Research On Live Migration Of Virtual Machines Based On KVM

In the data center and cloud computing environments where make use of virtualization,particularly server virtualization, live migration has been a very important mechanism. Live Migration enables the movement of virtual machines between different physical machines, and in the process of moving, the virtual machine remains running,that is, without interrupting the operation of the clients or services. This feature brings great flexibility. With the help of live migration, managers can easily achieve load-balancing, online infrastructure maintenance and upgrades,energy management,etc. The pre-copy strategy, as the main mechanism for the live migration, makes good tradeoffs among migration time and downtime, cannot work well or even fails when dirty memory pages are largely generated because of frequent update operations.In this paper, we propose a novel pre-copy strategy to reduce the total migration time and the total transferred data, while making the downtime acceptable. The key idea is to compute the frequency of each dirty page based on historical and current statistics data. A dirty page will be transferred only when its transfer weight is not greater than a predefined threshold, otherwise its transfer will be delayed. And we propose the definition of Transfer Number Limitation and Transfer Interval Limitation to reduce the duplication of data transmission.The experiment results show that the optimized strategy can significantly reduce the total transferred data, the total migration time and the number of iterations when memory pages update frequently. And the downtime is acceptable. The results show that the new strategy can reduce the total data transferred by about 58%?72% and the total migration time by about 58%?73% in case of write-intensive scenes compared with the original method.