Energy Saving Strategy Based on Profiling

Abstract

Constraints imposed by power consumption and the related costs are one of the key roadblocks to the design and development of next generation exascale systems. To mitigate these issues, strategies that reduce the power consumption of the processor are the need of the hour. Techniques such as Dynamic Voltage and Frequency Scaling (DVFS) exist which reduce the power consumption of a processor at runtime but they should be used in such a manner so that their overhead does not hamper application performance. In this paper, we propose an energy saving strategy which operates on timeslice basis to apply DVFS under a user defined performance constraint. Results show energy savings up to 7% when NAS benchmarks are tested on a laptop platform