Abstract

Performance and power management in High Performance Computing (HPC) has historically favored a race-to-idle approach in order to complete applications as quickly as possible, but this is not energy-eicient on modern systems. As we move toward exascale and hardware over-provisioning, power management is becoming more critical than ever for HPC system administrators, opening the door for more balanced approaches to performance and power management. We propose two projects to address balancing application performance and system power consumption in HPC during application runtime, using closed loop feedback designs based on the Self-aware Computing Model to observe, decide, and act.

Date

January 1, 1970

Authors

Connor Imes, Steven Hofmeyr, Henry Hofmann

Journal

Proceedings of the International Conference for High Performance Computing, Networking, Storage and Analysis (SC’17), Denver, CO, USA

Pages

12-17