一种能耗—性能协调优化的虚拟机重放置策略

在云计算环境中虚拟机重放置方法方面,现有多数算法通常聚焦单一目标的优化,而聚焦一个单一目标通常会牺牲其他目标来达到最优效果,因此有必要考虑多目标权衡的虚拟机重放置方法。以降低能耗和保证虚拟机的服务质量为目标,提出一种能耗—性能协调的虚拟机重放置优化算法,即能耗—性能优化配合降序最佳适应算法（energy-performance awareness best fit descending virtual machine relocating,EPAR）,把资源使用率转换为能耗,同时权衡了能耗和性能之间的关系。该算法在选择重放置虚拟机时使用自回归模型预测下一时间段的性能,有效避免了不必要的迁移。通...

Full description

Saved in:
Bibliographic Details
Published in计算机应用研究 Vol. 33; no. 11; pp. 3324 - 3328
Main Author 李湘 陈宁江 黄汝维 贾炅昊 闫承鑫
Format Journal Article
LanguageChinese
Published 广西大学 计算机与电子信息学院,南宁,530004 2016
Subjects
性能感知
能耗
自回归模型
虚拟机重放置
虚拟机重放置
自回归模型
performance awareness
autoregressive model
性能感知
能耗
virtual machine relocation
energy
Online AccessGet full text

Cover

More Information
Summary:在云计算环境中虚拟机重放置方法方面,现有多数算法通常聚焦单一目标的优化,而聚焦一个单一目标通常会牺牲其他目标来达到最优效果,因此有必要考虑多目标权衡的虚拟机重放置方法。以降低能耗和保证虚拟机的服务质量为目标,提出一种能耗—性能协调的虚拟机重放置优化算法,即能耗—性能优化配合降序最佳适应算法（energy-performance awareness best fit descending virtual machine relocating,EPAR）,把资源使用率转换为能耗,同时权衡了能耗和性能之间的关系。该算法在选择重放置虚拟机时使用自回归模型预测下一时间段的性能,有效避免了不必要的迁移。通过原型验证,EPAR算法能够在确保虚拟机服务的情况下,有效降低宿主机的能耗,避免不必要的虚拟机的迁移。
Bibliography:51-1196/TP
energy; performance awareness; virtual machine relocation; autoregressive model
Li Xiang, Chen Ningjiang, Huang Ruwei, Jia Jionghao, Yan Chengxin （School of Computer ＆ Electron Information, Guangxi University, Nanning 530004, China）
Most of existing virtual machine relocating algorithms usually focused on the optimization of single goal in cloud computing environment, usually sacrificed other goal to achieve the optimal effect. Therefore this paper presented a virtual ma- chine relocating strategy with multi-goal tradeoff. To decrease energy consumption and guarantee the service quality of the vir- tual machines, the paper put forward a virtual machine relocating algorithm of collaborating between energy and performance, called EPAR（ energy-performance awareness best fit descending virtual machine relocating）, in which resource utilization was converted into energy consumption and there was a tradeoff between energy consumption and performance at the same time. The algorithm effectively avoided the un
ISSN:1001-3695
DOI:10.3969/j.issn.1001--3695.2016.11.028