Study of resource utilization of applications in petascale system
Hao Jin · Illinois Digital Environment for Access to Learning and Scholarship (University of Illinois at Urbana-Champaign) · 2016
To improve the design of large-scale high performance computing systems for improving overall system efficiency, it is first important to understand the resource utilization in such systems. This project aims at understanding the resource utilization of Blue Waters, a 13.1 peta-FLOP system, by applications through mining and analysis of performance and job scheduling logs. This project studied utilization of CPU, memory and filesystem through detailed analysis on five applications with different needs for hardware that used a large number of nodes and ran for long time over the past few years. GPU utilization on XK nodes were not included due to the lack of data in the logs. Contribution of this study enables us to (1) find out applications with low resource utilization, (2) measure actual utilization of system, and (3) envision the design for next generation hybrid systems.