云中心异构服务器能耗受限下的性能优化策略

摘要

Aimed to the problem of optimal performance of heterogeneous servers under power constraint of cloud center,this paper proposed an energy-performance optimal model based on M/GI /1-PS queueing system,and designed an OPT algorithm of allocating power among servers to achieve optimal performance.Meanwhile by comparing OPT algorithm with other common heuristic power allocation methods (WP,EU,EMRT),via plenty of numerical simulation experiments,this paper analyzed the impact of energy consumption optimization on performance under traffic varying,and obtained power-performance relation curve.The experiments results show that the OPT algorithm can obtain better performance than other algorithms under the same power constraint condition,especially has 22% higher performance than EU algorithm.The result of this paper will provide theory basis and reference data for optimal resource allocation of cloud center.%面向能耗受限下云中心异构服务器的性能优化问题，提出了基于 M／GI ／1-PS 排队系统的云中心性能优化分析模型，并给出获得最优性能服务器能耗分配 OPT 算法；同时通过大量数值仿真实验将 OPT 算法与常用的启发式能耗分配方法（WP、EU、EMRT）进行对比，分析了不同流量下能耗优化对性能的影响，并获得能耗—性能变化曲线。实验结果表明，OPT 算法在同等能耗受限条件下比其他算法获得更优的性能，尤其是 EU 算法，性能高出约22％。结果可为云中心优化资源配置提供理论依据和参考数据。