Shrinker: Efficient Wide-Area Live Virtual Machine Migration using Distributed Content-Based Addressing

Live virtual machine migration is a powerful tool offered by virtualization technologies. However, current implementations restrict its usage to local area networks. Extensions to wide-area networks have been proposed with systems for transferring disk state and transparently migrating network connections. Still, wide-area live migration remains expensive to use because of the bandwidth bottleneck, especially when considering migrations of large clusters of virtual machines rather than single instances. Previous work has shown that VMs running identical or similar operating systems have a significant portion of their memory containing the same data. In this paper, we introduce Shrinker, a live virtual machine migration system leveraging this common data to improve migration efficiency between data centers interconnected by wide-area networks. Shrinker uses distributed content-based addressing to detect memory pages of the migrated VM that are already available on the destination site, removing the need to transfer these pages over a wide-area link. We have implemented Shrinker in the KVM hypervisor and present performance evaluation in a distributed environment. We show that it is able to reduce wide-area bandwidth usage by 30 to 40% and migration time by 20%