Workshop Program

High-End Computing Trends: From Speed, Scalability to Efficiency [slides]

Prof. Zhiwei Xu, Institute of Computing Technology (ICT) of the Chinese Academy of Sciences (CAS)

Modern high-end computing systems started in 1976 with the introduction of Cray-1. When one looks carefully at the 40-year history, two major phases of trends may be observed in high-end computing systems architecture development, each lasting roughly 20 years. The first is called the performance-first phase, lasting roughly from 1976 (Cray-1) to 1994. The number one priority in this phase is performance, or flops speed. The second is called the scalability-first phase, lasting roughly from 1994 (IBM SP-2) to present. The number one priority in this phase is scalability, including market scalability and systems scalability. Now we are entering a third phase: the efficiency-first phase. In the next 20 years, high-end computing systems architecture research needs to increase energy efficiency by 1000-10000 times, in addition to continuing performance and scalability advances. The research community should set a bold goal of achieving energy efficiency of 1-10 Tera Operations Per Joule (1-10 TOPJ), or 1-10 TOPS/W, by year 2035. This talk outlines four lines of work related to this goal: Makimoto's Wave, Elastic Processor, libraries, and application frameworks.

Zhiwei Xu is a professor and CTO of the Institute of Computing Technology (ICT) of the Chinese Academy of Sciences (CAS). His prior industrial experience included chief engineer of Dawning Corp. (now Sugon as listed in Shanghai Stock Exchange), a leading high-performance computer vendor in China. He currently leads “Cloud-Sea Computing Systems”, a strategic priority research project of the Chinese Academy of Sciences that aims at developing billion-thread computers with elastic processors. Zhiwei Xu holds a Ph.D. degree from the University of Southern California, an MS degree from Purdue University, and a BS degree from University of Electronic Science and Technology of China.

• "A Bandwidth-saving Optimization for MPI Broadcast Collective Operation", Huan Zhou, Vladimir Marjanovic, Christoph Niethammer and Jose Gracia. [slides]
• "HetroCV: Auto-tuning Framework and Runtime for Image Processing and Computer Vision Applications on Heterogeneous Platform", Daihou Wang, David J. Foran, Xin Qi and Manish Parashar.[slides]
• "Collective Computing for Scientific Big Data Analysis", Jialin Liu, Yong Chen and Surendra Byna. [slides]

• "High Performance Computing of Fast Independent Component Analysis for Hyperspectral Image Dimensionality Reduction on MIC-based Clusters", Minquan Fang, Yi Yu, Weimin Zhang, Heng Wu, Mingzhu Deng and Jianbin Fang.[slides]
• "Mimer and Schedeval: Comparison Tools for Static Schedulers and Streaming Applications on Concrete Manycore Architectures", Nicolas Melot, Johan Janzén and Christoph Kessler. [slides]
• "OpenJDK Meets Xeon Phi: A Comprehensive Study of Java HPC on Intel Many-core Architecture", Yang Yu, Tianyang Lei, Haibo Chen and Binyu Zang. [slides]
• "Communication Avoiding Power Scaling", John Leidel and Yong Chen. [slides]

• "Utility-Based Scheduling for Bulk Data Transfers between Distributed Computing Facilities", Xin Wang, Wei Tang, Rajkumar Kettimuttu and Zhiling Lan. [slides]
• "Alignment-Free Sequence Comparison over Hadoop for Computational Biology", Giuseppe Cattaneo, Gianluca Roscigno, Umberto Ferraro Petrillo, Raffaele Giancarlo. [slides]