Long-Term Analysis for Job Characteristics on the Supercomputer

Abstract

A deep understanding of the job characteristics and their impacts on the high performance computing system is one of the most critical steps for efficiently planning its design, development and optimization. However, frequent and regular characterization studies are insufficient in many HPC systems, which might make the study done by the system researchers inconsistent with the actual system features and application characteristics, and ultimately lead to the failure of the proposed strategy. Our study in this paper tries to bridge the gap by performing long-term analysis for job characteristics on a petascale ARM supercomputer, in this way, we get many meaningful findings and insights, which we believe can benefit the co-design of hardware and applications, and improve performance and experience of the job submitters in the HPC system.