IaaS云融合Q学习的自适应虚拟机部署  

Self-adaptive virtual machine deployment integrating Q-learning in IaaS clouds

作　　者：宋建松[1] 连玮[1] 陕粉丽[1] 闫慧鹏 SONG Jian-song;LIAN Wei;SHAN Fen-li;YAN Hui-peng(Department of Computer Science,Changzhi University,Changzhi 046011,China;Changzhi Branch,China Telecommunications Group Limited,Changzhi 046011,China)

机构地区：[1]长治学院计算机系,山西长治046011 [2]中国电信集团有限公司长治分公司,山西长治046011

出　　处：《计算机工程与设计》2021年第7期1851-1858,共8页Computer Engineering and Design

基　　金：国家自然科学基金项目(61773002)。

摘　　要：云环境中,仅根据当前负载需求降低活跃主机量,忽略负载变化时的未来资源需求,会导致过多非必要虚拟机迁移,增加SLA违例。为此,提出基于Q学习的自适应虚拟机部署算法。Q学习在无需先验知识前提下,可以自适应生成资源利用率阈值,根据自适应阈值动态地对主机超载状态做出决策,判断是否进行虚拟机迁移。通过现实负载流进行实验分析,实验结果表明,该算法可以降低主机能耗,同步减小虚拟机迁移量和SLA违例率。In cloud environment,the existing researches have focused on the number of active physical machines reduction accor-ding to their current workload requirements and neglected the future resource demands,which can lead to excessive unnecessary virtual machine migration and increase SLA violations.For this problem,a self-adaptive virtual machines deployment algorithm integrating Q-learning was proposed.On the premise of no prior knowledge,resource utilization threshold was self-adaptively generated by Q-learning,and a decision on the overload state of the host was dynamically made based on the adaptive threshold,to judge whether to carry out virtual machine migration.The algorithm was analyzed experimentally with real workload stream.Results show that the proposed algorithm can not only reduce the host energy consumption,but simultaneously reduce virtual machine migration and SLA violation rate.

关 键 词：虚拟机合并 虚拟机迁移 Q学习 服务等级协议 资源分配 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]