A Queueing-Theory-Based Bandwidth Allocation Algorithm for Live Virtual Machine Migration

Abstract

Live virtual machine migration has become a core system management tool to achieve load balancing and power saving among different servers. The recent studies mainly focus on the performance improvement for live VM migration. However, there has been little attention on the migration's impact on VMs co-located both source and destination servers during the migration. The impact, which is mainly caused by competition for sharing resources, lowers the real effect of live VM migration. To describe the real effect, we first use a weighted sum approach to formalize the performance of live VM migration and its impact. And then we use queueing theory to model the essential relationship between a VM's performance and it's residual bandwidth. Afterwards, based on the performance model, we propose a bandwidth allocation algorithm for live VM migration to minimize the sum. Our experiment results show our algorithm may improve the real effect of live VM migration by at most 71%, in comparison with the bandwidth allocation approach for live migration in QEMU/KVM.