Memory state transfer optimization for pre-copy based live VM migration

As an extremely powerful cloud platform management tool for load balancing, online maintenance, proactive fault tolerance, etc, the pre-copy based live migration of virtual machine has been widely used in multi commercial and open source hypervisor. This paper focuses on optimization of memory state transfer in the iterative copy stage of the pre-copy approach. Our approach uses historical dirty page information periodically collected from the starting of migration at a fixed time interval to make the hot dirty pages recognition more accurate. Further a weighted dirty page rate calculation method is proposed to determine whether a dirty page need be transferred to the target in the current iteration. Experiment results show the optimized approach is better than the existing pre-copy based approaches in terms of downtime, total migration time and the number of transferred pages.