2015年5月25日,Intel Jian Chen博士应邀来中国科学院深圳先进技术研究院云计算研究中心进行学术交流,作了题为“Scalable Performance Modeling for Large-scale MPI workloads”的学术讲座,报告由喻之斌研究员主持。


Abstract

Efficient and scalable performance modeling is essential to high-performance cluster computing. Existing methods heavily rely on traces for performance analysis, and are usually expensive, inefficient, and inscalable. In this talk, I present an innovative and scalable performance modeling framework, which includes a novel concept of critical-path candidates and a high-quality synthetic trace generator. The critical-path candidates refer to a group of paths that could potentially be the critical path. Using the instruction and communication counts as the metrics, the critical-path candidate captures the intrinsic computation and communication dependencies, and hence can be reused for exploring multiple design options. On the other hand, the trace generator leverages machine learning techniques to automatically recognize the scaling patterns from training traces, and generate the synthetic traces which mimic original trace behavior and extrapolate it to arbitrary scale. Together, they provides an end-to-end solution for efficient and scalable performance modeling of large-scale MPI workloads. 


Bio

Jian Chen joined Intel in 2011, and is currently a senior performance architect in Data Center Group. He earned his Ph.D. degree in computer engineering from the University of Texas at Austin in 2011, and received his M.E. and B.E. in electrical engineering from Shanghai Jiao Tong University in 2005 and 2002 respectively. His research interests include computer architecture, large-scale cluster performance modeling, workload characterization, and BigData applications. He has published more than 16 refereed papers. He is the recipient of ISPASS Best Paper Award 2015.