摘要
虚拟化的数据中心主要采用预拷贝(Pre-copy)算法进行虚拟机实时迁移.当虚拟机中运行负载较高或者网络传输带宽较低时,预拷贝算法固定的停机阈值严重影响实时迁移的性能.针对这个问题,提出一种自适应停机阈值机制,根据之前各拷贝轮中脏页率构成的时间序列,首先利用动态指数平滑法来预测后轮的脏页率,在预测脏页率超过网络传输带宽的情况下,再采用Mann-Kendall检验模型对脏页率变化趋势进行判断,根据判断结果确定停机切换的时机.实验结果证明,采用基于自适应阈值机制的预拷贝算法能在高负载低延迟场景下有效提高实时迁移性能.
Pre-copy is the dominant approach to carry out live migration of virtual machine. In the case of high workload in virtual machine and low network bandwidth, the fixed threshold setup for stop-copy in standard Pre-copy seriously degrades the performance of live migration. To address this problem, this paper proposes an adaptive threshold mechanism, which firstly predicts the dirty page rate of the next iteration by dynamic exponential smoothing when at the end of current iteration, then judges the changing trend of dirt- y page rate exploited with Mann-Kendall test model once the dirty page rate exceeds the network bandwidth, and finally figures out the time for stop-copy phrase according to the judgment result. Experiment results show that this optimized mechanism can effectively improve the performance of live migration under the condition of high workload and low network bandwidth.
出处
《小型微型计算机系统》
CSCD
北大核心
2015年第3期466-470,共5页
Journal of Chinese Computer Systems
基金
教育部博士点专项科研基金项目(20114101110007)资助
河南省科研重点项目(13A520562)资助
河南省创新人才项目(2011HASTIT003)资助
关键词
实时迁移
虚拟机
自适应阈值
预拷贝
live migration
virtual machine
adaptive threshold
pre-copy
