Towards consistency of virtual machine migration in Software-Defined Network

Virtualization has been widely used in Data Centers, which are running the applications that process data operations or provide services. Migration of virtual machines happens regularly for system load balance, scheduled maintenance, disaster recovery or even in case of unscheduled server downtime. However, packets are still forwarded along the previous path when the migration has completed, which leads to a long time connection interruption and can be abstracted as a consistency problem. We define Connection Recovery Time (CRT) as the duration from the time of migration initiation to the time of reconstructing forwarding rules for migration. In this paper, we analyze the reasons of long CRT in different kinds of Layer-2 switching implementations and propose a novel proactive mechanism to reduce CRT in Software-Defined Network (SDN). Finally, we evaluate the performance of our mechanism, and the CRT in our mechanism can be greatly reduced to the same order of magnitude with general TCP roundtrip time and the occupied hardware resources are hugely reduced.