摘要：Data centers hosting applications consume huge amounts of energy. Flexible management of virtual clusters is an effective way to reduce the overall power consumption of data centers. In order to save energy we focus on decreasing the total power-on time of physical nodes in data centers and increasing the runtime efficiency of physical nodes. For a specific physical node how long it is power-on is determined by the job with longest completion time. In this work, we formulate the problem into maximizing the amount of data which is processed by unit energy. AVC-based migration approach is proposed, which exploits the completion time of parallel jobs encapsulated by virtual clusters to guide virtual machines migrations. Algorithm VCGM is designed to construct the final mapping state which is energy-efficiency. In VCGM, all virtual clusters are grouped by their remaining time and every virtual cluster group is "compacted" into few physical nodes one by one. Then we adopt the low-cost perfect matching in the bipartite matching problem to obtain a migration plan. Algorithms are simulated and evaluated to verify the effectiveness of our energy efficiency scheme.
关键词：Virtual cluster migration ; resource consolidation ; green computing
