System energy analysis for shared memory multiprocessing applications

Abstract

This paper presents a detailed energy consumption analysis, considering the energy consumption related to CPU, cache memory and main memory of parallel applications on a 16-core HPC platform. The correlations between energy consumption, speedup, and execution time are also herein presented. Tests are conducted with the NAS parallel benchmarks using three different measurement tools: i) Perf, for the measurement of hardware cache memory events; ii) CACTI, used to estimate the cache memory energy consumption by access; and iii) PCM, for CPU and DRAM energy consumption estimates. Our results show that the lowest overall energy consumption occurs only when all physical cores are used, reducing by 62%, on average, the total system energy consumption when compared to the sequential version for the execution. Moreover, the cache memories results are even better, achieving a reduction of 80% in most of the cases, despite the increase in cache miss rate generated by the increased number of threads.