Abstract:
Thorough investigation of supercomputer resource utilization efficiency is of great practical importance. Every HPC system holder and every user face such a problem. An approach to overall performance analysis, including peculiarities of application runs, assignment of jobs to queues, and total resource utilization of supercomputer systems is proposed. This approach is based on the analysis of system monitoring data and is aimed at providing a number of means for the qualitative behavior evaluation of supercomputer applications and HPC systems as a whole.
Keywords:supercomputer, performance, efficiency, parallel programs, dynamic characteristics, workload, system monitoring.