Analysing the Migration Time of Live Migration of Multiple Virtual Machines

Workload consolidation is a technique applied to achieve energy efficiency in data centres. It can be realized via live migration of virtual machines (VMs) between physical servers with the aim to power off idle servers and thus, save energy. In spite of innumerable benefits, the VM migration process introduces additional costs in terms of migration time and the energy overhead. This paper investigates the influence of workload as well as interference effects on the migration time of multiple VMs. We experimentally show that the migration time is proportional to the volume of memory copied between the source and the destination machines. Our experiment proves that the VMs, which run simultaneously on the physical machine compete for the available resources, and thus, the interference effects that occur, influence the migration time. We migrate multiple VMs in all possible permutations and investigate into the migration times. When the goal is to power off the source machine it is better to migrate memory intensive VMs first. Kernel-based Virtual Machine (KVM) is used as a hypervisor and the benchmarks from the SPEC CPU2006 benchmark suite are utilized as the workload.