A detailed analysis of OpenMP runtime configurations for power constrained systems

Abstract

Application level power budget allocation is one way to overcome the power constraint problem in future HPC systems. This technique mainly depends on finding an optimal number of compute nodes and power level for each node. However, utilizing that power at node level requires optimization of the underlying programming model. OpenMP is the defacto standard for intra-node parallelism. In this paper, we investigate the impact of OpenMP runtime environment on the performance of OpenMP code at the different power level. We studied 28 OpenMP parallel regions from five NAS Parallel Benchmark (NPB) applications. Based on the study we show that for a given power level, a suitable selection of OpenMP runtime parameters can improve the execution time and energy consumption of a parallel region up to 67% and 72%, respectively. We also show that these fine grain improvements resulted in upto 26% execution time and 38% energy consumption improvement for a given OpenMP application.