WALM: Workload-Aware Live Migration of Virtual Machines

Abstract

Abstract Live Virtual Machine (VM) migration migrates VM from one physical machine to another while applications inside the VMs are in their execution state. It is crucial to migrate the VMs residing within the server as quickly as possible when an imminent server failure is detected. Depending on how the VM’s memory is migrated, VMs can undergo different types of migrations such as pre-copy, post-copy and hybrid. The nature of the workload significantly impacts the performance of VM migration. This paper presents Workload-Aware Live Migration of Virtual Machines (WALM), a new approach for choosing the most optimal migration method for a VM. WALM dynamically detects the nature of the VM workload by monitoring its CPU, network and memory usage and migrates the VM by choosing the most optimal migration method among pre-copy, post-copy and hybrid methods. Additionally, WALM also presents an intelligent hybrid migration method that dynamically chooses the most optimal point to switch from pre-copy to post-copy in hybrid migration. Experiments show that WALM successfully chooses the most optimal migration method for different types of workloads, optimizing the VM migration process. WALM outperforms pre-copy by 47% performance improvement for network-intensive workloads, over 30% improvement for CPU-intensive workloads and outperforms hybrid migration by 12% performance improvement for memory-intensive workloads.