Improving CPU Performance of Xen Hypervisor in Virtualized Environment

Abstract

In the large organizations, which are spread across large geographical area, virtualization plays an important role. This is because the amount of resources of a single physical server is large enough to be completely utilized by a single operating system and hence result in wastage of resources. The virtualization of hardware resources allows more than one virtualized servers to share same physical machine. A significant performance drop is observed in a virtualized operating system in comparison to when it runs directly on hardware. This depends on both the CPU technology as well as the virtualization technique used. The performance of the virtual machines also depends on the virtual CPU scheduling technique used. In XenServer, the Credit Scheduler assigns each virtual CPUs to physical CPUs asynchronously. But if the workload is concurrent, there is a need for synchronization. In this paper, we ran our guest operating system two biggest server virtualization platforms of recent times, namely, VMware ESXi and Citrix XenServer, which both use different approaches to virtualization. After analyzing different parameters in concurrent workload, we found that our guest performs better on VMware than XenServer. Then we proposed an improved virtual CPU scheduling algorithm for Xen hypervisor, which supports synchronization of concurrent programs and significantly reduces the CPU waste time in concurrent workload. 

Keywords: Cloud Computing, Performance, Virtualization, VMware, Xen, Virtual Machine Monitor, Hypervisor, Credit Scheduler, Virtual CPU Scheduling