Measurement of a scientific workload using the IBM hardware performance monitor

  • Authors:
  • Robert J. Bergeron

  • Affiliations:
  • NASA Ames Research Center, Moffett Field, CA

  • Venue:
  • SC '98 Proceedings of the 1998 ACM/IEEE conference on Supercomputing
  • Year:
  • 1998

Quantified Score

Hi-index 0.00

Visualization

Abstract

This paper presents data on the performance of the NAS SP2 IBM system using RS6000 hardware monitors and a performance measurement tool. The data collected showed that the SP2 averages about 1.3 Gflops, about 3% of peak. The report provides the relative usage for the various hardware units over the entire workload measured over a 9-month period. The workload displays moderate parallelism, with the most popular choice of nodes as 16. Although the monitor data provide a good snapshot of workload performance, causal correlations regarding key performance indicators appear difficult to draw from the current data due to the absence of I/O delay measurements.