delaware is engaged in interdisciplinary research with scientists and engineers in fields such as chemistry, chemical engineering, pharmaceutical sciences, seismology, and mathematics. The group’s mission is to increase scientific discovery by transforming the way scientific applications deploy high performance computing (HPC) including distributed and multi-core systems. Several research vignettes will be presented that show how we have been successful in addressing challenges in science such as classifying protein-ligand binding geometries of billions of ligands in molecular docking in linear time, performing accurate simulations of fully atomistic macromolecular systems at meso-length and time scales, and improving numerical reproducibility and stability of MD simulations on multi-core platforms.\n\nBio:\nMichela Taufer joined the University of Delaware in 2007 as an assistant professor. She was promoted to associate professor with tenure in 2012. She earned her master’s degrees in Computer Engineering from the University of Padova (Italy) and her doctoral degree in Computer Science from the Swiss Federal Institute of Technology (Switzerland). Taufer completed her postdoctoral studies as a La Jolla Interfaces in Sciences Fellow at the Center for Theoretical Biological Physics (CTBP), at the University of California, San Diego. Taufer’s research focuses on efficient computational algorithms and adaptive scheduling policies for scientific computing on HPC platforms including\nmulti-core, cloud, and volunteer computing. SUMMARY:Transforming computing algorithms and paradigms in HPC to enable more science out of our day-to-day simulations UID:1833 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20121003T150000 DTEND;TZID=America/Chicago:20121003T160000 DTSTAMP:20130525T020110 LOCATION:Building 240, 1404-1405, Argonne National Laboratory DESCRIPTION:We are interested in the fast solution of nonlinear ODE/DAE-constrained mixed-integer optimal control and model predictive control problems. Such problems frequently arise in industrial process control, and typically show significant potential for optimization. The hybrid and nonlinear nature of these problems however is challenging to deal with. We present a computational framework based on a direct and simultaneous method for optimal control and on a partial outer convexification reformulation of the problem. We show how to efficiently compute approximate solutions with feasibility and optimality certificates, and can typically do so without experiencing exponential runtime. The concept of real-time iterations also allows for a transfer of our framework to closed-loop control. Here, the computational performance is determined by the effort required to solve one nonconvex feedback QP in each real-time iteration. Block structures are exploited to significantly reduce this effort. We conclude with an outlook on current algorithmic developments in mixed-integer nonlinear model-predictive control. SUMMARY:Fast numerical methods for mixed-integer nonlinear model predictive control UID:1835 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20121010T150000 DTEND;TZID=America/Chicago:20121010T160000 DTSTAMP:20130525T020110 LOCATION:Building 240, 1404-1405, Argonne National Laboratory DESCRIPTION:Numerous structural materials are produced through solid state diffusional transformations. We discuss the modeling, the numerical methods and some numerical results in the diffusional evolution of microstructures in elastically stressed binary alloys. Equilibrium and growth shapes of a single precipitate embedded coherently in an infinite matrix are obtained for different materials and misfit strains. SUMMARY:Morphological Evolution of Particles in Stressed Solids UID:1837 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120925T140000 DTEND;TZID=America/Chicago:20120925T150000 DTSTAMP:20130525T020110 LOCATION:Building 240 Conference Room 1406-1407, Argonne National Laboratory DESCRIPTION:This talk presents a range of materials by design methods that are being employed to systematically discover new classes of organic materials for optoelectronic applications. These methods include large-scale data-mining, using empirical methods (embracing discrete mathematics) and high-throughput quantum-chemical calculations. Two case studies illustrate the predictive success of this work: the materials discovery of high-performance organic (1) non-linear optical materials, and (2) dyes for dye-sensitized solar cells. SUMMARY:Materials by Design for Optoelectronic Applications UID:1839 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20121010T103000 DTEND;TZID=America/Chicago:20121010T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, Seminar Room 4301, Argonne National Laboratory DESCRIPTION:Debugging large-scale HPC applications is challenging. Faults can come from hardware malfunctions, software bugs or unexpected runtime conditions. In addition, some faults only manifest at large scale when the application is executed with a large number of processes or with a large input data set. Most of the existing debugging tools scale poorly, and more importantly, they do not automate the process of finding the origin of failures; the developer have to manually inspect the state of a large number of processes to find the root-cause of problems.\n\nThis talk will present a probabilistic technique to detect and diagnose faults in large-scale parallel applications. Ignacio will present a methodology to model historic control-flow and timing information of MPI tasks using a semi-Markov model. When a failure occurs, his technique determines the faulty task(s) and code region(s) where the problem is first manifested. The technique isolates abnormal tasks and code regions by clustering MPI behavioral models and then by finding \'outliers\' within task clusters. He has implemented this technique in a tool called AutomaDeD and he has evaluated it against fault injections in the Sequoia and the NAS Parallel Benchmarks; AutomaDeD is able to identify the origin of faults 85% of the time. He also will show how AutomaDeD isolates in a few seconds the origin of a difficult-to-catch bug in a large scale molecular dynamics simulation code. The scalability of his technique has been demonstrated with over 32,000 MPI tasks in a BlueGene/P system.\n\nSpeaker Bio:\nIgnacio Laguna is a PhD Candidate at Purdue University in the School of Electrical and Computer Engineering working under the supervision of Professor Saurabh Bagchi. His research interests include fault detection and diagnosis in large-scale distributed applications and machine-learning for anomaly detection. He received the ACM & IEEE George Michael Memorial HPC Fellowship in 2011; this award honors exceptional PhD students throughout the world whose research focus area is HPC. SUMMARY:Probabilistic Fault Detection and Diagnosis in Large-Scale HPC Applications UID:1841 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20121010T133000 DTEND;TZID=America/Chicago:20121010T143000 DTSTAMP:20130525T020110 LOCATION:Building 240, TCS, Conference Room 1406 & 1407, Argonne National Laboratory DESCRIPTION:Processing large quantities of data is a common scenario for parallel applications. The limited performance of file I/O operations has a strong influence on the scalability of these parallel applications. While distributed memory applications are able to improve the performance of their I/O operations by using parallel I/O libraries, no support for parallel I/O operations for applications using shared-memory programming models such as OpenMP exists as of today. In this talk, I will present parallel I/O interfaces for OpenMP, along with the rationale for our design decisions and the interface specification. I will discuss optimizations performed to enable high performance I/O on shared memory machines. Finally, I will demonstrate the benefits of our approach through tests performed on different parallel file systems, for multiple benchmarks and application scenarios.\n\nBio: \nI am a PhD candidate in the Dept. of Computer Science at University of Houston, Texas. I work with Dr. Edgar Gabriel in the Parallel Software Technologies Lab (PSTL). In general, my field of research is parallel I/O. Specifically, I have been working on parallel I/O for shared memory programming models like OpenMP. I obtained my Masters in Computer Science from University of Houston in 2009 and my bachelors in Computer Engineering from University of Mumbai, India in 2005. SUMMARY:Specification and Performance Evaluation of Parallel I/O Interfaces for OpenMP UID:1843 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20121002T103000 DTEND;TZID=America/Chicago:20121002T113000 DTSTAMP:20130525T020110 LOCATION:CANCELED: Bldg 240 Seminar Room 4301, Argonne National Laboratory DESCRIPTION:In recent years, next-generation DNA sequencing capacity has completely outstripped our ability to computationally digest the resulting volume of data. Driven by the need to actually analyze the data, our lab has developed a suite\nof novel data structures and algorithms for graph compression and data reduction; in addition to being darned efficient on their own, our approaches make use of probabilistic data structures that enable substantially lower memory usage than the best possible exact approach. Using these approaches we have been able to scale de novo data assembly approaches down to cloud computing infrastructure, and we have also completed some of the largest de novo assemblies of metagenomes ever done. Last but not least, these approaches show the way to essentially infinite de novo assembly of environmental\nmicrobial data. SUMMARY:CANCELED: Streaming lossy compression of biological sequence data using probabilistic data structures UID:1845 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20121011T140000 DTEND;TZID=America/Chicago:20121011T150000 DTSTAMP:20130525T020110 LOCATION:Building 240, Conference Room 1406 and 1407, Argonne National Laboratory DESCRIPTION:In this proposal, GPU-accelerated applications are enabled to seamlessly interact with any GPU of a cluster independently of its exact physical location. This provides the possibility of sharing accelerators among different nodes, as GPU-accelerated applications do not fully exploit accelerator capabilities all the time, thus reducing power requirements. Furthermore, decoupling GPUs from nodes, creating pools of accelerators, brings additional flexibility to cluster deployments and allows accessing a virtually unlimited amount of GPUs from a single node, enabling, for example, GPU-per-core executions. Depending on the particular cluster needs, GPUs may be either distributed among computing nodes or consolidated into dedicated GPGPU servers, analogously to disk servers. In both cases, this proposal leads to energy, acquisition, maintenance, and space savings. Performance evaluations employing the rCUDA Framework, developed as a result of the research conducted during a 4-year predoctoral period, demonstrate the feasibility of this proposal within the HPC arena. SUMMARY:Virtualization of Accelerators in High Performance Clusters UID:1847 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20121009T143000 DTEND;TZID=America/Chicago:20121009T153000 DTSTAMP:20130525T020110 LOCATION:Building 240, Seminar Room 4301, Argonne National Laboratory DESCRIPTION:Recent interests in ultra-low concentration of nanoscale biosensors have introduced new types of fluid structure interaction (FSI) problems where molecular and electrokinetic phenomena must be effectively considered. This has motivated enthusiasm in the development of FSI-based tools capable of accounting for various physics. Pertinent to biosensors, the immersed molecular electrokinetic finite element method (IMEFEM) was developed to study three-dimensional motion and deformation of interacting objects immersed in a fluid at room temperature under an applied electric field [1, 2]. The IMEFEM framework incorporates fluctuating hydrodynamics [3] and electrokinetics [4] into the immersed finite element method (IFEM) [5], which is a finite element based formulation extended from the immersed boundary methods. I will present the technical details of the IMEFEM framework and provide several applications. Future insights on the applications will also be conveyed. SUMMARY:The Immersed Continuum Method and its Applications UID:1849 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20121016T103000 DTEND;TZID=America/Chicago:20121016T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, 1405, Argonne National Laboratory DESCRIPTION:In this talk I present an overview of the FEniCS Project, a free/open-source project for the development of innovative concepts and tools for automated scientific computing, with a particular focus on automated solution of differential equations by finite element methods. FEniCS has an extensive list of features for automated, efficient solution of differential equations, including automated solution of variational problems, automated error control and adaptivity, a comprehensive library of finite elements, high performance linear algebra, and many more. I will also present an overview of our work on finite element methods on cut meshes for the simulation of fluid flow and fluid-structure interaction. Cut mesh finite element methods allow the problem domain to be composed of a set of non-matching and overlapping meshes, which can be used for the formulation of very general and flexible methods for the simulation of fluid-structure interaction on complex and deforming geometries. SUMMARY:The FEniCS Project UID:1855 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20121016T133000 DTEND;TZID=America/Chicago:20121016T143000 DTSTAMP:20130525T020110 LOCATION:TCS Bldg. 240, Room 4301, Argonne National Laboratory DESCRIPTION:Current-generation DNA sequencers generate billions of reads per run. These machines operate in parallel, but most of the available analysis software tools for the genomics community run as one process with one or more threads and therefore are not matching the sequencing\nscalability. It is necessary to devise software that scale beyond one process with the help of message passing where numerous processes on many machines collaborate together by passing messages.\nHere, we present Ray Meta and Ray Communities and show how they can assemble de novo and profile samples with unmatched scalability. Ray Meta performs de novo assembly\nby traversing a distributed de Bruijn subgraph while Ray Communities utilizes virtual coloring to label distributed k-mer objects with known sequences after the assembly steps. Ray is built on top of the minimalist RayPlatform framework, which provides various services such as a modular plugin architecture to register handlers onto a\ndistributed state machine (handler tables), a virtual communicator (message aggregation), a virtual processor (user space threads), a virtual message router, and more. Also, RayPlatform provides valuable profiling information to any application using it. SUMMARY:Scalable Microbiome Metagenome Assembly and Profiling UID:1857 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20121019T100000 DTEND;TZID=America/Chicago:20121019T110000 DTSTAMP:20130525T020110 LOCATION:Building 240, 4301, Argonne National Laboratory DESCRIPTION:Current and future generations of large scale computers will be constrained by power -- this is driving an epochal change in computer architectures that rivals the transition to distributed memory machines in the late 1980s. Data movement, though important since the 1980s, will be central to the cost of computing on future generations of machines because memory use and data movement are responsible for most of the power budget of large scale computers. Future architectures are far from well understood but we can assume that memory movement and massive concurrency will be central to effective algorithms on these machines. With this in mind, we propose an equation solver algorithm that is a parallel extension of a low memory multigrid method proposed by Brandt in the 1970s (segregated refinement, with log^d(N) memory complexity). This method has the potential to radically reduce memory usage, and consequently data movement, and thereby use future machines effectively. This algorithm possesses massive concurrency, is amenable to data or task driven programming models, enforces good data locality, is amenable to loop fusion and is generally attractive in memory-centric computer cost models. We describe the algorithm, techniques to implement the method and show numerical results for a simple model problem to verify the correctness of the method. We further consider parallel data models on an example exa-scale machine model and case studies on the application of these methods to applications. SUMMARY:Memory movement optimized multigrid algorithms UID:1859 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20121023T100000 DTEND;TZID=America/Chicago:20121023T110000 DTSTAMP:20130525T020110 LOCATION:Building 240, Video Conference Room 7172, Argonne National Laboratory DESCRIPTION:Advances in optical communication and networking technologies, together with the computing and storage technologies, are dramatically changing the ways scientific research is conducted. A new term, e-Science, has emerged to describe “large-scale science carried out through distributed global collaborations enabled by networks, requiring access to very large scale data collections, computing resources, and high-performance visualization\".\n\nE-Science application workflows are complex and require schedulable and high-bandwidth connectivity with known future characteristics. Moreover, these workflows have performance requirements or metrics that have not been considered by conventional networking. For example, large file transfer may need guaranteed total turnaround time and the rate of progress.\nGiven the long duration of many requests, the network resources available may change before it is completed.\n\nWe develop a novel framework for provisioning a variety of e-Science applications that require complex workflows that span over multiple domains. Our framework provides guarantees on the performance while incurring minimal overhead, both necessary conditions for such a framework to be adopted in practice. SUMMARY:Network resource provisioning in e-Science networks UID:1863 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20121018T150000 DTEND;TZID=America/Chicago:20121018T160000 DTSTAMP:20130525T020110 LOCATION:TCS Building 240, Room 4301, Argonne National Laboratory DESCRIPTION:Abstract not available. SUMMARY:Predicting Cellular Growth Rate Using Genome-Scale Metabolic Models UID:1865 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20121025T100000 DTEND;TZID=America/Chicago:20121025T110000 DTSTAMP:20130525T020110 LOCATION:Building 240, Seminar Room 4301, Argonne National Laboratory DESCRIPTION:A wealth of social information produced by online social networks and user-generated content sharing services is currently fragmented across many diff erent proprietary applications. Combined, it could provide a more accurate representation of the social world that can be leveraged to enable novel socially-aware applications. We introduce Prometheus, a peer-to-peer service that collects social information from multiple sources into a distributed multigraph and exposes it through an interface that implements non-trivial social inferences. The system\'s socially-aware design serves multiple purposes. First, it allows users to manage their social information via socially-trusted peers, thus improving service availability. Second, it exploits naturally-formed social groups for improved end-to-end social inference performance and reduced message overhead. Third, it reduces the opportunity of malicious peers to influence requests in the system, thus constituting a more resilient solution to attacks.\n\nSocial applications can mine the Prometheus social graph to improve performance in search, provide recommendations, allow resource sharing and increase data privacy. The traversal of the de-centralized social graph in the network translates into a socially-informed routing in the peer-to peer layer. We defi ne the projection graph, the result of decentralizing a social graph onto a peer-to-peer network, and study its network properties (degree, node and edge betweenness centrality) and how they can be used to improve application and system performance. Experimental evaluation on real networks demonstrates the association between the properties of the social graph and the projection graph, and that the properties of the (dynamic) projection graph can be inferred from the properties of the (slower changing) social graph. Furthermore, it demonstrates with two application scenarios the usability of the projection graph in designing social search applications and unstructured P2P overlays. SUMMARY:On the Design of Socially-Aware Distributed Systems UID:1867 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20121025T100000 DTEND;TZID=America/Chicago:20121025T110000 DTSTAMP:20130525T020110 LOCATION:Building 240, Seminar Room 4301, Argonne National Laboratory DESCRIPTION:A wealth of social information produced by online social networks and user-generated content sharing services is currently fragmented across many di erent proprietary applications. Combined, it could provide a more accurate representation of the social world that can be leveraged to enable novel socially-aware applications. We introduce Prometheus, a peer-to-peer service that collects social information from multiple sources into a distributed multigraph and exposes it through an interface that implements non-trivial social inferences. The system\'s socially-aware design serves multiple purposes. First, it allows users to manage their social information via socially-trusted peers, thus improving service availability. Second, it exploits naturally-formed social groups for improved end-to-end social inference performance and reduced message overhead. Third, it reduces the opportunity of malicious peers to influence requests in the system, thus constituting a more resilient solution to attacks.\nSocial applications can mine the Prometheus social graph to improve performance in search, provide recommendations, allow resource sharing and increase data privacy. The traversal of the de-centralized social graph in the network translates into a socially-informed routing in the peer-to-peer layer. We de fine the projection graph, the result of decentralizing a social graph onto a peer-to-peer network, and study its network properties (degree, node and edge betweenness centrality) and how they can be used to improve application and system performance. Experimental evaluation on real networks demonstrates the association between the properties of the social graph and the projection graph, and that the properties of the (dynamic) projection graph can be inferred from the properties of the (slower changing) social graph. Furthermore, it demonstrates with two application scenarios the usability of the projection graph in designing social search applications and unstructured P2P overlays. SUMMARY:On the Design of Socially-Aware Distributed Systems UID:1869 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20121024T150000 DTEND;TZID=America/Chicago:20121024T160000 DTSTAMP:20130525T020110 LOCATION:Building 240, 4301, Argonne National Laboratory DESCRIPTION:We present a global optimization algorithm, Branch-and-Sandwich, for optimistic bilevel programming problems which satisfy a regularity condition in the inner problem. The functions involved are assumed to be nonconvex and twice continuously differentiable. The proposed approach can be interpreted as the exploration of two solution spaces (corresponding to the inner and the outer problems) using a single branch-and-bound tree. A novel branching scheme is developed such that classical branch-and-bound is applied to both spaces without violating the hierarchy in the decisions and the requirement for (global) optimality in the inner problem. To achieve this, the well-known features of branch-and-bound algorithms are customized appropriately. For instance, two pairs of lower and upper bounds are computed: one for the outer optimal objective value and the other for the inner value function. The proposed bounding problems do not grow in size during the algorithm and are obtained from the corresponding problems at the parent node. The theoretical properties of the algorithm are investigated and finite $\\epsilon$-convergence to a global solution of the bilevel problem is proved. Thirty-four literature problems are tackled successfully. SUMMARY:Branch-and-Sandwich : A Deterministic Global Optimization Algorithm for Optimistic Bilevel Programming Problems UID:1873 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20121031T150000 DTEND;TZID=America/Chicago:20121031T160000 DTSTAMP:20130525T020110 LOCATION:Building 240, 1407, Argonne National Laboratory DESCRIPTION:The confluence of three technical advances in the past decade has led to a breakthrough in high resolution imaging with x-rays. The advent of brilliant third-generation synchrotron x-ray sources enabled pioneering advances in coherent x-ray experiments. The availability of highly sensitive megapixel x-ray cameras made it possible to record x-ray diffraction patterns in exquisite detail. But while even a candle produces some coherent light and photographic film can be used to record diffraction patterns, the most significant recent advance is development of computational phase retrieval algorithms for recovering images from the recorded diffraction data. Interest by the x-ray imaging community in the basic algorithms, conceived in the 1970s and 1980s for optical and electron imaging, surged when it became evident that enough coherent x-ray photons could be detected to image objects at unprecedented resolution. X-ray coherent diffractive imaging has already made an impact in the condensed matter and life sciences. Fueled by a new generation of ultra-bright x-ray laser sources and fast x-ray detectors, it is poised to revolutionize our understanding of the relationships between structure and function in materials and biological systems, for example, how electron spins interact on the femtosecond scale in magnetic devices, and how sub-cellular organelles function within biological cells. This capability offers tremendous opportunity yet it exposes fresh challenges that currently limit progress. Much work lies ahead to develop efficient algorithms for analyzing complex and noisy diffraction data sets, and practical tools for effective coordination between investigators and analysis efforts. This talk reviews the current status of the field and discusses how algorithmic advances can enable science by coherent diffractive imaging. SUMMARY:Opportunities in Coherent Diffractive Imaging: Help wanted! UID:1875 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20121107T150000 DTEND;TZID=America/Chicago:20121107T160000 DTSTAMP:20130525T020110 LOCATION:Building 240, 4301, Argonne National Laboratory DESCRIPTION:After an implicit time discretization, Cahn-Hilliard Equations become linear fourth order elliptic boundary value problems with essential and natural boundary conditions. The solutions of such problems has weaker regularity than the solutions of the problems with only Dirichlet boundary conditions. The numerical analysis of such problems is, therefore, more subtle. In this talk we will present a quadratic C^0 interior penalty method for linear fourth order elliptic boundary value problems with the boundary conditions of Cahn-Hilliard type. C^0 interior penalty methods are discontinuous Galerkin methods that use Lagrange elements for higher order equations. Convergence analysis, adaptive methods and multigrid methods will be discussed. Numerical results for phase separation and image processing will be presented. SUMMARY:A Quadratic C^0 Interior Penalty Method for Cahn-Hilliard Equations UID:1877 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20121114T150000 DTEND;TZID=America/Chicago:20121114T160000 DTSTAMP:20130525T020110 LOCATION:Building 240, 1404-1405, Argonne National Laboratory DESCRIPTION:Computer experiments simulate the engineering systems by implementing the mathematical models governing the systems in computers. Recently, experiments having large number of input variables and experimental runs started to emerge. In the existing literature, kriging has been commonly used for approximating the complex computer models, but it has limitations for dealing with the large-scale experiments due to its computational complexity and numerical stability. In this talk, we present three new modeling approaches: regression-based inverse distance weighting (RIDW), Kernel approximation, and Approximate Kriging. The proposed methods are shown to be computationally more efficient and numerically more stable than kriging while producing comparable prediction performance. SUMMARY:Analysis of Large-Scale Computer Experiments UID:1879 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20121108T103000 DTEND;TZID=America/Chicago:20121108T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, Seminar Room 4301, Argonne National Laboratory DESCRIPTION:In high performance computing, scientific applications need to make progress despite frequent failures. Thus, long running executions are periodically checkpointed to stable storage. Nowadays, the overhead imposed by parallel file system based checkpointing is about 25% of execution time. In future exascale supercomputers, checkpointing will become prohibitively time consuming. We developed a fault tolerance interface that exploits the features of large scale hybrid systems implementing a low-overhead high-frequency multi-level checkpoint that uses a Topology-Aware Reed-Solomon encoding algorithm with modern local storage devices, advanced clustering techniques and Fault Tolerance Dedicated Threads. Finally, we develop an exascale study using our performance model and we show that our approach can guarantee low overhead in future extreme scale systems. SUMMARY:Fast Checkpoint for Extreme Scale Supercomputers UID:1883 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20121108T140000 DTEND;TZID=America/Chicago:20121108T150000 DTSTAMP:20130525T020110 LOCATION:Building 240, TCS, Conference Room 1404 & 1405, Argonne National Laboratory DESCRIPTION:Power-hungry Graphics processing unit (GPU) accelerators are ubiquitous in high performance computing data centers today. The advent of GPU virtualization frameworks have introduced new opportunities for better management of GPU resources by decoupling them from application execution. However, there are significant challenges in improving the energy efficiency and managing the peak power usage of GPU-enabled server clusters. The underlying system infrastructure shows complex power consumption characteristics depending on the placement of GPU workloads across various compute nodes, power-phases and cabinets in a datacenter. Furthermore, GPU resources need to be scheduled dynamically in the face of time-varying resource demand while keeping the power usage below the peak power constraints. In this talk, Palden will describe the implementation and evaluation of an automated power manager that addresses the above challenges. For GPU virtualization, it uses VOCL, a framework for virtualized execution of OpenCL applications.\n\nBio:\nPalden Lama is a PhD candidate at the Dept. of Computer Science, University of Colorado at Colorado Springs. His research interest is in developing self-managing autonomic resource provisioning methods to control the performance of heterogeneous Internet applications and the power consumption of the underlying servers in a virtualized data center. He is supervised by Dr. Xiaobo Zhou. As a research aide, he is working with Dr. Pavan Balaji, Argonne National Laboratory. \n SUMMARY:Power-Aware Dynamic Placement and Migration in Virtualized GPU Environments UID:1885 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20121113T103000 DTEND;TZID=America/Chicago:20121113T113000 DTSTAMP:20130525T020110 LOCATION:Building 240/Conference Room 1405, Argonne National Laboratory DESCRIPTION:The mechanisms that control carbon (C) and nitrogen (N) cycling between the atmosphere and the biosphere remain central themes in ecosystem ecology. Mycorrhizal fungi (root mutualists) can improve plant uptake of inorganic N in soil, but their capacity to acquire organic N remains less clear. We developed a new fluorescence technique (quantum dots) that can be used to quantify mycorrhizal uptake of organic N. Specifically, we observed quantum dot-labeled organic N within soil hyphae, plant roots and shoots using quantitative imaging techniques in both field and laboratory settings. This seminar will explore the use of quantum dot conjugates to address similar ecological based questions. SUMMARY:\"A PEAK INSIDE THE BLACK BOX: QUANTUM DOTS HIGHLIGHT UPTAKE OF ORGANIC NITROGEN BY MYCORRHIZAL FUNGI IN THE FIELD\" UID:1887 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20121115T103000 DTEND;TZID=America/Chicago:20121115T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, Seminar Room 4301, Argonne National Laboratory DESCRIPTION:Many partial differential equation (PDE)-based engineering and scientific applications have multiple requirements for the finite element mesh discretizing the geometric domain. For example, such requirements may include having non-inverted mesh elements, elements that are well-shaped, elements with uniform element size, and/or elements which yield small PDE interpolation error. Despite there being multiple mesh requirements for various PDE applications, most traditional mesh optimization algorithms optimize only a single objective function and hence improve only one aspect of the mesh. There are several existing multiobjective mesh optimization methods, however most of the existing methods are not flexible and/or not easy to use in practice. In this talk, we will present a mesh optimization framework for mesh quality improvement and mesh untangling, which is flexible and easy to implement in practice. We will extend the idea of mesh optimization framework for mesh deformation problems. SUMMARY:A Multiobjective Mesh Optimization Framework for Mesh Quality Improvement, Mesh Untangling, and Deformation UID:1891 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20121203T100000 DTEND;TZID=America/Chicago:20121203T110000 DTSTAMP:20130525T020110 LOCATION:Building 240/Conference Room 1404, Argonne National Laboratory DESCRIPTION:For ~3.8 billion years, microbial processes etched their indelible mark upon the geochemistry of Planet Earth, giving rise to extreme transformations - from the formation of granite during the early Archaean to the rise in atmospheric oxygen with oxygenic photosynthesis. Even now, atmospheric/climactic homeostasis and the non-equilibrium chemical speciation of most of the elements in the planet’s crust & oceans are due, almost exclusively, to microbial metabolism. In addition to being the architects of the physiochemical world, prokaryotes have a 2 billion-year evolutionary head start on ‘higher’ organisms, exhibiting an incredible diversity of physiologies, with population sizes and generation times that are more amenable to experimentation than macro-organisms. Thanks to emerging technologies (next-generation sequencing and high-resolution mass-spectrometry), scientists are beginning to appreciate the overwhelming complexity of microbial regulatory processes, behaviors, and interactions in the environment. A more thorough understanding of the ecology, evolution, and biochemistry of these prolific nano-machines can help us tackle some of the biggest questions in biology and Earth science. The Earth Microbiome Project (EMP) is a systematic attempt to characterize patterns of global microbial taxonomic and functional diversity, with the aim of processing over 200,000 samples from around the planet. In its first two years, the EMP has sequenced over 10,000 samples (almost 1 billion sequences, representing ~280,000 OTUs), which have been uploaded to a publically available database (www.microbio.me & github), with another ~50,000 samples in the processing queue. This work will enhance our capacity to model global biotic and abiotic dynamics, and expand the theoretical framework underlying ecology and evolution. SUMMARY:The Earth Microbiome Project: Planetary-scale systems ecology UID:1895 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20121119T083000 DTEND;TZID=America/Chicago:20121119T180000 DTSTAMP:20130525T020110 LOCATION:Building 240 1416, Argonne National Laboratory DESCRIPTION:On November 19 - November 21, 2012 the eighth workshop of the INRIA-Illinois Joint Laboratory on Petascale Computing will gather top researchers in HPC from INRIA, ANL and the University of Illinois to explore research problems related to post-petascale supercomputers and present results of their joint work\nobtained since the previous workshop (6 months ago). The workshop will feature technical sessions on seven topics: Big data, applications, I/O and Big systems, Parallel Programming/Runtime (Hybrid, Hierarchical, Heterogeneous), Performance analysis (modeling, simulation), Resilience, Numerical Algorithms and Libraries, and HPC Clouds. Other potential topics of collaboration will be presented and discussed. SUMMARY:INRIA-Illinois Joint Laboratory on Petascale Computing UID:1897 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20121129T103000 DTEND;TZID=America/Chicago:20121129T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, 1404-1405, Argonne National Laboratory DESCRIPTION:Economic systems have many dimensions due to differences in people, products, and geography. There is also much uncertainty in economic systems. Any optimality analysis of an economic problem must be able to incorporate this high dimensionality.\n\nI will describe current efforts in solving such problems. The key idea is Value Function Iteration from the dynamic programming literature, but computational implementation of that approach faces challenges in approximating a value function, computing expectations, and solving huge numbers of relatively small optimization problems. Parallel computing is necessary to solve even modest size problems. A unique feature of economics problems is that the domain of the solution is not known a priori, making it desirable to adopt flexible methods that can solve for the domain of the solution as well as the solution.\n\nThe result is a combination of methods from approximation theory, quadrature theory, and simulation methods, all coordinated in a manner suitable for massively parallel environments. \n SUMMARY:Methods for Solving Large Dynamic Optimization Problems in Economics UID:1899 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20121219T150000 DTEND;TZID=America/Chicago:20121219T160000 DTSTAMP:20130525T020110 LOCATION:Building 240, 1404-1405, Argonne National Laboratory DESCRIPTION:Detailed computational models, massively parallelized calculations, and enormously collaborative simulation projects are increasingly integral to the advancement of science. However, the quality and caliber of this work is limited by a workforce lacking formal training in a software development skill suite that is becoming increasingly essential. To address this unmet need, a number of organizations, including Software Carpentry, have developed online resources and short courses addressing software development best practices such as version control and test driven code development, as well as basic skills such as UNIX mobility [1, 2]. In addition to contributions such as a \"Driver\'s License for High Performance Computing\"[3], Software Carpentry conducts ‘Boot Camps’ at research institutions around the world. These boot camps seek to provide time efficient introductions to essential programming languages and tools without turning “biochemists and mechanical engineers into computer scientists” [4].\n\n[1] K. Huff, A. Scopatz, N. Preston, P.P.H. Wilson. “Rapid Peer Education of a Computational Nuclear Engineering Skill Suite.” Transactions of the American Nuclear Society Annual Conference. Hollywood, FL. June 2011. \n[2] G.V. Wilson, D.A. Aruliah, C.T. Brown, N.P. Chue Hong, M. Davis, R.T. Guy, S.H.D. Haddock, K.D. Huff, I. Mitchell, M. Plumbley, B. Waugh, E.P. White, and P.P.H. Wilson. “Best Practices For Scientific Computing.” arXiv:1210.0530 [cs.MS]. \n[3] Gregory V. Wilson. \"Alpha Test of Driver’s License Exam.\" Software Carpentry Blog. August 16, 2012, http://software-carpentry.org/2012/08/alpha-test-of-drivers-license-exam/. \n[4] Gregory V. Wilson. \"Software Carpentry: Essential Software Skills for Research Scientists,\" September 6, 2006, http://nanohub.org/resources/1811.\n SUMMARY:Peer Education For Scientific Computing UID:1907 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20121128T150000 DTEND;TZID=America/Chicago:20121128T160000 DTSTAMP:20130525T020110 LOCATION:Building 240, 1404-1405, Argonne National Laboratory DESCRIPTION:Algebraic based multigrid methods offer a flexible medium for adapting to a wide range of problems, yet traditional approaches have designed the multigrid components for basic, isotropic, and well-behaved phenomena. As a result, out-of-the-box multilevel preconditioners do not handle a wide range of problems. Moreover, the standard compoenents of the multigrid hierarchy are also not directly suited for data parallel computing (e.g. on a GPU or other high-throughput computing unit). In this talk, we highlight some recent advances in generalizing multigrid, and detail an approach to exposing fine-grained parallelism in the multigrid hierarchy. SUMMARY:Toward a more robust, data parallel algebraic multigrid method UID:1905 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20130206T150000 DTEND;TZID=America/Chicago:20130206T160000 DTSTAMP:20130525T020110 LOCATION:Building 240, 1404-1405, Argonne National Laboratory DESCRIPTION:We will discuss some exciting new developments in the studies of high-order tensors and their coordinate-dependent form, hypermatrices. We argue that a subject that parallels matrix theory and numerical linear algebra is emerging from these recent developments. Examples will be drawn from areas of interest to this audience: physics (elementary particles, Yang-Baxter equations), chemistry (fluorescence spectroscopy, density matrix renormalization group), and optimization (self-concordance, higher-order optimality conditions). SUMMARY:Tensors and hypermatrices in physics, chemistry, and optimization UID:1909 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20121204T133000 DTEND;TZID=America/Chicago:20121204T143000 DTSTAMP:20130525T020110 LOCATION:TCS 240, Room 4301, Argonne National Laboratory DESCRIPTION:The development of improved materials is an expensive and risky venture. Although qualitative models are useful for understanding how materials perform, they cannot rank performance. It is difficult to justify the expense of an exploratory synthesis effort when the improved performance cannot be estimated. Virtual High Throughput Screening (VHTS) uses quantitative atomistic models estimate properties that correlate with experiment to rapidly screen candidate materials. VHTS will focus effort, increase efficiency, and lower the risk of a materials development effort. This talk will present results from the VHTS of zeolites for air separation and metal organic frameworks for carbon sequestration. Current efforts in the dissolution of Li-ion batteries, electrolytes for Li-Air batteries and catalysts for biomass to fuels will also be discussed. SUMMARY:Virtual High Throughput Screening of Materials UID:1913 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20121211T103000 DTEND;TZID=America/Chicago:20121211T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, Seminar Room 4301, Argonne National Laboratory DESCRIPTION:The numerical simulation of compressible, turbulent flows requires an algorithm with high spectral resolution to capture all relevant length scales, as well as yield non-oscillatory solutions across discontinuities. Compact schemes (Lele, J. Comput. Phys., 1992) have significantly higher spectral resolutions than non-compact schemes of the same order of accuracy. A new class of weighted, non-linear compact schemes is presented in this talk that use solution-dependent WENO weights to yield high-resolution, non-oscillatory solutions. Fifth-order CRWENO schemes are derived and applied to scalar conservation laws and the inviscid Euler equations (Ghosh & Baeder, SIAM J. Sci. Comput., 2012). The numerical properties are verified for benchmark problems and compared to those of the WENO schemes. Significant improvements are observed in the resolution of smaller length scales and discontinuities, as well as preservation of flow features over large convection distances. The schemes are integrated into a structured, finite-volume Navier-Stokes solver and applied to problems of practical relevance. Steady and unsteady flows around 2D airfoils and 3D wings/rotors are solved. Improvements are observed in the resolution of near-blade and wake flow features. The schemes are applied to overset meshes and it is verified that no additional modifications are necessary for the application of CRWENO schemes to such domains. The direct numerical simulation (DNS) of canonical turbulent flows - the decay of isotropic turbulence and the shock-turbulence interaction - are attempted and the results presented. The CRWENO schemes show significant improvements in the resolution of smaller length scales. Overall, it is demonstrated that the CRWENO schemes are well-suited for problems with a large range of length scales. SUMMARY:Compact-Reconstruction Weighted Essentially Non-Oscillatory Schemes for Hyperbolic Conservation Laws UID:1915 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20121213T103000 DTEND;TZID=America/Chicago:20121213T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, Conference Room 4301, Argonne National Laboratory DESCRIPTION:The amount of scientific data produced through simulations and experiments in engineering and medicine is continuously growing. These datasets often comprise multiple variables (or fields) that describe the physics of the considered phenomena. The purpose of visualization is to provide the user with graphical tools to discover and analyze the remarkable features exhibited by these datasets. In my thesis, I approach these features from the point of view of their characterization as geometric structures. These structures often take the form of manifolds in the spatial domain. My work is premised on the idea that characterizing and visualizing these manifolds can enable or significantly improve the interpretation and analysis of spatial scientific data across problem domains. Defining and extracting these structures is a challenging task, however. This is mainly due to the ambiguous nature of the features of interest and the computational cost of their extraction from very large-scale spatio-temporal domains. In addition, most visualization scenarios require interactivity to allow the user to steer the visual analysis, and provide a seamless exploration of the domain. In my work, I aim to devise new scalable and high-performance methods for the identification, extraction, and visualization of salient structures from 3D datasets. Applications of this research span fluid dynamics, medical imaging, and combustion research.\n\nThis talk will include a discussion of the interactive visualization of Lagrangian Coherent structures (LCS) using a massively parallel computation of the Finite-Time Lyapunov Exponent. Since LCS structures correspond to ridges of FTLE, the discussion will extend to describing parallel methods for the interactive visualization and extraction of crease manifolds from scientific data in general. The talk will also include the topic of multifiled feature definition and visualization. Finally, I will cover some new scalable ideas concerning the adaptive refinement and approximation of the flow map. SUMMARY:High Performance Structure Extraction for Visualization UID:1917 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20121214T103000 DTEND;TZID=America/Chicago:20121214T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, Conference Room 4301, Argonne National Laboratory DESCRIPTION:Efficient load balancing methods are required to obtain scalability in many scientific software applications. One such application is NWChem\'s coupled-cluster module, which allows for detailed study of chemical problems by iteratively solving the Schrodinger equation with an accurate ansatz. In this case, relevant task information can be obtained just before execution with negligible cost, which suggests a static mapping of task groups to processors can be a simple and more efficient alternative to centralized dynamic load balancing. The distributed tensor contractions are block sparse, and an a priori inspection can quickly assign cost estimations to tasks based on characteristics such as their dimensions. Architecture-specific and empirically driven performance models of the dominant SORT and DGEMM routines serve as a cost estimator for a once-per-simulation static partitioning process. This inspector/executor technique has been demonstrated, improving the NWChem coupled-cluster module’s execution time by as much as 50% at scale. The technique is applicable to any scientific application requiring load balance where performance models or estimations of kernel execution times are available. SUMMARY:Inspector-Executor Load Balancing Algorithms for Block-Sparse Tensor Contractions UID:1919 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20121217T100000 DTEND;TZID=America/Chicago:20121217T110000 DTSTAMP:20130525T020110 LOCATION:Building 240, Conference Room 1404, Argonne National Laboratory DESCRIPTION:We are investigating the genome diversity and genetics of the common foodborne human pathogen Salmonella. Closer inspection of the evolution and phylogenetic structure of this diverse species will help us to understand how the genetic differences between isolates of this species impact on survival strategies of the bacterium. In addition to these evolutionary studies, we are also developing genetic methods to efficiently screen fitness phenotypes of mutants. In our approaches we focus not only on mechanisms of animal infection but also seek to characterize the interaction of Salmonella with the host microbiome and the bacterium’s survival mechanisms outside animal hosts. SUMMARY:Salmonella: Comparative Genomics and High-throughput Genetics UID:1921 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20121217T103000 DTEND;TZID=America/Chicago:20121217T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, Conference Room 1406 and 1407, Argonne National Laboratory DESCRIPTION:There is a known direct relationship between the size of computing resources and their failure rate. As the scale of these platforms become increasingly extreme, the requirements for application fault tolerance are following the same trend. Automatic approaches have a small investment cost, but their scalability is questionable at the magnitude of future machines. More promising techniques toward improving the resilience of application\'s intrinsic algorithms have been developed, but they currently receive no support from the programming model, and without such support, they are bound to fail. This talk discusses two approaches to failure mitigation, one in the context of the current MPI Standard (Version 3.0) and one using an extension to the MPI standard, called Checkpoint-on-Failure (CoF) and User Level Failure Mitigation (UFLM) respectively. Experiments demonstrate the capabilities of these two techniques, and highlight that a fault-aware MPI implementation can have little to no impact on performance for a range of applications, while producing satisfactory recovery times when failures occur. SUMMARY:Toward Recovery Capabilities in Message Passing Environments UID:1923 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20121217T103000 DTEND;TZID=America/Chicago:20121217T113000 DTSTAMP:20130525T020110 LOCATION:Building 240 / Conference Room 4301, Argonne National Laboratory DESCRIPTION:We describe and present results of the implementation of the surface and volume polarization for electrostatics (SVPE) and the iso-density surface solvation models. Unlike most other implementations of the solvation models where the solute and the solvent are described with multiple numerical representations, our implementation uses a multiresolution, adaptive multiwavelet basis to describe both the solute and the solvent.\n\nUsing the geometric structure from SVPE, together with the iso-density solvation model, the effects of solvation on the static properties of a molecule physisorbed on a spherical polarizable continuum particle, with a static dielectric constant is investigated. The effective polarizability of the physisorbed molecule is enhanced by a factor of 105 in vacuo and by only 102 when solvated. The variation of the polarizability of the molecules with respect to the changes in their environment illustrates the importance of electrostatic interaction in the enhancement of the effective polarizability. In the course of the investigation, the solute and the continuous body are represented with the same adaptive multi-wavelet basis functions, thereby, within the user specified precision, eliminating basis set error. This requires reformulation to use integral equations throughout as well as a conscious management of numerical properties of the basis.\n\nClick here to add this event to your calendar.\n SUMMARY:Confinement Effects of Solvation on a Molecule Physisorbed on a Polarizable Continuum Particle UID:1925 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20121220T093000 DTEND;TZID=America/Chicago:20121220T103000 DTSTAMP:20130525T020110 LOCATION:Building 240, Conference Room 1404-1405, Argonne National Laboratory DESCRIPTION:In the first part of the talk, I will focus on search algorithms in empirical performance tuning of computer codes. The increasing complexity, heterogeneity, and rapid evolution of modern computer architectures present obstacles for achieving high performance of scientific codes on different machines. Empirical performance tuning is a viable approach to obtain high-performing code variants based on their measured performance on the target machine. The search for the best code variant can be formulated as a numerical optimization problem. Two classes of algorithms are available to tackle this problem: global and local algorithms. I will present an experimental study of some global and local search algorithms on a number of problems from the recently introduced SPAPT test suite. The results show that local search algorithms are particularly attractive, where finding high-preforming code variants in a short computation time is crucial. In the second part of the talk, I will present the use of machine learning techniques to reduce computationally expensive simulations in chemical compound space and exascale workload characterization studies. Finally, I will give an overview of my future research plans.\n\n SUMMARY:Search Algorithms in Empirical Performance Tuning and Machine Learning for Computationally Expensive Simulations. UID:1931 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20121220T093000 DTEND;TZID=America/Chicago:20121220T103000 DTSTAMP:20130525T020110 LOCATION:Building 240, Conference room 1404, Argonne National Laboratory DESCRIPTION:In the first part of the talk, I will focus on search algorithms in empirical performance tuning of computer codes. The increasing complexity, heterogeneity, and rapid evolution of modern computer architectures present obstacles for achieving high performance of scientific codes on different machines. Empirical performance tuning is a viable approach to obtain high-performing code variants based on their measured performance on the target machine. The search for the best code variant can be formulated as a numerical optimization problem. Two classes of algorithms are available to tackle this problem: global and local algorithms. I will present an experimental study of some global and local search algorithms on a number of problems from the recently introduced SPAPT test suite. The results show that local search algorithms are particularly attractive, where finding high-preforming code variants in a short computation time is crucial. In the second part of the talk, I will present the use of machine learning techniques to reduce computationally expensive simulations in chemical compound space and exascale workload characterization studies. Finally, I will give an overview of my future research plans. SUMMARY:Search Algorithms in Empirical Performance Tuning and Machine Learning for Computationally Expensive Simulations. UID:1933 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20121220T093000 DTEND;TZID=America/Chicago:20121220T103000 DTSTAMP:20130525T020110 LOCATION:Building 240, Conference Room 1404-1405, Argonne National Laboratory DESCRIPTION:In the first part of the talk, I will focus on search algorithms in empirical performance tuning of computer codes. The increasing complexity, heterogeneity, and rapid evolution of modern computer architectures present obstacles for achieving high performance of scientific codes on different machines. Empirical performance tuning is a viable approach to obtain high-performing code variants based on their measured performance on the target machine. The search for the best code variant can be formulated as a numerical optimization problem. Two classes of algorithms are available to tackle this problem: global and local algorithms. I will present an experimental study of some global and local search algorithms on a number of problems from the recently introduced SPAPT test suite. The results show that local search algorithms are particularly attractive, where finding high-preforming code variants in a short computation time is crucial. In the second part of the talk, I will present the use of machine learning techniques to reduce computationally expensive simulations in chemical compound space and exascale workload characterization studies. Finally, I will give an overview of my future research plans.\n\n SUMMARY:Search Algorithms in Empirical Performance Tuning and Machine Learning for Computationally Expensive Simulations. UID:1935 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20121220T093000 DTEND;TZID=America/Chicago:20121220T103000 DTSTAMP:20130525T020110 LOCATION:Building 240, Conference Room 1404-1405, Argonne National Laboratory DESCRIPTION:In the first part of the talk, I will focus on search algorithms in empirical performance tuning of computer codes. The increasing complexity, heterogeneity, and rapid evolution of modern computer architectures present obstacles for achieving high performance of scientific codes on different machines. Empirical performance tuning is a viable approach to obtain high-performing code variants based on their measured performance on the target machine. The search for the best code variant can be formulated as a numerical optimization problem. Two classes of algorithms are available to tackle this problem: global and local algorithms. I will present an experimental study of some global and local search algorithms on a number of problems from the recently introduced SPAPT test suite. The results show that local search algorithms are particularly attractive, where finding high-preforming code variants in a short computation time is crucial. In the second part of the talk, I will present the use of machine learning techniques to reduce computationally expensive simulations in chemical compound space and exascale workload characterization studies. Finally, I will give an overview of my future research plans.\n\n SUMMARY:Search Algorithms in Empirical Performance Tuning and Machine Learning for Computationally Expensive Simulations. UID:1937 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20130109T150000 DTEND;TZID=America/Chicago:20130109T160000 DTSTAMP:20130525T020110 LOCATION:Building 240, 1404-1405, Argonne National Laboratory DESCRIPTION:The increasing geometric complexities in modern applications pose significant challenges to numerical analysis and computational science when using traditional techniques with either structured or unstructured meshes. We present our recent work towards a unified theoretical and algorithmic framework for accurate and stable numerical discretizations and efficient solution techniques for partial differential equations (PDEs) over complex geometries. This talk is composed of two parts. The first part will describe a unified and versatile framework of high-order accurate, stable, and efficient numerical methods based on local weighted-least squares (WLS) approximations and a global weighted-residual formulation for discretizing elliptic and parabolic PDEs over complex and curved geometries. This framework generalizes the finite difference methods to unstructured meshes in a coherent way, and generalizes the finite element methods to have relaxed requirements on mesh quality and to deliver higher-order convergence on curved geometries. The second part describes a hybrid geometric+algebraic multigrid method (or HyGA) for weighted-residual methods with hierarchical basis functions. HyGA combines the rigor, accuracy and runtime-and-memory efficiency of geometric multigrid with the robustness and flexibility of algebraic multigrid, and at the same time is relatively easy to implement. We also discuss the intricate interactions between the WLS-based PDE discretizations with multigrid solvers, including the structure of the linear systems and the solutions of rank-deficient linear systems. SUMMARY:Unified Discretizations and Solution Techniques for Numerical Computations over Complex Geometries UID:1939 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20130108T103000 DTEND;TZID=America/Chicago:20130108T113000 DTSTAMP:20130525T020110 LOCATION:Building 240 / Conference Room 4301, Argonne National Laboratory DESCRIPTION:This work presents results from numerical studies of dynamics in three classical non-linear field theories, each of which possesses stable, localized solutions called solitons. We are mainly interested in two of the theories, known as Skyrme models, which have had application in various areas of physics. The third, which describes the dynamics of a complex scalar field and its solitonic solutions (named Q-balls), is principally viewed as a model problem for the development of solution techniques. In all cases, complicated, time dependent, non-linear partial differential equations in several spatial dimensions must be solved, and this necessitates a computational approach. A particular focus of the work is the simulation of high-energy collisions of the solitons. The chief contributions of this work come from simulations performed within the context of a Skyrme model in two spatial dimensions. We concentrate on the rich phenomenology seen in high-energy scattering of pairs of these objects, and the outcome of head-on and off-axis collisions. The study of instabilities seen in previous simulations of Skyrme models is of central interest. Our results confirm that the governing partial differential equations become of mixed hyperbolic-elliptic type for interactions at sufficiently high-energy. We present strong evidence for the loss of energy conservation and smoothness of the dynamical fields in these instances. This supports the conclusion that the initial value problem at hand becomes ill posed, so that the observed instabilities result from the nature of the equations themselves, and are not numerical artifacts. Our calculations incorporate parallel adaptive mesh refinement, which allows us to deal efficiently with the significant dynamical range exhibited in the simulations.\n\nClick on the icon below to add this even to your calendar. SUMMARY:Ultra-relativistic Scattering of Solitons in Non-linear Field Theory UID:1941 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120109T150000 DTEND;TZID=America/Chicago:20120109T160000 DTSTAMP:20130525T020110 LOCATION:Building 240, 1404-1405, Argonne National Laboratory DESCRIPTION:TBA SUMMARY:TBA UID:1943 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120108T103000 DTEND;TZID=America/Chicago:20120108T113000 DTSTAMP:20130525T020110 LOCATION:Building 240 / Conference Room 4301, Argonne National Laboratory DESCRIPTION:We describe and present results of the implementation of the surface and volume polarization for electrostatics (SVPE) and the iso-density surface solvation models. Unlike most other implementations of the solvation models where the solute and the solvent are described with multiple numerical representations, our implementation uses a multiresolution, adaptive multiwavelet basis to describe both the solute and the solvent.\n\nUsing the geometric structure from SVPE, together with the iso-density solvation model, the effects of solvation on the static properties of a molecule physisorbed on a spherical polarizable continuum particle, with a static dielectric constant is investigated. The effective polarizability of the physisorbed molecule is enhanced by a factor of 105 in vacuo and by only 102 when solvated. The variation of the polarizability of the molecules with respect to the changes in their environment illustrates the importance of electrostatic interaction in the enhancement of the effective polarizability. In the course of the investigation, the solute and the continuous body are represented with the same adaptive multi-wavelet basis functions, thereby, within the user specified precision, eliminating basis set error. This requires reformulation to use integral equations throughout as well as a conscious management of numerical properties of the basis.\n\nClick here to add this seminar to your calendar. SUMMARY:Confinement Effects of Solvation on a Molecule Physisorbed on a Polarizable Continuum Particle UID:1945 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120108T103000 DTEND;TZID=America/Chicago:20120108T113000 DTSTAMP:20130525T020110 LOCATION:Building 240 / Conference Room 4301, Argonne National Laboratory DESCRIPTION:We describe and present results of the implementation of the surface and volume polarization for electrostatics (SVPE) and the iso-density surface solvation models. Unlike most other implementations of the solvation models where the solute and the solvent are described with multiple numerical representations, our implementation uses a multiresolution, adaptive multiwavelet basis to describe both the solute and the solvent.\n\nUsing the geometric structure from SVPE, together with the iso-density solvation model, the effects of solvation on the static properties of a molecule physisorbed on a spherical polarizable continuum particle, with a static dielectric constant is investigated. The effective polarizability of the physisorbed molecule is enhanced by a factor of 105 in vacuo and by only 102 when solvated. The variation of the polarizability of the molecules with respect to the changes in their environment illustrates the importance of electrostatic interaction in the enhancement of the effective polarizability. In the course of the investigation, the solute and the continuous body are represented with the same adaptive multi-wavelet basis functions, thereby, within the user specified precision, eliminating basis set error. This requires reformulation to use integral equations throughout as well as a conscious management of numerical properties of the basis.\n\nClick here to add this seminar to your calendar. SUMMARY:Confinement Effects of Solvation on a Molecule Physisorbed on a Polarizable Continuum Particle UID:1947 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120108T103000 DTEND;TZID=America/Chicago:20120108T000008 DTSTAMP:20130525T020110 LOCATION:Building 240 / Conference Room 4301, Argonne National Laboratory DESCRIPTION:This work presents results from numerical studies of dynamics in three classical non-linear field theories, each of which possesses stable, localized solutions called solitons. We are mainly interested in two of the theories, known as Skyrme models, which have had application in various areas of physics. The third, which describes the dynamics of a complex scalar field and its solitonic solutions (named Q-balls), , is principally viewed as a model problem for the development of solution techniques. In all cases, complicated, time dependent, non-linear partial differential equations in several spatial dimensions must be solved, and this necessitates a computational approach. A particular focus of the work is the simulation of high energy collisions of the solitons. The chief contributions of this work come from simulations performed within the context of a Skyrme model in two spatial dimensions. We concentrate on the rich phenomenology seen in high-energy scattering of pairs of these objects, and the outcome of head-on and off-axis collisions. The study of instabilities seen in previous simulations of Skyrme models is of central interest. Our results confirm that the governing partial differential equations become of mixed hyperbolic- elliptic type for interactions at sufficiently high-energy. We present strong evidence for the loss of energy conservation and smoothness of the dynamical fields in these instances. This supports the conclusion that the initial value problem at hand becomes ill-posed, so that the observed instabilities result from the nature of the equations themselves, and are not numerical artifacts. Our calculations incorporate parallel adaptive mesh refinement, which allows us to deal efficiently with the significant dynamical range exhibited in the simulations.\n\nClick here to add this event to your calendar. SUMMARY:Ultra-relativistic Scattering of Solitons in Non-linear Field Theory UID:1949 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20130107T150000 DTEND;TZID=America/Chicago:20130107T160000 DTSTAMP:20130525T020110 LOCATION:Building 240, 1406--1407, Argonne National Laboratory DESCRIPTION:Accurate predictive simulations of complex real world applications require numerical approximations to first, oppose the curse of dimensionality and second, converge quickly in the presence of steep gradients, sharp transitions, bifurcations or finite discontinuities in high-dimensional random parameter spaces. In this talk we present a novel multi-dimensional multi-resolution sparse grid adaptive wavelet stochastic collocation method (AWSCM), that utilizes hierarchical multi-scale piecewise Riesz basis functions constructed from interpolating wavelets. The basis for our non-intrusive method forms a stable multi-scale splitting and thus, optimal adaptation is achieved. More importantly, when the dimension of this stochastic domain becomes moderately large, we show that utilizing a hierarchical sparse-grid AWSCM (sg-AWSCM) not only combats the curse of dimensionality but, in contrast to the standard sg-SCMs built from global Lagrange-type interpolating polynomials, maintains fast convergence without requiring sufficiently regular stochastic solutions. Error estimates and numerical examples will used to compare the efficiency of the method with several other well-known techniques. SUMMARY:An adaptive wavelet stochastic collocation method for irregular solutions of PDEs with random input data UID:1951 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120108T103000 DTEND;TZID=America/Chicago:20120108T113000 DTSTAMP:20130525T020110 LOCATION:Building 240 / Conference Room 4301, Argonne National Laboratory DESCRIPTION:This work presents results from numerical studies of dynamics in three classical non-linear field theories, each of which possesses stable, localized solutions called solitons. We are mainly interested in two of the theories, known as Skyrme models, which have had application in various areas of physics. The third, which describes the dynamics of a complex scalar field and its solitonic solutions (named Q-balls), , is principally viewed as a model problem for the development of solution techniques. In all cases, complicated, time dependent, non-linear partial differential equations in several spatial dimensions must be solved, and this necessitates a computational approach. A particular focus of the work is the simulation of high energy collisions of the solitons. The chief contributions of this work come from simulations performed within the context of a Skyrme model in two spatial dimensions. We concentrate on the rich phenomenology seen in high-energy scattering of pairs of these objects, and the outcome of head-on and off-axis collisions. The study of instabilities seen in previous simulations of Skyrme models is of central interest. Our results confirm that the governing partial differential equations become of mixed hyperbolic- elliptic type for interactions at sufficiently high-energy. We present strong evidence for the loss of energy conservation and smoothness of the dynamical fields in these instances. This supports the conclusion that the initial value problem at hand becomes ill-posed, so that the observed instabilities result from the nature of the equations themselves, and are not numerical artifacts. Our calculations incorporate parallel adaptive mesh refinement, which allows us to deal efficiently with the significant dynamical range exhibited in the simulations.\n\nClick on the icon below to add this seminar presentation to your calendar. SUMMARY:Ultra-relativistic Scattering of Solitons in Non-linear Field Theory UID:1955 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120108T103000 DTEND;TZID=America/Chicago:20120108T113000 DTSTAMP:20130525T020110 LOCATION:Building 240 / Conference Room 4301, Argonne National Laboratory DESCRIPTION:This work presents results from numerical studies of dynamics in three classical non-linear field theories, each of which possesses stable, localized solutions called solitons. We are mainly interested in two of the theories, known as Skyrme models, which have had application in various areas of physics. The third, which describes the dynamics of a complex scalar field and its solitonic solutions (named Q-balls), , is principally viewed as a model problem for the development of solution techniques. In all cases, complicated, time dependent, non-linear partial differential equations in several spatial dimensions must be solved, and this necessitates a computational approach. A particular focus of the work is the simulation of high energy collisions of the solitons. The chief contributions of this work come from simulations performed within the context of a Skyrme model in two spatial dimensions. We concentrate on the rich phenomenology seen in high-energy scattering of pairs of these objects, and the outcome of head-on and off-axis collisions. The study of instabilities seen in previous simulations of Skyrme models is of central interest. Our results confirm that the governing partial differential equations become of mixed hyperbolic- elliptic type for interactions at sufficiently high-energy. We present strong evidence for the loss of energy conservation and smoothness of the dynamical fields in these instances. This supports the conclusion that the initial value problem at hand becomes ill-posed, so that the observed instabilities result from the nature of the equations themselves, and are not numerical artifacts. Our calculations incorporate parallel adaptive mesh refinement, which allows us to deal efficiently with the significant dynamical range exhibited in the simulations.\n\nClick on the icon below to add this seminar presentation to your calendar. SUMMARY:Ultra-relativistic Scattering of Solitons in Non-linear Field Theory UID:1957 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120108T103000 DTEND;TZID=America/Chicago:20120108T113000 DTSTAMP:20130525T020110 LOCATION:Building 240 / Conference Room 4301, Argonne National Laboratory DESCRIPTION:This work presents results from numerical studies of dynamics in three classical non-linear field theories, each of which possesses stable, localized solutions called solitons. We are mainly interested in two of the theories, known as Skyrme models, which have had application in various areas of physics. The third, which describes the dynamics of a complex scalar field and its solitonic solutions (named Q-balls), , is principally viewed as a model problem for the development of solution techniques. In all cases, complicated, time dependent, non-linear partial differential equations in several spatial dimensions must be solved, and this necessitates a computational approach. A particular focus of the work is the simulation of high energy collisions of the solitons. The chief contributions of this work come from simulations performed within the context of a Skyrme model in two spatial dimensions. We concentrate on the rich phenomenology seen in high-energy scattering of pairs of these objects, and the outcome of head-on and off-axis collisions. The study of instabilities seen in previous simulations of Skyrme models is of central interest. Our results confirm that the governing partial differential equations become of mixed hyperbolic- elliptic type for interactions at sufficiently high-energy. We present strong evidence for the loss of energy conservation and smoothness of the dynamical fields in these instances. This supports the conclusion that the initial value problem at hand becomes ill-posed, so that the observed instabilities result from the nature of the equations themselves, and are not numerical artifacts. Our calculations incorporate parallel adaptive mesh refinement, which allows us to deal efficiently with the significant dynamical range exhibited in the simulations.\n\nClick on the icon below to add this seminar presentation to your calendar. SUMMARY:Ultra-relativistic Scattering of Solitons in Non-linear Field Theory UID:1959 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120107T103000 DTEND;TZID=America/Chicago:20120107T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, Seminar Room 4301, Argonne National Laboratory DESCRIPTION:Graph algorithms have been playing a crucial role for solving problems in several domains including scientific computing, data mining, social networks, computational biology, and image decomposition. To deal with the emerging big data, it has been of significant importance to develop scalable parallel graph algorithms capable of utilizing the parallelism of todays multicore, many-core, and heterogeneous machines. In this talk, I will present the Union-Find (UF) algorithm, which is popularly known as one of the best ways to solve the minimum spanning tree (MST) problem. We performed a comprehensive study to compare all of the different versions of the UF algorithm and proposed the best way of implementing the UF algorithm in general and demonstrated that a somewhat forgotten simple algorithm is the fastest, in spite of the fact that its worst-case time complexity is inferior to that of the commonly accepted “best” algorithms. To solve the MST problem using big data, I will then briefly present scalable frameworks for the UF algorithm both on shared memory and distributed memory computers with reasonable speedup.\n\nLater I will talk on a well-known density based clustering algorithm, named DBSCAN, capable of discovering arbitrary shaped clusters and eliminating noise data. DBSCAN has been popular in image processing for many scientific domains, such as astronomy, biology, and earth science. I will show how a graph algorithmic technique (the parallel UF algorithm) can be used to break the access sequentiality of DBSCAN. Using data sets from millennium simulation runs consisting of billions of stars and galaxies, I show that parallel DBSCAN significantly out-performs previous algorithms, achieving speedups up to 25.97 using 40 cores on shared memory architecture, and speedups up to 5,765 using 8,192 cores on distributed memory architecture. SUMMARY:Scalable Parallel Graph Algorithms UID:1961 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120108T103000 DTEND;TZID=America/Chicago:20120108T120000 DTSTAMP:20130525T020110 LOCATION:Building 240, Conferene Room 1416, Argonne National Laboratory DESCRIPTION:High fidelity mathematical and computer modeling of coupled multiphysics\nproblems require solutions to large systems of stiff, nonlinear, coupled\nequations. ’Black-box’ coupling strategies introduce several sources of\nerror in the solution fields stemming from inaccurate and in some cases\ninconsistent treatment of the spatio-temporal nonlinear coupling terms.\nChoosing the right software and algorithmic options to resolve the inaccuracies\nare usually physics and problem dependent and hence a multiphysics framework\nneeds to encompass several coupling schemes with varying stability and accuracy\ntraits to solve dependent systems in order to analyze, compare and choose\nan optimal strategy. The schemes for solving the coupled systems are similar\nin principle for most approaches and with reusable design principles, accurate\nresolution of the spatio-temporal scales are possible. Different methodologies\nand the relevant numerics that are commonly employed for bringing together various\nphysics components will be explored.\n\nFinally, details of a coupling framework in development based on PETSc and MOAB\nlibraries, CouPE (Coupled Physics Environment), for driving existing high-fidelity\nvalidated physics codes will be provided whose design aims to obtain tightly-coupled\nphysical solutions by employing loosely-coupled software interfaces. Some relevant\nresults obtained from nuclear engineering and radiation-diffusion problems with\nCouPE will be shown. The talk will conclude with proposed enhancements to such a\nglass-box framework to make it amenable to couple both existing and new physics solvers. SUMMARY:Designing a glass-box, PDE-based coupled multiphysics framework UID:1963 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120108T103000 DTEND;TZID=America/Chicago:20120108T120000 DTSTAMP:20130525T020110 LOCATION:Building 240, Conference Room 1416, Argonne National Laboratory DESCRIPTION:High fidelity mathematical and computer modeling of coupled multiphysics problems require solutions to large systems of stiff, nonlinear, coupled equations. ’Black-box’ coupling strategies introduce several sources of error in the solution fields stemming from inaccurate and in some cases\ninconsistent treatment of the spatio-temporal nonlinear coupling terms. Choosing the right software and algorithmic options to resolve the inaccuracies are usually physics and problem dependent and hence a multiphysics framework needs to encompass several coupling schemes with varying stability and accuracy traits to solve dependent systems in order to analyze, compare and choose an optimal strategy. The schemes for solving the coupled systems are similar\nin principle for most approaches and with reusable design principles, accurate resolution of the spatio-temporal scales are possible. Different methodologies and the relevant numerics that are commonly employed for bringing together various physics components will be explored.\n\nFinally, details of a coupling framework in development based on PETSc and MOAB libraries, CouPE (Coupled Physics Environment), for driving existing high-fidelity\nvalidated physics codes will be provided whose design aims to obtain tightly-coupled physical solutions by employing loosely-coupled software interfaces. Some relevant\nresults obtained from nuclear engineering and radiation-diffusion problems with CouPE will be shown. The talk will conclude with proposed enhancements to such a glass-box framework to make it amenable to couple both existing and new physics solvers. SUMMARY: Designing a glass-box, PDE-based coupled multiphysics framework UID:1965 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120108T103000 DTEND;TZID=America/Chicago:20120108T120000 DTSTAMP:20130525T020110 LOCATION:Building 240, Conference Room 1416, Argonne National Laboratory DESCRIPTION:High fidelity mathematical and computer modeling of coupled multiphysics problems require solutions to large systems of stiff, nonlinear, coupled equations. ’Black-box’ coupling strategies introduce several sources of error in the solution fields stemming from inaccurate and in some cases\ninconsistent treatment of the spatio-temporal nonlinear coupling terms. Choosing the right software and algorithmic options to resolve the inaccuracies are usually physics and problem dependent and hence a multiphysics framework needs to encompass several coupling schemes with varying stability and accuracy traits to solve dependent systems in order to analyze, compare and choose an optimal strategy. The schemes for solving the coupled systems are similar\nin principle for most approaches and with reusable design principles, accurate resolution of the spatio-temporal scales are possible. Different methodologies and the relevant numerics that are commonly employed for bringing together various physics components will be explored.\n\nFinally, details of a coupling framework in development based on PETSc and MOAB libraries, CouPE (Coupled Physics Environment), for driving existing high-fidelity\nvalidated physics codes will be provided whose design aims to obtain tightly-coupled physical solutions by employing loosely-coupled software interfaces. Some relevant\nresults obtained from nuclear engineering and radiation-diffusion problems with CouPE will be shown. The talk will conclude with proposed enhancements to such a glass-box framework to make it amenable to couple both existing and new physics solvers. SUMMARY: Designing a glass-box, PDE-based coupled multiphysics framework UID:1967 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120110T103000 DTEND;TZID=America/Chicago:20120110T120000 DTSTAMP:20130525T020110 LOCATION:Building 240 Conference Room 1416, Argonne National Laboratory DESCRIPTION:As the term ``big data\'\' appears more and more frequently in our daily life and research activities, it changes our knowledge of how large the scale of data can be and challenges our traditional practice of data handling. One such challenge occurs in statistical analysis, which drives the extraction of hidden information and assists human understanding of the underlying principles behind the data. Many analysis techniques suffer from the poor scalability of the numerical algorithms and thus pose significant difficulties for their applications to data that are ``big enough.\'\' In this talk, we start from a basic statistics principle---maximum likelihood estimation---to illustrate why traditional numerical calculations fail to provide answers and how novel numerical algorithms (including matrix function evaluation, trace approximation, and the solution of fully dense linear systems) are derived to widen the applicability of the principle to large-scale data in practice. Big data provides a fresh opportunity for numerical analysts to develop algorithms with a central goal of scalability in mind. Accompanied with the increasing computing power of high performance computers, parallelization is of the same importance in this process. The development of scalable and parallelizable numerical algorithms is key for convincing statisticians and data analysts to apply the powerful statistical theories on large-scale data that they currently feel uncomfortable to handle. SUMMARY:Big Data Analytics---A New Opportunity for Numerical Analysis and High Performance Computing UID:1969 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120123T100000 DTEND;TZID=America/Chicago:20120123T110000 DTSTAMP:20130525T020110 LOCATION:Bldg 240 Room 1404, Argonne National Laboratory DESCRIPTION:Predictive models depend on high quality input data, but not all data are of similar quality nor are all of the data amenable to computational analysis without extensive cleaning, interpretation and normalization. Phenotypic data is a prime example. These data are more complex than sequence data, occur in a wide variety of forms, often use non-uniform descriptors that change over time and are scattered about, mainly in the scientific and technical literature or in specialized databases. Incorporating these data into repositories such as the DOE Kbase requires not only expertise in harvesting and modeling the data, but also knowledge in interpreting the data in the correct biological context. While it is generally agreed that access to such data would be invaluable for genome and metagenome analysis, capturing it is a non-trivial undertaking.\n\nWe are approaching this problem in a stepwise fashion, by first creating a standardized terminology of phenotypes for Bacteria and Archaea, derived from the taxonomic literature. To date, we have developed tokenizers that work well on a variety of document and data types that have allowed us to compile a list of approximately 40,000 terms that were used in the published descriptions of 5,750 type strains of Bacteria and Archaea. These terms are being placed into a phenotypic ontology that will be incorporated into the NamesforLife database, from which it will become available for transclusion into community resources such as the Kbase, into machine generated descriptions for publication, and served over top of published literature using annotation services that were initially developed for biological names. Terminological services can also be integrated into end-user applications that will allow for easier capture of phenotypic data that is normalized and persistently linked to the appropriate bio-sample at the earliest possible point in the discovery process. And, like the names and taxonomic concepts applied to the organisms themselves, it becomes feasible to provide this information for other uses as a semantic service via DOIs, automatically resolving any semantic ambiguities that arise over time. It also becomes feasible to index the literature and other digital resources based on broad phenotypic concepts rather than the descriptive terms.\n SUMMARY:Some thoughts on the creation of a phenotypic ontology of prokaryotes UID:1981 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20130123T130000 DTEND;TZID=America/Chicago:20130123T140000 DTSTAMP:20130525T020110 LOCATION:Building 240, Conference Room 4301, Argonne National Laboratory DESCRIPTION:As the scale and complexity of computing systems increases,\napplication programmers must not only be experts in their application domain, but also have the systems knowledge required to address problems arising from parallelism, scale, power, energy, and reliability concerns. Existing computational models are derived from three basic mechanisms -- a mechanism for parallelism, a mechanism for communication and a mechanism for synchronization. Because such models ignore new constraints such as power-efficiency, resilience and ease of programming, they are ill-suited for future scalable systems.\n\nSEEC proposes a radical approach that is capable of meeting these new constraints by adding a self-awareness dimension to computational models. Self-aware computational models automatically adjust their behavior in response to environment stimuli to meet user specified goals and relieve the programmer from having to worry about such issues. This talk presents SEEC, a computational model designed to support the development of self-aware computing systems. The SEEC model is unique in that it supports a decoupled approach where applications and systems developers separately contribute to the development of a self-aware system, each focusing on their area of expertise. Using SEEC, applications directly specify their goals through a standard programming API while system components (e.g. runtime, OS, hardware, & applications themselves) specify possible actions. Given a set of goals and actions, SEEC uses analysis and decision engines (e.g., adaptive feedback control systems\nand machine learning engines) to monitor application progress and select actions to meet goals optimally (e.g. meeting performance goals with minimal power consumption).\n\nThis talk will describe the SEEC computational model and discuss how systems developers and applications writers use SEEC, illustrate several systems built with the model, show results, and discuss future work.\n SUMMARY:SEEC: A Framework for the Self-aware Management of Goals and Constraints in Modern Computing Systems UID:1983 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120128T103000 DTEND;TZID=America/Chicago:20120128T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, Conference Center Room 1404/1405, Argonne National Laboratory DESCRIPTION:Cloud service providers charge customers based upon the amount of resources used or reserved. However, there are no guarantees on the quality-of-service (QoS) that the given resources will provide. Due to the highly dynamic nature of Internet workloads, increasing complexity of cloud-hosted applications, multi-tier architectures and the complex dynamics of underlying shared infrastructure, it is significantly challenging to manage application level performance in the Cloud. The situation is further complicated by the fact that cloud service providers need to control the power consumption in their data centers to avoid power capacity overload of high density servers, lower electricity costs and reduce their carbon footprint.\n\nI have developed novel middleware approaches to autonomic management of virtualized resources for power and performance control in the Cloud. Firstly, I designed self-adaptive and efficient resource provisioning techniques based on machine learning and control theoretical techniques to guarantee a high percentile performance of multi-tier web applications in the face of highly dynamic workloads. I also developed an automation tool for resource allocation and configuration of Hadoop framework for cost-efficient big data processing in the Cloud. Secondly, I designed a system that simultaneously controls the power consumption and the performance of multi-tier applications in a virtualized server system. Furthermore, I developed a power-aware framework for managing scientific workloads in virtualized GPU computing environments, with the help of emerging technologies that provide GPU accelerators as virtualized computing resources. Lastly, I proposed and developed non-invasive, energy efficient and highly scalable mechanisms to achieve performance isolation of heterogeneous applications in a multi-tenant cloud system.\n SUMMARY:Probabilistic Fault Detection and Diagnosis in Large-Scale HPC Applications UID:1991 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20130220T150000 DTEND;TZID=America/Chicago:20130220T160000 DTSTAMP:20130525T020110 LOCATION:Building 240, 4301, Argonne National Laboratory DESCRIPTION:Quantum chemistry faces many challenges on petascale supercomputers due to a changing landscape of processing architecture and the enormous scale of these systems. Over the past 10 years, systems have grown in size by a factor of a thousand in some cases and within a node there is significantly more parallelism. This talk will start by considering the NWChem software package and it\'s coupled-cluster (CC) codes, which are scalable to 100,000 cores on Cray supercomputers and have rich scientific capability. From here, we consider three research projects to develop new coupled-cluster algorithms for modern supercomputers. The first of these is the development of CC for GPUs, where we observe approximately an order-of-magnitude speedup with respect to existing codes for CCSD at workstation scale. The second project is a complete redesign of tensor contraction algorithms using novel techniques from dense linear algebra. This completely eliminates load-imbalance and allows for topology-aware mapping on systems with torus interconnects (e.g. Blue Gene and \"K\"). Finally, we demonstrate how inspector-executor techniques can be used to eliminate dynamic task scheduling in NWChem CC codes, which can lead to a substantial reduction in time-to-solution.\n\nThis work has been done in collaboration with Eugene DePrince (Georgia Tech), Edgar Solomonik (Berkeley), Devin Matthews (Texas), David Ozog (Oregon), Pavan Balaji (Argonne) and Jim Dinan (Argonne). SUMMARY:Algorithms and Software for Quantum Chemistry at Petascale and Beyond UID:1997 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:00000000T150000 DTEND;TZID=America/Chicago:00000000T160000 DTSTAMP:20130525T020110 LOCATION:Building 240, 1404-1405, Argonne National Laboratory DESCRIPTION:TBA SUMMARY:TBA UID:1999 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20130313T150000 DTEND;TZID=America/Chicago:20130313T160000 DTSTAMP:20130525T020110 LOCATION:Building 240, 1406-1407, Argonne National Laboratory DESCRIPTION:Scalable data analysis and visualization (collectively called analysis) depends on lightweight, custom tasks that can be tightly integrated with both computational science applications and existing analysis tools. DIY (Do-it-Yourself Analysis) is a scalable library of data-movement algorithms for domain decomposition, parallel I/O, and efficient communication that permits data analysis to be parallelized and executed as a data-parallel program. Parallel analyses are then executed in situ with full-scale simulations and in tandem with existing visualization and analysis packages. DIY has enabled the parallelization of serial analysis algorithms previously considered difficult to scale efficiently, or for which no parallel counterpart existed. These include parallel particle tracing for steady and unsteady flows, information entropic analysis, topological construction, computation of Lagrangian coherent structures, and mesh generation from N-body particle simulations. This seminar covers the basics of DIY: data decomposition into blocks, assignment of blocks to processes, support for multiple domains, grouping of blocks into neighborhoods, creation of custom DIY datatypes, communication mechanisms, and integration with other libraries. SUMMARY:Do-It-Yourself Parallel Data Analysis UID:2001 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120215T143000 DTEND;TZID=America/Chicago:20120215T153000 DTSTAMP:20130525T020110 LOCATION:BUILDING 240/CONFERENCE ROOM 1404, Argonne National Laboratory DESCRIPTION:Microbial electrosynthesis is the conversion of CO2­ to chemicals by autotrophic microorganisms using reducing equivalents from a negatively poised electrode. The technology is in the nascent stage of development, but has the potential to be a sustainable, carbon-consuming platform for industrial chemical and fuel production. To date, microbial electrosynthesis of acetate, the first step in order to generate liquid fuels from CO2, has been characterized by low rates and yields. In order for microbial electrosynthesis systems to become commercially relevant, long-term operation and improved rates must be demonstrated. To address these issues, bioelectrochemical reactors were operated chronoamperometrically at a cathode potential of -590 mV vs. SHE for over 200 days. Hydrogen, acetate, and methane were produced by adaptively evolved microbiomes at rates 10-100x what has been previously reported. Other products observed were formate, propionate, and butyrate. Acetobacterium spp. dominated the active microbial population on the cathodes (50-60%), with members of Sulfurospirillum and the Rhodobacteraceae also prevalent. Further experimental results demonstrate stability, resilience, and improved performance of electrosynthetic biocathodes following long-term operation. Sustained product formation at faster rates by a carbon-capturing microbiome is a key milestone addressed in this work that advances microbial electrosynthesis systems towards commercialization. SUMMARY:Microbial Electrosynthesis by Autotrophic Microbiomes UID:2003 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120208T103000 DTEND;TZID=America/Chicago:20120208T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, Conference Center, Room 1406 , Argonne National Laboratory DESCRIPTION:There is an ever-increasing need for exploring large-scale graph data sets in computational sciences, social networks, and business analytics. One of the most widely used graph searching algorithms is breadth-first search (BFS), which serves as a building block for a great many graph algorithms such as minimum spanning tree, betweenness centrality, and shortest paths. However, BFS is notoriously known challenging to optimize because of its poor locality and low computation to communication ratio. In this talk I will present several of our findings to improve the efficiency of BFS algorithms: exploiting different levels of parallelism to provide latency hiding and maximize memory bandwidth; reducing communication in distributed BFS to improve its scalability. SUMMARY:Large-Scale Graph Traversal Algorithms on Multi-core Clusters UID:2005 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120111T100000 DTEND;TZID=America/Chicago:20120111T110000 DTSTAMP:20130525T020110 LOCATION:Bldg. 240, TCS Room 1416, Argonne National Laboratory DESCRIPTION:Korea Meteorological Administration (KMA) launched a 9-year project in 2011 to develop Korea’s own global NWP system with the total funding of about 100 million US dollars. To lead the effort, Korea Institute of Atmospheric Prediction Systems (KIAPS) was established by KMA as a non-profit foundation. As implied by the name of the institute, the main goal of the project is to develop a global operational NWP system and to replace the current KMA NWP system (UK Met Office Unified Model system) with Korea’s own system. The project is planned to be carried out in three stages. The first stage (2011-2013) is a period to set up the\ninstitute, to recruit researchers, to make plans for research and development, and to design the basic structure and explore/develop NWP-related basic technologies. The second stage (2014-2016) aims at developing the modules for dynamical core, physical parameterizations, and data assimilation systems as well as the modeling system framework and couplers to connect the core modules in a systematic and efficient way, and eventually building a prototype NWP system. At the third stage (2017-2019), the prototype NWP system will be evaluated, refined, and tuned for operational use at KMA as well as developing post-processing systems. At seminar, the organization of KIAPS and 9-year roadmap are briefly introduced, action plans for the development of NWP system are presented. SUMMARY:Global NWP Model Development at KIAPS UID:2007 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20130306T150000 DTEND;TZID=America/Chicago:20130306T160000 DTSTAMP:20130525T020110 LOCATION:Building 240, 4301, Argonne National Laboratory DESCRIPTION:M2ACS is a new research project, involving 22 senior investigators from 5 institutions. Its central research concept is the one of multi-faceted mathematics -- whereby new mathematics is discovered by sustained investigation of the multiple facets presented by the grand challenge subproblems, such as nonconvexity, stochasticity, integrality, spatio-temporal variability -- from the perspective of all the mathematics discipline area involved. We will discuss the project scope, some of its challenges and how we anticipate it to impact mathematics and complex energy systems practice. SUMMARY:The Multifaceted Mathematics for Complex Energy Systems (M2ACS) Project UID:2017 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20130218T103000 DTEND;TZID=America/Chicago:20130218T113000 DTSTAMP:20130525T020110 LOCATION:Building 240 / Conference Room 4301, Argonne National Laboratory DESCRIPTION:MapReduce parallel programming model has been widely adopted, including scientific data analysis and management. Recently, lightweight, fast, in-memory MapReduce runtime systems have been proposed for shared memory systems. Such in-memory MapReduce runtime systems have the potential to alleviate the parallel programming challenges. However, what factors affect performance and what performance bottlenecks exist for a given program are not well understood. In this talk, I will present a practical performance model that captures key performance factors, important trends, and behavior of in-memory MapReduce on multi-core architectures. I will discuss how our analytical model discovers several important findings and implications for system designers, performance tuners and programmers.\n\nIf time permits, I will share my experiences in applying analytical models for understanding performance and energy trade-offs in other execution paradigms, in particular performing data analysis on emerging storage devices such as Solid State Drives (SSDs). I will show that how analytical models can be used to understand the feasibility and design trade-offs in designing such systems.\n SUMMARY:A Performance Modeling Approach for Analyzing In-Memory MapReduce Workloads on Multi-core Architecture UID:2011 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120218T103000 DTEND;TZID=America/Chicago:20120218T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, Conf. Center, Rooms 1406 and 1407, Argonne National Laboratory DESCRIPTION:Scientific simulations and experiments are producing increasingly large so-called multifield datasets that are comprised of scalar, vector, and tensor fields tracked over time. The visualization literature provides a variety of surface-based feature characterizations. While these models have proven extremely valuable in the analysis of individual fields, their use in practice is hindered by two major challenges.\n\nFirst, the extraction of these structures is often non-trivial as their precise computation requires careful analysis of the spatial variation of convoluted functions and their nonlinear invariants. Examples of such characterizations include so-called crease surfaces and level sets in nonlinear fields. In addition, these surfaces can be non-orientable, with a nontrivial topology, and a complex set of boundaries. To address these issues we have introduced a novel scalable front propagation strategy that offers strong guarantees on the approximation quality. Our solution is suitable for arbitrarily complex surfaces, and it ensures an excellent geometric quality for the produced mesh.\n\nA second, and arguably more severe limitation of these structure definitions is that they do not support the visual analysis of multifield datasets. In that context, we have proposed a geometrically motivated, multifield feature definition that characterizes the relationship between the features exhibited by the individual fields in the spatial domain of the problem. Our algorithm leverages the existing theory of skeleton derivation to simplify and fuse the surface structures from the constitutive fields into a coherent and visually effective data description. This work also introduced a new method for non-rigid surface registration tailored to the specific nature of surfaces extracted from computational fluid dynamics datasets. Temporal matching and spatial clustering enable the discovery of subtle interaction patterns between the different fields and their evolution over time.\n\nFinally, we document the unified visual analysis achieved by our methods in the context of several multifield problems from large-scale time-varying simulations, and we discuss future research directions. SUMMARY:Fast Extraction and Analysis of Surface-based Structures UID:2013 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20130213T150000 DTEND;TZID=America/Chicago:20130213T160000 DTSTAMP:20130525T020110 LOCATION:Building 240, 1404-1405, Argonne National Laboratory DESCRIPTION:Simulation and analysis of stiff PDE requires robust and algorithmically scalable solvers. Multilevel methods are required to provide this algorithmic scalability, but appear in various forms due to additional problem structure, physical regime, coupling strengths, discretizations, problem size, and hardware. I will discuss software components developed in PETSc to support combining multilevel methods with equation splitting and multirate structure, as well as recent advances in robustness and generality, performance implications of an extremely low-communication variant of multigrid, and a new approach to resilience that enables local reconstruction of fine-grid state from arbitrarily coarse checkpoints. SUMMARY:Composability, performance, and resilience in multilevel solvers UID:2015 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20130222T100000 DTEND;TZID=America/Chicago:20130222T110000 DTSTAMP:20130525T020110 LOCATION:Building 240, 1404-1405, Argonne National Laboratory DESCRIPTION:The talk will overview two prototypical problems of multiscale computational science for which a mathematical approach can help. Both topics are related to materials science in a broad sense. The first topic is multiscale-in-space in nature: random composite materials are often prohibitively expensive to simulate numerically. A category of new, efficient numerical approaches that elaborates on the possibly small disorder present in such materials will be described. The second topic is multiscale-in-time in nature: the efficiency of the parallel replica method, simulating long term dynamics of potentially metastable evolutions, will be examined mathematically. The talk is joint work with a number of colleagues from Ecole des Ponts, INRIA, and several other institutions. SUMMARY:Two mathematical topics in mutiscale science: random composites and metastable states UID:2019 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20130416T080000 DTEND;TZID=America/Chicago:20130416T000008 DTSTAMP:20130525T020110 LOCATION:TCS Conference Center, Argonne National Laboratory DESCRIPTION:GlobusWORLD is the annual gathering for users and developers of Globus technologies, including the Globus Toolkit and Globus Online. This year\'s conference focuses on \"moving, syncing, and sharing\" research data at scale. We invite research computing center and campus cluster administrators, as well as HPC resource owners and managers to see how Globus services can simplify data management for end users. SUMMARY:GlobusWORLD 2013 - Big Data: Move it. Sync it. Share it. UID:2021 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120312T103000 DTEND;TZID=America/Chicago:20120312T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, Seminar Room 4301, Argonne National Laboratory DESCRIPTION:In high energy Li-ion batteries, Li deposits formed on the anode surface during cell charging can cause poor cycling and sever safety problem. The formation of Li deposits leads to a large decrease of reversible capacity and worst a short-circuiting phenomenon as deposits grow towards to the cathode. Even for the graphite anode instead of lithium metal there is still a great chance for lithium to be deposited when Li-ion batteries undergo a high rate charge. During lithium electrodeposition processes, it is believed that the morphology and growth of electrodeposits are mainly determined by the kinetics of the heterogeneous electrode reaction, electrode surface states, Ohmic potential drop and mass species transports inside the bulk materials.\n In this talk, I will present a nonlinear phase field model for predicting electrode-electrolyte interface motion and microstructure evolution during the electrochemical deposition involving highly nonequilibrium processes. Without considering the solid-electrolyte interface (SEI) layer effect, this model is able to simulate and predict the lithium deposits formation and growth in Li-ion batteries during charging operations. The electrodeposition rate implicitly follows the Butler-Volmer kinetic with a diffuse electrode-electrolyte interface description. The consuming of Li+ concentration and electric potential drop across the interface is correlated to the electrochemical reaction kinetic. The local variations of ion concentration and overpotential cause the instability of electrode surface which eventually produce fiber-like lithium deposits growth. The effects of charged current density and reaction rate constant on the deposits morphology are discussed in details. SUMMARY:Phase Field Modeling of Lithium Electrodeposition in Lithium-Ion Batteries UID:2027 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120313T103000 DTEND;TZID=America/Chicago:20120313T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, Seminar Room 4301, Argonne National Laboratory DESCRIPTION:For every class of polycrystalline materials, the scientific study of grain boundaries as well as the increasingly widespread practice of grain boundary engineering relies heavily on visual representation for the analysis of boundary statistics and their connectivity. Traditional methods of grain boundary representation neglect the full complexity of misorientation information and often rely on boundary classification schemes of dubious physical significance. This talk will highlight my recent work toward significantly advancing our ability to represent grain boundary information. Beginning with an understanding of the topology of the group space of misorientations, we developed new methods of mapping and visualizing boundary character. This new technique allows micrographs or maps of grain boundaries to be presented in a colorized form which, at a glance, reveal all of the misorientation information in an entire grain boundary network, as well as the connectivity among different boundary misorientations. Finally, I will introduce the extension of recently developed hyperspherical harmonic formulation for the description of grain boundary misorientation and boundary-plane statistics. These new and improved methods of representing grain boundary information, which are already being incorporated in commercial software packages, are expected to be powerful tools for grain boundary network analysis as the practice of grain boundary engineering becomes a routine component of the materials design paradigm. SUMMARY:Theory as a Tool for Microstructural Design: Grain Boundary Engineering UID:2029 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120319T103000 DTEND;TZID=America/Chicago:20120319T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, Conference Room 4301, Argonne National Laboratory DESCRIPTION:Selecting meaningful features is central in the analysis of scientific data. Today\'s multivariate scientific datasets are often large and complex making it difficult to define general features of interest significant to scientific applications.\nTo address this problem, we propose three general, spatiotemporal metrics to quantify the significant properties of data features--concentration, continuity and co-occurrence, named collectively as CO3. We implemented an interactive visualization system to investigate complex multivariate time-varying data from satellite remote sensing with great spatial resolutions, as well as from real-time continental-scale power grid monitoring with great temporal resolutions. The system integrates CO3 metrics with an elegant multi-space user interaction tool to provide various forms of quantitative user feedback. Through these, the system supports an iterative user-driven analysis process. Our findings demonstrate that the CO3 metrics are useful for simplifying the problem space and revealing potential unknown possibilities of scientific discoveries by assisting users to effectively select significant features and groups of features for visualization and analysis. Users can then comprehend the problem better and design future studies using newly discovered scientific hypotheses.\n\nBio:\n\nBio:\nJian Huang is an associate professor in the Department of Electrical Engineering and Computer Science and also associate director of the NSF funded remote data analysis and visualization center at the University of Tennessee, Knoxville. His research focuses on large data visualization, multivariate visualization, and parallel, distributed and remote visualization. He received a BEng in electrical engineering from the Nanjing University of Posts and Telecom, China in 1996. During his subsequent graduate study at the Ohio State University, he was awarded an MS degree in computer science and an MS degree in biomedical engineering, both in 1998, and a PhD degree in computer science in 2001. Dr. Huang\'s research has been funded by National Science Foundation and Department of Energy. Dr. Huang is a recipient of DOE Early Career Principal Investigator Award in 2004.\n SUMMARY:Interactive Selection of Multivariate Features in Large Spatiotemporal Data UID:2031 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20130318T150000 DTEND;TZID=America/Chicago:20130318T160000 DTSTAMP:20130525T020110 LOCATION:Building 240, 1406-1407, Argonne National Laboratory DESCRIPTION:The Deepwater Horizon (DwH) was not only the largest oil spill along the US coast, but perhaps also the most important oceanographic event over the past several decades. Numerical modeling and prediction of pollutant transport require a good understanding of the nature of multi-scale processes in the upper ocean. The Consortium for Advanced Research of Transport of Hydrocarbons in the Environment (CARTHE, carthe.org) have conducted the largest simultaneous drifter sampling attempted in oceanography to date in order to find out the scale-dependent dispersion characteristics near the DwH site. Upper ocean boundary layer simulations obtained with Nek5000 have been used to estimate the numbers and durations of drifter trajectories needed for this oceanic experiment. A total of 317 drifters, an order of magnitude larger than typical, have been deployed during so-called Grand Lagrangian Deployment (GLAD). The primary finding of GLAD is that dispersion depends on local processes in the region, in contrast with results from operational ocean models that indicate the dominance of various large eddies. Several other oceanic applications of Nek5000 will also be discussed. SUMMARY:On the Nature of Dispersion in the Ocean UID:2035 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120315T103000 DTEND;TZID=America/Chicago:20120315T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, Conf. Center, Rooms 1406 and 1407, Argonne National Laboratory DESCRIPTION:The Message Passing Interface (MPI) is an extremely successful parallel programming model, and it is expected to be a core component for exascale applications. In spite of its success, there are growing gaps between MPI and system and application architectures. In this talk, I will describe current and ongoing efforts to advance the state of the art in scalable parallel programming models through MPI. I will discuss the recently released MPI 3.0 standard, with a focus on the new one-sided communication interface. This interface enables a broad range asynchronous data access techniques and provides portable support for globally accessible distributed data, two parallel programming idioms that will be of increasing importance on future systems. SUMMARY:How I Learned to Stop Worrying About Exascale and Love MPI UID:2037 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120423T083000 DTEND;TZID=America/Chicago:20120423T170000 DTSTAMP:20130525T020110 LOCATION:Searle 240, University of Chicago DESCRIPTION:For Charles Darwin, the HMS Beagle was the ship for the historic voyage in 1831 that inspired his later work on the origin of species. For the University of Chicago, \"Beagle\" is an 18,000-core Cray XE6 supercomputer that is one of the most powerful supercomputers dedicated to biomedical research. Since its launch in early 2011, Beagle has supported over 80 research projects, more than 300 users and has resulted in over 30 publications. On April 23rd, a \"Day of the Beagle\" symposium will discuss the innovative research made possible to date by Beagle, with speakers representing the fields of genomics, neurobiology, molecular biology and medicine. \n\n8:30 – 9:00 Continental breakfast\n\n9:00 – 9:30 Ian Foster & Conrad Gilliam\nOverview of the Cray Beagle and Diversity of Science on Beagle\n\n9:30 – 10:15 Kazutaka Takahashi & Nicholas Hatsopoulos (PI) \nComputation and analysis of time varying large-scale networks of spiking neurons in primary motor cortex using massively-parallel supercomputers\n\n10:15 – 10:30 Coffee Break\n\n10:30 – 11:15 Yves Lussier (PI) & Haiquan Li \nIntergenic DNA’s role in complex disease unveiled by supercomputer-obligated analyses of BIG data\n\n11:15 – 12:00 Greg Tietjen & Ka Yee Lee (PI) \nElucidating the molecular details of phosphatidylserine membrane recognition in immune response\n\n12:00 – 13:00 Lunch \n\n12:45 – 13:00 Esmael Jafari Haddadian \nUsing Beagle in an actual undergraduate biology course\n\n13:00 – 13:45 Benoit Roux (PI) \nTheoretical studies of biomolecular systems on Beagle: highly-scalable computational strategies\n\n13:45 – 14:30 Edwin Munro (PI), TaeYoon Kim\nUsing agent-based models and supercomputing to probe self-organized contractility and stress dissipation in active cytoskeletal networks\n\n14:30 – 14:45 Coffee Break\n\n14:45 – 15:30 Jason J. Pitt, Chai Bandlamudi & Kevin White (PI) \nGenome sequence analysis workflows\n\n15:30 – 16:15 Elizabeth McNally (PI), Megan Puckelwartz & Lorenzo Pesce\nSupercomputing for the parallelization of whole genome sequence analysis\n\n16:15 – 16:45 Carlos Sosa (Cray)\nCray Supercomputers – Life Sciences \n\n16:45 – 17:00 Ian Foster – Closing Remarks / Future Directions\n SUMMARY:Scientific Discovery on the Computation Institute/BSD Cray “Beagle” Supercomputer UID:2043 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120423T083000 DTEND;TZID=America/Chicago:20120423T170000 DTSTAMP:20130525T020110 LOCATION:Searle 240, University of Chicago DESCRIPTION:For Charles Darwin, the HMS Beagle was the ship for the historic voyage in 1831 that inspired his later work on the origin of species. For the University of Chicago, \"Beagle\" is an 18,000-core Cray XE6 supercomputer that is one of the most powerful supercomputers dedicated to biomedical research. Since its launch in early 2011, Beagle has supported over 80 research projects, more than 300 users and has resulted in over 30 publications. On April 23rd, a \"Day of the Beagle\" symposium will discuss the innovative research made possible to date by Beagle, with speakers representing the fields of genomics, neurobiology, molecular biology and medicine. \n\n8:30 – 9:00 Continental breakfast\n\n9:00 – 9:30 Ian Foster & Conrad Gilliam\nOverview of the Cray Beagle and Diversity of Science on Beagle\n\n9:30 – 10:15 Kazutaka Takahashi & Nicholas Hatsopoulos (PI) \nComputation and analysis of time varying large-scale networks of spiking neurons in primary motor cortex using massively-parallel supercomputers\n\n10:15 – 10:30 Coffee Break\n\n10:30 – 11:15 Yves Lussier (PI) & Haiquan Li \nIntergenic DNA’s role in complex disease unveiled by supercomputer-obligated analyses of BIG data\n\n11:15 – 12:00 Greg Tietjen & Ka Yee Lee (PI) \nElucidating the molecular details of phosphatidylserine membrane recognition in immune response\n\n12:00 – 13:00 Lunch \n\n12:45 – 13:00 Esmael Jafari Haddadian \nUsing Beagle in an actual undergraduate biology course\n\n13:00 – 13:45 Benoit Roux (PI) \nTheoretical studies of biomolecular systems on Beagle: highly-scalable computational strategies\n\n13:45 – 14:30 Edwin Munro (PI), TaeYoon Kim\nUsing agent-based models and supercomputing to probe self-organized contractility and stress dissipation in active cytoskeletal networks\n\n14:30 – 14:45 Coffee Break\n\n14:45 – 15:30 Jason J. Pitt, Chai Bandlamudi & Kevin White (PI) \nGenome sequence analysis workflows\n\n15:30 – 16:15 Elizabeth McNally (PI), Megan Puckelwartz & Lorenzo Pesce\nSupercomputing for the parallelization of whole genome sequence analysis\n\n16:15 – 16:45 Carlos Sosa (Cray)\nCray Supercomputers – Life Sciences \n\n16:45 – 17:00 Ian Foster – Closing Remarks / Future Directions\n SUMMARY:Scientific Discovery on the Computation Institute/BSD Cray “Beagle” Supercomputer UID:2045 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120423T083000 DTEND;TZID=America/Chicago:20120423T170000 DTSTAMP:20130525T020110 LOCATION:Searle 240b, University of Chicago DESCRIPTION:For Charles Darwin, the HMS Beagle was the ship for the historic voyage in 1831 that inspired his later work on the origin of species. For the University of Chicago, \"Beagle\" is an 18,000-core Cray XE6 supercomputer that is one of the most powerful supercomputers dedicated to biomedical research. Since its launch in early 2011, Beagle has supported over 80 research projects, more than 300 users and has resulted in over 30 publications. On April 23rd, a \"Day of the Beagle\" symposium will discuss the innovative research made possible to date by Beagle, with speakers representing the fields of genomics, neurobiology, molecular biology and medicine. \n\nScientific Discovery on the Computation Institute/BSD Cray “Beagle” Supercomputer\n\n8:30 – 9:00 Continental breakfast\n\n9:00 – 9:30 Ian Foster & Conrad Gilliam\nOverview of the Cray Beagle and Diversity of Science on Beagle\n\n9:30 – 10:15 Kazutaka Takahashi & Nicholas Hatsopoulos (PI) \nComputation and analysis of time varying large-scale networks of spiking neurons in primary motor cortex using massively-parallel supercomputers\n\n10:15 – 10:30 Coffee Break\n\n10:30 – 11:15 Yves Lussier (PI) & Haiquan Li \nIntergenic DNA’s role in complex disease unveiled by supercomputer-obligated analyses of BIG data\n\n11:15 – 12:00 Greg Tietjen & Ka Yee Lee (PI) \nElucidating the molecular details of phosphatidylserine membrane recognition in immune response\n\n12:00 – 13:00 Lunch \n\n12:45 – 13:00 Esmael Jafari Haddadian \nUsing Beagle in an actual undergraduate biology course\n\n13:00 – 13:45 Benoit Roux (PI) \nTheoretical studies of biomolecular systems on Beagle: highly-scalable computational strategies\n\n13:45 – 14:30 Edwin Munro (PI), TaeYoon Kim\nUsing agent-based models and supercomputing to probe self-organized contractility and stress dissipation in active cytoskeletal networks\n\n14:30 – 14:45 Coffee Break\n\n14:45 – 15:30 Jason J. Pitt, Chai Bandlamudi & Kevin White (PI) \nGenome sequence analysis workflows\n\n15:30 – 16:15 Elizabeth McNally (PI), Megan Puckelwartz & Lorenzo Pesce\nSupercomputing for the parallelization of whole genome sequence analysis\n\n16:15 – 16:45 Carlos Sosa (Cray)\nCray Supercomputers – Life Sciences \n\n16:45 – 17:00 Ian Foster – Closing Remarks / Future Directions\n SUMMARY:Scientific Discovery on the Computation Institute/BSD Cray “Beagle” Supercomputer UID:2047 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120320T103000 DTEND;TZID=America/Chicago:20120320T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, Seminar Room 4301, Argonne National Laboratory DESCRIPTION:Inverse problems are ubiquitous in science and engineering, especially in the field of Geoscience. Algorithms that we have developed are applicable to seismic imaging, contaminant source identification, hydraulic tomography, etc. This work tackles these inverse problems using the Geostatistical approach that stochastically models unknowns as random fields. However, due to high computational costs in identifying small scale features, these methods are challenging. In addition, it is necessary to quantify the corresponding predictive uncertainty to provide a sound basis for management or policy decision making. My approach uses Hierarchical matrices to efficiently represent dense covariance matrices and solves the resulting intermediary system of equations using preconditioned iterative methods. The resulting cost is reduced from O(N^2) to O(N log N), where N is the number of unknowns to be determined. In addition, I will describe some recent work on Krylov solvers for shifted systems that accelerates the solution of \"forward problems\" and as a result, the inverse problem. Moreover, a technique to perform uncertainty quantification by estimating the diagonals of the posterior covariance matrix will be presented. I will illustrate these algorithms with examples from contaminant source identification and hydraulic tomography. SUMMARY:Fast algorithms for inverse problems and uncertainty quantification UID:2051 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120321T103000 DTEND;TZID=America/Chicago:20120321T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, Seminar Room 4301, Argonne National Laboratory DESCRIPTION:In this talk we outline the challenges facing typical linear inverse problems from physical sciences: limited amount of data compared to the number of unknowns in the linear system, noise in the right hand side data, poor conditioning of the matrices and shortage of computational resources for the large data sets that are typically encountered. We discuss algorithms and practical strategies to address these problems and illustrate with examples from an inverse problem in geotomography. We start with classical $\\ell_2$ based minimization and go on to discuss different classes of algorithms (convex and non-convex) for sparsity constrained minimization motivated by theoretical results in compressive sensing. We then discuss approaches for mixed penalty regularization and the use of wavelet based constraints. \n\nFinally, we turn our attention to computation and discuss practical strategies for implementing the algorithms discussed with very large matrices using techniques such as wavelet based compression and randomized low rank singular value decompositions. We conclude by discussing additional applications and how the mentioned techniques can take advantage of today\'s multiprocessor/core architectures. SUMMARY:Linear Inverse Problems: Regularization Strategies and Their Practical Implementation UID:2055 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20120326T103000 DTEND;TZID=America/Chicago:20120326T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, Seminar Room 4301, Argonne National Laboratory DESCRIPTION:This talk will present recent progress related to the theory and applications of the multilevel optimization approach (MG/OPT). The intent of MG/OPT is to use calculations on coarser levels to accelerate the progress of the optimization on the finest level. Significant speedup by using MG/OPT comparing to other existing techniques has shown for a particular type of tessellation problems called centroidal Voronoi tessellations(CVTs) and a variety of other constrained optimization problems. SUMMARY:Novel Optimization-Based Multilevel Framework on CVT Construction UID:2057 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20130326T103000 DTEND;TZID=America/Chicago:20130326T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, Seminar Room 4301, Argonne National Laboratory DESCRIPTION:This talk will present recent progress related to the theory and applications of the multilevel optimization approach (MG/OPT). The intent of MG/OPT is to use calculations on coarser levels to accelerate the progress of the optimization on the finest level. Significant speedup by using MG/OPT comparing to other existing techniques has shown for a particular type of tessellation problems called centroidal Voronoi tessellations(CVTs) and a variety of other constrained optimization problems. SUMMARY:Novel Optimization-Based Multilevel Framework on CVT Construction UID:2059 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20130326T150000 DTEND;TZID=America/Chicago:20130326T160000 DTSTAMP:20130525T020110 LOCATION:Building 240, 7172, Argonne National Laboratory DESCRIPTION:We present a mechanism for describing and solving collections of optimization problems that are linked by equilibrium conditions. Included in this class are classical models such as the PIES model and agent based formulations arising from Nash Games. We demonstrate this mechanism in the context of energy planning problems, specifically for capacity expansion, hydro operation, and transmission line switching. We show how to incorporate stochastic information into these systems and give examples of their use and their possible extensions to hierarchical modeling. SUMMARY:MOPEC: Multiple Optimization Problems with Equilibrium Constraints UID:2063 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20130403T150000 DTEND;TZID=America/Chicago:20130403T160000 DTSTAMP:20130525T020110 LOCATION:Building 240, 1404-1405, Argonne National Laboratory DESCRIPTION:Advances in optical communication and networking technologies, together with the computing and storage technologies, are dramatically changing the ways scientific research is conducted. A new term, e-Science, has emerged to describe “large-scale science carried out through distributed global collaborations enabled by networks, requiring access to very large scale data collections, computing resources, and high-performance visualization\". Many large-scale scientific and commercial applications produce large amounts of data, of the order of terabytes to petabytes, which must be transferred across wide-area networks. These e-Science application workflows may be complex and require schedulable and high-bandwidth connectivity with known future characteristics. Moreover, these workflows have performance requirements or metrics that have not been considered by conventional networking. For example, large file transfer may need guaranteed total turnaround time and the rate of progress. Given the long duration of many requests, the network resources available may change before it is completed. This seminar covers problems ranging from basic bandwidth scheduling and path computation problems to complex workflow scheduling problems. SUMMARY:Network-centric resource scheduling in e-Science networks UID:2065 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20130319T103000 DTEND;TZID=America/Chicago:20130319T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, Seminar Room 4301, Argonne National Laboratory DESCRIPTION:Selecting meaningful features is central in the analysis of scientific data. Today\'s multivariate scientific datasets are often large and complex making it difficult to define general features of interest significant to scientific applications. To address this problem, we propose three general, spatiotemporal metrics to quantify the significant properties of data features--concentration, continuity and co-occurrence, named collectively as CO3. We implemented an interactive visualization system to investigate complex multivariate time-varying data from satellite remote sensing with great spatial resolutions, as well as from real-time continental-scale power grid monitoring with great temporal resolutions. The system integrates CO3 metrics with an elegant multi-space user interaction tool to provide various forms of quantitative user feedback. Through these, the system supports an iterative user-driven analysis process. Our findings demonstrate that the CO3 metrics are useful f or simplifying the problem space and revealing potential unknown possibilities of scientific discoveries by assisting users to effectively select significant features and groups of features for visualization and analysis. Users can then comprehend the problem better and design future studies using newly discovered scientific hypotheses.\n\nBio:\nJian Huang is an associate professor in the Department of Electrical Engineering and Computer Science and also associate director of the NSF funded remote data analysis and visualization center at the University of Tennessee, Knoxville. His research focuses on large data visualization, multivariate visualization, and parallel, distributed and remote visualization. He received a BEng in electrical engineering from the Nanjing University of Posts and Telecom, China in 1996. During his subsequent graduate study at the Ohio State University, he was awarded an MS degree in computer science and an MS degree in biomedical engineering, both in 1998, and a PhD degree in computer science in 2001. Dr. Huang\'s research has been funded by National Science Foundation and Department of Energy. Dr. Huang is a recipient of DOE Early Career Principal Investigator Award in 2004. SUMMARY:Interactive Selection of Multivariate Features in Large Spatiotemporal Data UID:2067 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20130605T150000 DTEND;TZID=America/Chicago:20130605T160000 DTSTAMP:20130525T020110 LOCATION:Building 240, 1404-1405, Argonne National Laboratory DESCRIPTION:TBA SUMMARY:Computational Challenges in Energy Storage Research UID:2069 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20130118T103000 DTEND;TZID=America/Chicago:20130118T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, TCS Conference Room 1404, Argonne National Laboratory DESCRIPTION:Cloud computing represents an opportunity for a powerful convergence in which the societal investment in massive HPC systems can be brought to bare in a general purpose fashion. Cloud computing provides arbitrary consumers on-demand access to the aggregate computational power of data-center scale systems with compelling economic advantages. These clouds will increasingly adopt the kind of highly integrated and balanced systems pioneered by and used in HPC. Reflexively, cloud economics and the potential for broad utility and amortization can help foster the development and deployment of extreme scale systems. In this talk we describe two aspects of our work in exploring this convergence. In the first part we describe the Massachusetts Open Cloud (MOC), a new \"open cloud\" concept targeting big data and HPC applications. In the second part of the talk we describe our work on a new system software architecture and runtime that targets the challenges associated with developing software that can efficiently leveraging such systems for a broad spectrum of applications and use cases. We describe this system software effort in the context of the lessons from our past systems research. SUMMARY:Cloud and HPC System Convergence UID:2071 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20130603T110000 DTEND;TZID=America/Chicago:20130603T120000 DTSTAMP:20130525T020110 LOCATION:Building 240 Conference Room 1404-1405, Argonne National Laboratory DESCRIPTION:First, a short outlook on the research at IFP Energies nouvelles in the domain of piston engine CFD will be given. Research activities range from system simulation to LES, with model development activities addressing all related aspects as aerodynamics, two-phase flows, combustion, polluntant formation and exhaust gas after-treatment.\n\nThe main part of the presentation will then concern the development of the LES code AVBP and related models for studying non-cyclic phenomena in piston engines. In the last 10 years, LES has become a key topic for exploring these phenomena hitherto not addressed in CFD. This is for one related to the development of massively parallel computers that made such simulations possible, and a direct consequence of the unique potential LES has to address the impact of local instantaneous flow phenomena on the operation of complex devices as piston engine or gas turbines.\n\nAfter recalling some key requirements on numerics in relation with LES, the application of LES and the AVBP code to study cyclic variability in spark-ignition engines will be presented. Emphasis will be put on how it allows identifying their sources, and how this knowledge is used at IFPEN to formulate system simulation models able to reproduce them. Then, the development of a LES model for Diesel combustion will be presented, and its application to a Diesel jet studied in the frame of the Engine Combustion Network. will be discussed.\n\nPerspectives far LES research and for applying LES to industrial subjects will close the presentation. SUMMARY:CFD research at IFPEN, with a focus on exploring non-cyclic piston engine combustion with Large-Eddy Simulation. UID:2073 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20130409T103000 DTEND;TZID=America/Chicago:20130409T113000 DTSTAMP:20130525T020110 LOCATION:Building 240 Conference Room 4301, Argonne National Laboratory DESCRIPTION:The human brain remains as one of the great frontiers of science – how does this organ upon which we all depend so critically, actually do its job? A great deal is known about the underlying technology – the neuron – and we can observe large-scale brain activity through techniques such as magnetic resonance imaging, but this knowledge barely starts to tell us how the brain works. Something is happening at the intermediate levels of processing that we have yet to begin to understand, but the essence of the brain\'s information processing function probably lies in these intermediate levels. To get at these middle layers requires that we build models of very large systems of spiking neurons, with structures inspired by the increasingly detailed findings of neuroscience, in order to investigate the emergent behaviours, adaptability and fault-tolerance of those systems.\n\nWhat has changed, and why could we not do this ten years ago? Simply put, it is now possible to build ensembles of a million cores in a University research environment, something that was impossible a decade ago.\n\nBiological inspiration draws us to two parallel, synergistic directions of enquiry:\n\n• How can massively parallel computing resources accelerate our understanding of brain function?\n• How can our growing understanding of brain function point the way to more efficient parallel, fault-tolerant computation?\n\nWe start from the following question: what will happen when processors become so cheap that there is, in effect, an unlimited supply of them? The goal is now to get the job done as quickly and/or energy-efficiently as possible, and as many processors can be brought into play as is useful; this may well result in a significant number of processors doing identical calculations, or indeed nothing at all - they are a free resource.\n\nOne significant outcome of the research to date is the recognition that a machine with the architecture outlined above can do much, much more than we had originally anticipated. Any problem that can be represented as a set, or grid, of interacting elements - be they biological neurons, or abstract mathematical entities - can benefit enormously from the computational power that a million processors can bring to bear. Complex system models, molecular dynamics; application domains traditionally the territory of (extremely expensive) high-performance computer systems are proving to be biddable to the capabilities of this machine. We call this new domain atomic computing, indicative of the small scale of the individual processing elements that combine to produce such powerful effects.\n\n SUMMARY:Computing beyond a million processors UID:2075 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20130409T103000 DTEND;TZID=America/Chicago:20130409T113000 DTSTAMP:20130525T020110 LOCATION:Building 240 Conference Room 4301, Argonne National Laboratory DESCRIPTION:The human brain remains as one of the great frontiers of science – how does this organ upon which we all depend so critically, actually do its job? A great deal is known about the underlying technology – the neuron – and we can observe large-scale brain activity through techniques such as magnetic resonance imaging, but this knowledge barely starts to tell us how the brain works. Something is happening at the intermediate levels of processing that we have yet to begin to understand, but the essence of the brain\'s information processing function probably lies in these intermediate levels. To get at these middle layers requires that we build models of very large systems of spiking neurons, with structures inspired by the increasingly detailed findings of neuroscience, in order to investigate the emergent behaviours, adaptability and fault-tolerance of those systems.\n\nWhat has changed, and why could we not do this ten years ago? Simply put, it is now possible to build ensembles of a million cores in a University research environment, something that was impossible a decade ago.\n\nBiological inspiration draws us to two parallel, synergistic directions of enquiry:\n\n• How can massively parallel computing resources accelerate our understanding of brain function?\n• How can our growing understanding of brain function point the way to more efficient parallel, fault-tolerant computation?\n\nWe start from the following question: what will happen when processors become so cheap that there is, in effect, an unlimited supply of them? The goal is now to get the job done as quickly and/or energy-efficiently as possible, and as many processors can be brought into play as is useful; this may well result in a significant number of processors doing identical calculations, or indeed nothing at all - they are a free resource.\n\nOne significant outcome of the research to date is the recognition that a machine with the architecture outlined above can do much, much more than we had originally anticipated. Any problem that can be represented as a set, or grid, of interacting elements - be they biological neurons, or abstract mathematical entities - can benefit enormously from the computational power that a million processors can bring to bear. Complex system models, molecular dynamics; application domains traditionally the territory of (extremely expensive) high-performance computer systems are proving to be biddable to the capabilities of this machine. We call this new domain atomic computing, indicative of the small scale of the individual processing elements that combine to produce such powerful effects.\n\n SUMMARY:Computing beyond a million processors UID:2077 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20130312T103000 DTEND;TZID=America/Chicago:20130312T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, Seminar Room 4301, Argonne National Laboratory DESCRIPTION:\nIn high energy Li-ion batteries, Li deposits formed on the anode surface during cell charging can cause poor cycling and sever safety problem. The formation of Li deposits leads to a large decrease of reversible capacity and worst a short-circuiting phenomenon as deposits grow towards to the cathode. Even for the graphite anode instead of lithium metal there is still a great chance for lithium to be deposited when Li-ion batteries undergo a high rate charge. During lithium electrodeposition processes, it is believed that the morphology and growth of electrodeposits are mainly determ ined by the kinetics of the heterogeneous electrode reaction, electrode surface states, Ohmic potential drop and mass species transports inside the bulk materials.\nIn this talk, I will present a nonlinear phase field model for predicting electrode-electrolyte interface motion and microstructure evolution during the electrochemical deposition involving highly nonequilibrium processes. Without considering the solid-electrolyte interface (SEI) layer effect, this model is able to simulate and predict the lithium deposits formation and growth in Li-ion batteries during charging operations. The electrodeposition rate implicitly follows the Butler-Volmer kinetic with a diffuse electrode-electrolyte interface description. The consuming of Li+ concentration and electric potential drop across the interface is correlated to the electrochemical reaction kinetic. The local variations of ion concentration and overpotential cause the instability of electrode surface which eventually produce fiber-like lithium deposits growth. The effects of charged current density and reaction rate constant on the deposits morphology are discussed in details. SUMMARY:Phase Field Modeling of Lithium Electrodeposition in Lithium-Ion Batteries UID:2079 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20130405T103000 DTEND;TZID=America/Chicago:20130405T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, Conference Room 4301, Argonne National Laboratory DESCRIPTION:High-performance GPU architectures have been widely used for scientific computation due to their high computational throughput, large memory bandwidth and outstanding performance-per-watt figures. VOCL(Virtual OpenCL) is a framework, based on OpenCL programming model, which supports the transparent utilization of local and remote GPUs, and it also supports virtual GPU migration. This talk will present new progress in VOCL framework. The new design of VOCL is able to abstract and manage local and remote GPUs efficiently in GPU cluster systems.\n\nBio: \nYan Li is a Ph.D. student at the Institute of Software, Chinese Academy of Sciences. His research interests mainly focus on the development of tools and libraries for applications on GPU-based heterogeneous computing platforms, such as fast Fourier transformation, numerical computation and data-parallel algorithms. He has been working in the programming models and runtime systems group at Argonne for about one year focusing on the VOCL project. SUMMARY:Abstraction and Management of Resources in Virtual GPU Environments UID:2101 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20130411T103000 DTEND;TZID=America/Chicago:20130411T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, Seminar Room 4301, Argonne National Laboratory DESCRIPTION:In 1990, Tardella showed that a sufficient condition for a function to attain its minimum over a polyhedron is that the function is concave along the edges of the polyhedron. We show how this result can be used to formulate two important discrete optimization problems as continuous quadratic programs. These programs possess the beautiful property that local optimality of a feasible point can be checked in polynomial time. Algorithms for solving these these programs will also be discussed. SUMMARY:Continuous Edge-Concave Quadratic Programming Formulations of Discrete Optimization Problems UID:2103 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20130313T103000 DTEND;TZID=America/Chicago:20130313T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, Seminar Room 4301, Argonne National Laboratory DESCRIPTION:For every class of polycrystalline materials, the scientific study of grain boundaries as well as the increasingly widespread practice of grain boundary engineering relies heavily on visual representation for the analysis of boundary statistics and their connectivity. Traditional methods of grain boundary representation neglect the full complexity of misorientation information and often rely on boundary classification schemes of dubious physical significance. This talk will highlight my recent work toward significantly advancing our ability to represent grain boundary information. Beginning with an understanding of the topology of the group space of misorientations, we developed new methods of mapping and visualizing boundary character. This new technique allows micrographs or maps of grain boundaries to be presented in a colorized form which, at a glance, reveal all of the misorientation information in an entire grain boundary network, as well as the connectivity among different boundary misorientations. Finally, I will introduce the extension of recently developed hyperspherical harmonic formulation for the description of grain boundary misorientation and boundary-plane statistics. These new and improved methods of representing grain boundary information, which are already being incorporated in commercial software packages, are expected to be powerful tools for grain boundary network analysis as the practice of grain boundary engineering becomes a routine component of the materials design paradigm.\n SUMMARY:Theory as a Tool for Microstructural Design: Grain Boundary Engineering UID:2083 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20130107T103000 DTEND;TZID=America/Chicago:20130107T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, Seminar Room 4301, Argonne National Laboratory DESCRIPTION:\nGraph algorithms have been playing a crucial role for solving problems in several domains including scientific computing, data mining, social networks, computational biology, and image decomposition. To deal with the emerging big data, it has been of significant importance to develop scalable parallel graph algorithms capable of utilizing the parallelism of todays multicore, many-core, and heterogeneous machines. In this talk, I will present the Union-Find (UF) algorithm, which is popularly known as one of the best ways to solve the minimum spanning tree (MST) problem. We performed a comprehensive study to compare all of the different versions of the UF algorithm and proposed the best way of implementing the UF algorithm in general and demonstrated that a somewhat forgotten simple algorithm is the fastest, in spite of the fact that its worst-case time complexity is inferior to that of the commonly accepted “best” algorithms. To solve the MST problem using big data, I will then briefly present scalable frameworks for the UF algorithm both on shared memory and distributed memory computers with reasonable speedup.\n\nLater I will talk on a well-known density based clustering algorithm, named DBSCAN, capable of discovering arbitrary shaped clusters and eliminating noise data. DBSCAN has been popular in image processing for many scientific domains, such as astronomy, biology, and earth science. I will show how a graph algorithmic technique (the parallel UF algorithm) can be used to break the access sequentiality of DBSCAN. Using data sets from millennium simulation runs consisting of billions of stars and galaxies, I show that parallel DBSCAN significantly out-performs previous algorithms, achieving speedups up to 25.97 using 40 cores on shared memory architecture, and speedups up to 5,765 using 8,192 cores on distributed memory architecture. SUMMARY: Scalable Parallel Graph Algorithms UID:2085 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20130218T103000 DTEND;TZID=America/Chicago:20130218T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, Conference Center, Room 1406 and 140, Argonne National Laboratory DESCRIPTION:Scientific simulations and experiments are producing increasingly large so-called multifield datasets that are comprised of scalar, vector, and tensor fields tracked over time. The visualization literature provides a variety of surface-based feature characterizations. While these models have proven extremely valuable in the analysis of individual fields, their use in practice is hindered by two major challenges.\n\nFirst, the extraction of these structures is often non-trivial as their precise computation requires careful analysis of the spatial variation of convoluted functions and their nonlinear invariants. Examples of such characterizations include so-called crease surfaces and level sets in nonlinear fields. In addition, these surfaces can be non-orientable, with a nontrivial topology, and a complex set of boundaries. To address these issues we have introduced a novel scalable front propagation strategy that offers strong guarantees on the approximation quality. Our solution is suitable for arbitrarily complex surfaces, and it ensures an excellent geometric quality for the produced mesh.\n\nA second, and arguably more severe limitation of these structure definitions is that they do not support the visual analysis of multifield datasets. In that context, we have proposed a geometrically motivated, multifield feature definition that characterizes the relationship between the features exhibited by the individual fields in the spatial domain of the problem. Our algorithm leverages the existing theory of skeleton derivation to simplify and fuse the surface structures from the constitutive fields into a coherent and visually effective data description. This work also introduced a new method for non-rigid surface registration tailored to the specific nature of surfaces extracted from computational fluid dynamics datasets. Temporal matching and spatial clustering enable the discovery of subtle interaction patterns between the different fields and their evolution over time.\n\nFinally, we document the unified visual analysis achieved by our methods in the context of several multifield problems from large-scale time-varying simulations, and we discuss future research directions. SUMMARY:Fast Extraction and Analysis of Surface-based Structures UID:2087 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20130123T103000 DTEND;TZID=America/Chicago:20130123T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, Seminar Room 4301, Argonne National Laboratory DESCRIPTION:Electronic structure theory concerns the description of molecular properties according to the postulates of quantum mechanics. For practical purposes, this is realized entirely through numerical computation, the scope of which is constrained by computational costs that increases rapidly with the size of the system.\n\nThe significant progress made in this field over the past decades have been facilitated in part by the willingness of chemists to forego some mathematical rigour in exchange for greater efficiency. While such compromises allow large systems to be computed feasibly, there are lingering concerns over the impact that these compromises have on the quality of the results that are produced. This research is motivated by two key issues that contribute to this loss of quality, namely i) the numerical errors accumulated due to the use of nite precision arithmetic and the application of numerical approximations, and ii) the reliance on iterative methods that are not guaranteed to converge to the correct solution. SUMMARY:Rigorous Numerical Approaches in Electronic Structure Theory UID:2089 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20130320T103000 DTEND;TZID=America/Chicago:20130320T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, Seminar Room 4301, Argonne National Laboratory DESCRIPTION:Inverse problems are ubiquitous in science and engineering, especially in the field of Geoscience. Algorithms that we have developed are applicable to seismic imaging, contaminant source identification, hydraulic tomography, etc. This work tackles these inverse problems using the Geostatistical approach that stochastically models unknowns as random fields. However, due to high computational costs in identifying small scale features, these methods are challenging. In addition, it is necessary to quantify the corresponding predictive uncertainty to provide a sound basis for management or pol icy decision making. My approach uses Hierarchical matrices to efficiently represent dense covariance matrices and solves the resulting intermediary system of equations using preconditioned iterative methods. The resulting cost is reduced from O(N^2) to O(N log N), where N is the number of unknowns to be determined. In addition, I will describe some recent work on Krylov solvers for shifted systems that accelerates the solution of \"forward problems\" and as a result, the inverse problem. Moreover, a technique to perform uncertainty quantification by estimating the diagonals of the posterior covariance matrix will be presented. I will illustrate these algorithms with examples from contaminant source identification and hydraulic tomography.\n SUMMARY:Fast Algorithms for Inverse Problems and Uncertainty Quantification UID:2091 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20130321T103000 DTEND;TZID=America/Chicago:20130321T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, Seminar Room 4301, Argonne National Laboratory DESCRIPTION:\nIn this talk we outline the challenges facing typical linear inverse problems from physical sciences: limited amount of data compared to the number of unknowns in the linear system, noise in the right hand side data, poor conditioning of the matrices and shortage of computational resources for the large data sets that are typically encountered. We discuss algorithms and practical strategies to address these problems and illustrate with examples from an inverse problem in geotomography. We start with classical $\\ell_2$ based minimization and go on to discuss different class es of algorithms (convex and non-convex) for sparsity constrained minimization motivated by theoretical results in compressive sensing. We then discuss approaches for mixed penalty regularization and the use of wavelet based constraints.\n\nFinally, we turn our attention to computation and discuss practical strategies for implementing the algorithms discussed with very large matrices using techniques such as wavelet based compression and randomized low rank singular value decompositions. We conclude by discussing additional applications and how the mentioned techniques can take advantage of today\'s multiprocessor/core architectures. SUMMARY:Linear Inverse Problems: Regularization Strategies and Their Practical Implementation UID:2093 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20130208T103000 DTEND;TZID=America/Chicago:20130208T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, Conference Center, Room 1406, Argonne National Laboratory DESCRIPTION:There is an ever-increasing need for exploring large-scale graph data sets in computational sciences, social networks, and business analytics. One of the most widely used graph searching algorithms is breadth-first search (BFS), which serves as a building block for a great many graph algorithms such as minimum spanning tree, betweenness centrality, and shortest paths. However, BFS is notoriously known challenging to optimize because of its poor locality and low computation to communication ratio. In this talk I will present several of our findings to improve the efficiency of BFS algori thms: exploiting different levels of parallelism to provide latency hiding and maximize memory bandwidth; reducing communication in distributed BFS to improve its scalability. SUMMARY:Large-Scale Graph Traversal Algorithms on Multi-core Clusters UID:2095 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20130128T103000 DTEND;TZID=America/Chicago:20130128T113000 DTSTAMP:20130525T020110 LOCATION:Building 240, Conf. Center, Rooms 1404 and 1405, Argonne National Laboratory DESCRIPTION:Cloud service providers charge customers based upon the amount of resources used or reserved. However, there are no guarantees on the quality-of-service (QoS) that the given resources will provide. Due to the highly dynamic nature of Internet workloads, increasing complexity of cloud-hosted applications, multi-tier architectures and the complex dynamics of underlying shared infrastructure, it is significantly challenging to manage application level performance in the Cloud. The situation is further complicated by the fact that cloud service providers need to control the power consumption in their data centers to avoid power capacity overload of high density servers, lower electricity costs and reduce their carbon footprint.\n\nI have developed novel middleware approaches to autonomic management of virtualized resources for power and performance control in the Cloud. Firstly, I designed self-adaptive and efficient resource provisioning techniques based on machine learning and control theoretical techniques to guarantee a high percentile performance of multi-tier web applications in the face of highly dynamic workloads. I also developed an automation tool for resource allocation and configuration of Hadoop framework for cost-efficient big data processing in the Cloud. Secondly, I designed a system that simultaneously controls the power consumption and the performance of multi-tier applications in a virtualized server system. Furthermore, I developed a power-aware framework for managing scientific workloads in virtualized GPU computing environments, with the help of emerging technologies that provide GPU accelerators as virtualized computing resources. Lastly, I proposed and developed non-invasive, e nergy efficient and highly scalable mechanisms to achieve performance isolation of heterogeneous applications in a multi-tenant cloud system. SUMMARY: Probabilistic Fault Detection and Diagnosis in Large-Scale HPC Applications UID:2097 SEQUENCE:0 END:VEVENT BEGIN:VEVENT DTSTART;TZID=America/Chicago:20130318T103000