EAGER: Workload Analysis of Blue Waters

Blue Waters is the largest NSF-supported leadership-class supercomputer. It was acquired and is operated by the University of Illinois. Its purpose is to greatly advance fundamental understanding represented in "grand challenge" problems within and across a wide range of science and engineering. It does this by offering a computing system and environment suitable for and capable of solving the most challenging computational and data analysis problems. Given the important and unique role that Blue Waters plays in the nation's research portfolio, it is important to have a detailed technical understanding of its scientific workload. This award to State University of New York (SUNY) at Buffalo undertakes such a study. This workload characterization will guide performance optimization at the software and system configuration level to maximize job performance and workflow, as well as to help inform future computer architecture research and development. Additionally, the analysis could concretely inform the system balance trade-offs of future leadership-class system deployments. This study will leverage prior Blue Waters studies and data, but will incorporate a comprehensive approach by extending the NSF-funded and publicly available XDMoD (XD Metrics on Demand) service to include data specific to Blue Waters architecture. The results of this study will not only provide detailed operational and performance analytics for Blue Waters, but will also be used as a template for similar studies carried out on other advanced high performance computing systems that are revolutionizing computational science. This "transfer of knowledge" will be facilitated through the use of Open XDMoD for the proposed analysis, which is already in wide use by HPC centers worldwide. In addition, one of the outcomes of this project will be the modification of Open XDMoD to ingest job level performance data from the OVIS/LDMS monitoring framework. OVIS/LDMS is widely deployed on Cray HPC systems and this outcome will enable those centers with Cray systems to fully leverage Open XDMoD to provide comprehensive resource management.