Recent Successes and Changes of the HPCMP Sustained Systems Performance Test

The sustained systems performance (SSP) test has been implemented on certain High Performance Computing Modernization Program (HPCMP) HPC systems in order to quantitatively evaluate updates to system software, hardware repairs, job queuing policy modifications, and revisions to the job scheduler as necessary. The test employs codes used in the system acquisition cycle with proven migration capability to HPCMP HPC systems and non-empirical tests for numerical accuracy. Metrics such as compilation time, queue wait time, benchmark execution time, and total test throughput time are gathered and compared against metric data from previous tests to monitor the systems under test while minimizing impact to the users. Jobs failing to execute properly or in anomalously short or long times are investigated, and the results are reported to system administrators and center directors at each center for appropriate actions. During the past year, the SSP test has been instrumental in surfacing configuration issues with the PBS scheduler and performance issues on several HPC systems. Additionally, the frequency of the SSP test on systems procured in Technology Insertion 2009 (TI-09) and thereafter has increased, with attendant changes in the test cases comprising the test. The SSP test continues to play an important role in monitoring the quality of service delivering HPC to HPCMP users at the system, DoD Supercomputing Resource Center, and vendor levels.