SUPReMM Overview

An available Open XDMoD enhancement is SUPReMM (integrated HPC systems usage and performance of resources monitoring and modeling), which queries system hardware counters to collect a range of performance information, including memory usage, filesystem usage, interconnect fabric traffic, and CPU performance. Typically, this information is acquired at the job’s start in the prolog, at the job’s end in the epilog, and synchronously across all nodes periodically. The recommended collection period is every 30 seconds. SUPReMM provides a large variety of job performance metrics that give the HPC center directors and support personnel insight into the performance of all applications running on the cluster, without the need to recompile end user applications.