Interference and Topology-Aware VM Live Migrations in Software-Defined Networks

Abstract

Live migration of virtual machines (VMs) serves as the key technique that enables the resource management, server maintenance, and load balancing in cloud datacenters, while introducing a sufficient amount of performance degradation to both the source and destination physical machines (PMs) during and after the migration. Existing works generally focus on either minimizing the VM migration time (or downtime) or reducing the performance degradation for an individual VM migration from the host aspect. Jointly optimizing the performance of multiple concurrent VM migrations and the performance interference among VMs from the network aspect, however, receives comparably little attention. Leveraging the benefits of Software-Defined Network (SDN), in this paper, we propose ITEM, an Interference and Topology-awarE VM live Migration strategy to minimize the overall VM migration cost, which jointly optimizes the VM migration time during the migration and the VM performance interference after the migration. Specifically, our ITEM strategy seeks to decide the optimal migration destination PM, the migration path, and the allocated network bandwidth for a single migration (ITEM-S) and a set of concurrent migrations (ITEM-M). Extensive experiments using OpenFlow in the Mininet virtual network environment show that our ITEM strategy can significantly reduce the migration cost and improve the network bandwidth utilization, in comparison to the state-of-the-art VM migration strategies, yet with acceptable runtime overhead.