Workshop: Workshop on Operational Data Analytics in HPC (HPC-ODA)
Authors: Strahinja Trecakov (New Mexico State University, New Mexico Consortium) and Mohammad Al-Tahat and Jonathan Cook (New Mexico State University)
Abstract: Most HPC clusters collect and save job accounting data that can offer insights into how the cluster is used. Many others have looked at this and other data, with particular results and insights. This study adds a novel approach, looking at job data through a four-way job classification, and then analyzing cluster utilization through this lens. Additionally, we infer some amount of application usage data, and infer parallel configurations being utilized by users.
Some important results, among others, of our analysis are: a large fraction of HPC usage is from single-node jobs; user provided job walltime limits are largely unrelated to actual job walltime; and parallelism configurations (inferred) seem to be a place for significant user training and improvement.
Back to Workshop on Operational Data Analytics in HPC (HPC-ODA) Archive Listing
Back to Full Workshop Archive Listing