An experimental study of hybrid energy-aware scheduling in a cloud testbed

Abstract

Open-source cloud management often has a high degree of centralization and limited support of power management. This paper develops a hybrid approach that combines suspending idle server and load-balancing of virtual machines to minimize energy consumption in the cloud environment. We have conducted a series of experiments evaluating the performance of the hybrid approach under different synthetic CPU loads and memory usage. We also studied the effect of traditional task scheduling schemes in conjunction with our hybrid approach on energy consumption. Experimental results show that the hybrid technique can increase power saving by approximately 10-20% with various server configurations, workloads and task scheduling schemes, while meeting a variety of service requirements of different applications.