An agile framework adaptive to complicated memory workloads for VM migration

A vital advantage of virtual machines (VMs) is live migration - the ability to transfer VMs from one physical machine to another as the VMs continue to offer service. Some well-known techniques have been proposed for live migration, such as pre-copy and post-copy. Unfortunately, these classical techniques are not agile enough in the face of complicated workloads since they are designed to work well in a specific workload. When the workloads of VMs get complicated, even a good migration algorithm may not run very well. This paper proposes an agile framework which determines the type of workloads of a VM by learning the statistics of usages of memory pages and then the framework chooses an appropriate improved algorithm to complete a VM live migration. The experimental results show that the framework is able to identify the type of workloads, and improves the performance of VM live migration.