虚拟机实时迁移的研究

Research on Live Migration of Virtual Machine

虚拟机的实时迁移技术就是把虚拟机完整的从源物理主机迁移拷贝到另外一台物理主机上,在负载均衡和灾难恢复方面起到了重要作用。预拷贝算法实现了虚拟机的实时迁移,但在高负载场景下,一些内存页会被反复传送,严重影响了迁移效率,延长了总迁移时间。针对此问题,提出了基于脏页预测算法,在高负载场景下,采用动态指数平滑法对脏页面进行工作集预测,减少脏页面的传送,对于高频脏页,直接放入最后停机拷贝时传送,从而减少实时迁移的时间,提高迁移效率。实验结果表明,改进后的算法能在高负载场景下有效提高虚拟机实时迁移的性能。

Live migration technology of virtual machine is to finish the migration copy of virtual machine from the source physical host to another, which plays an important role in load balancing and disaster recovery. The live migration is realized by pre-copy,but in the case of high workload,some memory pages will be transmitted over and over again, which affects processing performance seriously and extends the total migration time. For this problem,a dirty page prediction algorithra is proposed which is used to predict the dirty page by dynamic exponential smoothing method,reducing the repeated transmission of dirty pages. For dirty page with high frequency,directly into the final copy down the transmission, the number of live migration time is reduced and the efficiency of migration is improved. Experiment results show that the improved algorithm can effectively rise the performance of virtual machine migration in high scenarios.