Factor Analysis of Applications Performance Data for the Cray Y-MP
G. Delic, Richard I. Haller · The International Journal of Supercomputer Applications and High Performance Computing · 1996
Factor analysis was applied to supercomputer applica tions performance data recorded by the hpm hardware performance monitor on the Cray Y-MP for a sample of 146 user codes. The analysis reduced the volumi nous hpm data to five dominant factors which were used as performance metrics for applications code ex ecution on the Y-MP. Case studies of Fortran kernels demonstrated that the proposed metrics accurately reflect changes in performance due to source code modifications. Therefore, the factor-analysis model so lution provides a supercomputer performance metric that is both parsimonious and simple to interpret as an alternative to the complexities of hpm-derived perfor mance attributes. Metrics such as these should prove to be of value in code optimization, user training, and characterization of site-specific workloads.