摘要
针对虚拟化集群上日益严重的功耗问题,在定义集群、物理机、操作点、能耗、任务集、虚拟机等概念的基础上,提出了形式化的虚拟机(Virtual Machine,VM)调度策略,并有针对性地利用动态电压和频率调整(Dynamic Voltage and Frequency Scaling,DVFS)技术对普通算法进行了扩展改进,提出一个功耗敏感的VM调度算法。首先,利用FFD(First-Fit Decreasing)算法解决虚拟机的初始布局问题,然后尽可能地将虚拟机部署在低电压的物理机上,此外加入性能感知策略,尽量将物理机的电压调节至刚好满足虚拟机性能需求,以避免"奢侈"能耗的浪费,同时增加了虚拟机调度失败时灵活调整电压的规则,以减少错误先验知识对虚拟机后续部署的误导。在理想模型中的仿真实验表明:与现有部署算法相比,本算法具有更好的节能效应,以及相对不大的性能损失。
Aiming at the power problem of virtualized cluster, based on the defining of cluster,physical machine,opera- ting point,energy,jobs,virtual machine, the formal scheduling policy of Virtual Machines （VMs） was proposed to re- search the deployment scheme. And the common algorithm was extended and modified with Dynamic voltage and fre- quency scaling （DVFS） technology for energy-awareness. Using the FFD （First-Fit Decreasing） algorithm, the virtual machines were arranged in frequency sequence,then deployed on a physical machine with lowest voltage. Using perfor- mance apperceive policy, the voltage physical machines were set just enough for the total requirement of virtual ma- chines to reduce ＂luxury＂ energy respectively. And using the voltage update rules after no virtual machines deployed, the misdirection of the inaccurate heuristic information was avoided. The modified algorithm was researched with exper- iments in an ideal model. Finally the performance and energy of the modified algorithm were compared with the old. The experimental results show that the modified algorithm meets the need of the jobs,and saves a great deal of energy.
出处
《计算机科学》
CSCD
北大核心
2014年第8期38-41,46,共5页
Computer Science
基金
国家"863"高技术研究发展计划基金项目(2013AA01A213)资助
关键词
虚拟化
DVFS
功耗
调度
Virtualization, DVFS, Power, Scheduling
