Personalized Benchmarks for High Performance Computing Applications

Computer Science

Participants: Philip Carns, Robert Ross, Marianne Winslett

The goal of this project is to develop methods to provide personalized supercomputing I/O benchmarks and analytics that represent the entire range of production workloads. To do this we will instrument a broad sampling of jobs at production HPC facilities, automate analysis of that instrumentation, and create customized benchmarks that reflect the aggregate workload of the system.

Heat map showing the relationship between I/O volume and I/O throughput for production applications on a large-scale HPC system. This analysis suggests that relatively few applications make optimal use of the system's I/O capability.
Heat map showing the relationship between I/O volume and I/O throughput for production applications on a large-scale HPC system. This analysis suggests that relatively few applications make optimal use of the system’s I/O capability.
Personalized Benchmarks for High Performance Computing Applications