Desktop workload characteristics and their utility in optimizing virtual machine placement in cloud

Abstract

Optimally placing virtual machines on numerous physical servers saves clouds' hardware resources. In the case of the virtual desktop cloud, the placement task becomes more difficult because the workload of virtual desktop changes quickly over time.In this paper, we first study the workload of CPU, memory, and hard disks of 172 machines that provide virtual desktops to remote users. We also thoroughly analyze CPU usage metric and show that office users' desktops often repeat a certain pattern every workday. We then evaluate the virtual machine placement algorithms including PBA, which utilizes the correlation between the CPU usage patterns to find the suitable server for a virtual desktop, and First Fit, a widely used placement algorithm in cloud infrastructure.