Workshop Program

An Ecosystem for the New HPC: Heterogeneous Parallel Computing

Prof. Wu-chun Feng, Virginia Tech

With processor core counts doubling every 24 months and penetrating all markets from high-end servers in supercomputers to desktops and laptops down to even mobile phones, we sit at the dawn of a world of ubiquitous parallelism, one where extracting performance via parallelism is paramount. That is, the "free lunch" to better performance, where programmers could rely on substantial increases in single-threaded performance to improve software, is over. The burden falls on developers to exploit parallel hardware for performance gains. But how do we lower the cost of extracting such parallelism, particularly in the face of the increasing heterogeneity of processing cores? To address this issue, this talk will present a vision for an ecosystem for delivering accessible and personalized supercomputing to the masses, one with a heterogeneity of (hardware) processing cores on a die or in a package, coupled with enabling software that tunes the parameters of the processing cores with respect to performance, power, and portability. Initial results of different aspects of the ecosystem will be demonstrated via a benchmark suite of computational dwarfs and applications on platforms ranging from the embedded or mobile space to the datacenter, such as Virginia Tech's very own GPU-accelerated supercomputer, HokieSpeed, which debuted on the Green500 as the most energy-efficient (i.e., greenest) commodity supercomputer in the U.S. and at a fraction of the cost of the fastest supercomputer in the world.

Wu Feng is the Elizabeth & James E. Turner Fellow and Professor of Computer Science, of Electrical & Computer Engineering, and of Health Sciences at Virginia Tech, where he directs the Synergy Laboratory and serves as the VT site co-director of the NSF Center for High-Performance Reconfigurable Computing (CHREC). He is a recognized leader in creating systems software (e.g., CoreTSAR, EnergyFit, EcoDaemon), middleware (e.g., MOON), tools (e.g., CU2CL, OpenDwarfs), and ecosystems (e.g., ParaMEDIC) to enable high-performance computing from the embedded space to the datacenter and cloud. To demonstrate the efficacy of his interdisciplinary research, he has applied the above to many application domains, including the life sciences (e.g., mpiBLAST and Compute the Cure for Cancer), seismology (e.g., WaveProp), and neuroscience (e.g., TDM).

Dr. Feng received a B.S. in Electrical & Computer Engineering and Music (Honors) and an M.S. in Computer Engineering from Penn State University in 1988 and 1990, respectively. He earned a Ph.D. in Computer Science from the University of Illinois at Urbana-Champaign in 1996. His previous professional stints include IBM T.J. Watson Research Center, NASA Ames Research Center, Vosaic, University of Illinois at Urbana-Champaign, Purdue University, The Ohio State University, Orion Multisystems, and Los Alamos National Laboratory. He is a Distinguished Member of the ACM, Senior Member of the IEEE, and two-time designee of HPCwire's Top People to Watch List.

• [09:15am - 09:45am] "Local Search to Improve Task Mapping", Evan Balzuweit, David P. Bunde, Vitus J. Leung, Austin Finley, and Alan C. S. Lee. [slides]
• [09:45am - 10:15am] "A Framework for Tracking Memory Accesses in Scientific Applications", Antonio J. Pena and Pavan Balaji. [slides]

[10:15am - 10:45am] Coffee Break

• [10:45am - 11:15am] "Checksumming Strategies for Data in Volatile Memories", Humayun Arafat, Sriram Krishnamoorthy and P. Sadayappan. [slides]

• [11:15am - 11:45am] "Pruning strategies in adaptive off-line tuning for optimized composition of components on heterogeneous systems", Lu Li, Usman Dastgeer and Christoph Kessler. [slides]
• [11:45am - 12:15pm] "Energy and Power Characterization of Parallel Programs Running on Intel Xeon Phi", Joal Wood, Ziliang Zong, Qijun Gu and Rong Ge. [slides]
• [12:15pm - 12:45pm] "Delivering Parallel Programmability to the Masses via the Intel MIC Ecosystem: A Case Study", Kaixi Hou, Hao Wang and Wu-Chun Feng. [slides]

• [1:45pm - 2:15pm] "Using Working Set Reorganization to Manage Storage Systems with Hard and Solid State Disks", Junjie Chen, Jialin Liu, Philip Roth and Yong Chen. [slides]
• [2:15pm - 2:45pm] "Benchmarking the Performance of Scientific Applications with Irregular I/O at the Extreme Scale", Stephen Herbein, Scott Klasky and Michela Taufer. [slides]
• [2:45pm - 3:15pm] "Improving Data Movement Performance for Sparse Data Patterns on Blue Gene/Q Supercomputer", Huy Bui, Eun-Sung Jung, Venkatram Vishwanath, Jason Leigh and Michael Papka. [slides]

[3:15pm - 3:45pm] Coffee Break

• [3:45pm - 4:15pm] "Decoupled I/O for Data-Intensive High Performance Computing", Chao Chen, Yong Chen, Kun Feng, Yanlong Yin, Hassan Eslami, Rajeev Thakur, Xian-He Sun, and William D. Gropp. [slides]

• [4:15pm - 4:45pm] "A Distributed Dataflow Model for Task-uncoordinated Parallel Program Execution", Lucas Wilson and Jeffery Von Ronne. [slides]
• [4:45pm - 5:15pm] "Beehive: A Framework for Graph Data Analytics on Cloud Computing Platforms", Anand Tripathi, Vinit Padhye and Tara Sasank Sunkara. [slides]
• [5:15pm - 5:45pm] "NestedMP: Taming Complex Configuration Space of Degree of Parallelism for Nested-Parallel Programs", Jiangzhou He, Wenguang Chen and Zhizhong Tang. [slides]

Copyright (C): Pavan Balaji, Argonne National Laboratory