Improve Energy Efficiency by Processor Overclocking and Memory Frequency Scaling

High power consumption has become one of most important issues for supercomputer systems. Although the performance of supercomputer systems has been greatly improved in the recent years, a large amount of energy consumption causes not very high energy efficiency of the supercomputer systems. Computer nodes are one of major components of supercomputer systems. Therefore, how to improve the energy efficiency of a computer node is our concern. It is very important to do the tradeoff between performance and energy consumption of a computer node for better energy efficiency. We take energy delay product (EDP) as our metric for energy efficiency. The lower EDP means the better energy efficiency. For a shared-memory multicore processor system, we utilize processor overclocking and memory frequency scaling to achieve better performance and lower power consumption, that brings about the better energy efficiency for our benchmarks. But, we can not ensure to achieve better energy efficiency at any time we use processor overclocking, and we also find that scaling down memory frequency combined with processor overclocking cannot make sure to reach higher energy efficiency and even make the energy efficiency worse. In order to achieve higher energy efficiency for a shared-memory multicore processor system, we propose a processor overclocking and memory frequency scaling based Holistic Energy-Efficient (HEE) algorithm, which determines when we should use processor overclocking to improve energy efficiency and when we could combine processor overclocking with memory frequency scaling for a better energy efficiency. We build a real experimental platform to evaluate the effectiveness of our HEE algorithm. From the detailed experimental results, our HEE algorithm can achieve an average 13% energy efficiency improvement and the maximal energy efficiency increase is up to 25.6%.