Template-based memory deduplication method for inter-data center live migration of virtual machines

Abstract

Live migration of virtual machines (VMs) can benefit data centers through load balancing, fault tolerance, energy saving, etc. Although live migration between geographically distributed data centers can enable optimized scheduling of resources in a large area, it remains expensive and difficult to implement. One of the main challenges is transferring the memory state over WAN. There is a conflict between the low data transmission speed over WAN and the rapid change of memory contents. This paper proposes a novel live migration method with page-count-based data deduplication, which takes advantage of the fact that VMs running same or similar operating systems and other software tend to have identical memory pages. Template pages are selected based on number of occurrences of each page across multiple VMs and indexed by content hash. When a memory page is transferred, the source host first compares it with the templates. If a match is identified, the source host transfers the index instead of the data of the memory page. The experimental results show that our approach reduces the migration time by 27% and the data transferred by 38% on average compared to the default method of QEMU-KVM.