QoS-driven Probabilistic Runtime Evaluations of Virtual Machine Placement on Hosts

Abstract

We tackle the cloud providers challenge of virtual machine placement when the client experienced Quality of Service (QoS) is of paramount importance and resource demand of virtual machines varies over time. To this end, this work investigates approaches that leverage measured dynamic data for placement decisions. Relying on dynamic data to guide decisions has, on the one hand, the potential to optimize hardware utilization, while, on the other hand, increases the risk on the provided QoS. In this context, we present three probabilistic methods for evaluation of host suitability to allocate new virtual machines. We also present experiments results that illustrate the differences in the outcomes of presented approaches.