An Analytical Model for Virtualization

Virtualization is a technique that partitions a physical computer server into two or more virtual subsystems (or guests), each of which behaves as if it were a separate physical system. Through virtualization, more efficient usage of resources can often be achieved. However, virtualization comes with cost. In particular, the virtualization manager and the individual operating systems installed on the virtual subsystems consume resources, such as CPU computing power. Since all the guests share the same resources in a given computer server (consisting of CPUs and peripherals such as disks), a research issue is how many guests can be comfortably supported. In this paper we present an analytical model to study the performance of virtualization with respect to the parameters including job service time distributions, the number of guests, the virtualization overhead, the memory thrashing overhead, and the disk access speed.