Harnessing Memory Page Distribution for Network-Efficient Live Migration

Live migration of Virtual Machines (VM) is an important part of cloud and data center management that allows load balancing, physical machine (PM) maintenance, and consolidation. Maintaining low network usage spikes and fast times during live migration are necessary goals to ensure that the performance of the migrating VM as well as that of other VMs in the cluster is not degraded. In this paper we present a proactive memory sharing protocol that utilizes the commonality of memory pages between VMs due to similar resident Operating Systems and applications to achieve these goals. The proposed method identifies and distributes important shared pages in a cluster over time substantially reducing network usage and enabling fast times during migration. We implement our approach on a test cluster and demonstrate the time and bandwidth benefits achieved with real world loads for both pre and post copy live migration.