Release-time aware VM placement

Abstract

Consolidating virtual machines (VMs) on as few physical machines (PMs) as possible so as to turn into sleep as many PMs as possible can make significant energy savings in cloud centers. Although traditional online bin packing heuristics, such as Best Fit (BF), have been used to reduce the number of active PMs, they share one common limitation; they do not account for VM release times, which can lead to an inefficient usage of energy resources. In this paper, we propose several extensions to the original BF heuristic by accounting for VMs' release times when making VM placement decisions. Our comparative studies conducted on Google traces show that, when compared to existing heuristics, the proposed heuristic reduces energy consumption and enhances utilization of cloud servers.