Abstract: With the increasing energy consumption in current data centers, many emerging energy-saving mechanisms have been proposed to reduce the energy consumption, but most of these methods assume data center is in a homogeneous environment. However, most of current data centers are heterogeneous as different types of servers are purchased at different time in reality. An energy-efficient method named Primary Virtual Machine Allocation Policy (PVMAP) was proposed, with the performance/power introduced as a parameter to indicate the energy efficiency of each server. The server of high energy efficiency would be fully utilized with high priority in the dynamic Virtual Machine (VM) consolidation. Also the consolidation process would try to minimize the VM migrations and running hosts in the end. The simulation results demonstrate that the PVMAP can guarantee the energy conservation and Quality of Service (QoS) at the same time, and it has better stability and extensibility.

Key words: data center, virtual machine scheduling, performance/power, CloudSim

摘要： 随着数据中心的快速发展，其能耗问题已经愈发突出，数据中心节能机制已成为研究热点；但大多节能机制并未充分考虑数据中心的异构性，如不同时间购置的服务器之间存在差异。为此引入代表服务器能耗效率的能效比(Performance/Power)作为参数，提出一种基于虚拟机调度的节能算法PVMAP，动态整合虚拟机时优先充分使用能效比高的服务器，从而尽量减少虚拟机迁移次数和同时运行的服务器数量。仿真实验结果表明，算法能够在节能的同时保证服务质量(QoS)，比其他算法具有更好的稳定性和可扩展性。

关键词: 数据中心, 虚拟机调度, 能效比, CloudSim