Performance Analysis of Virtual Machine Monitoring System

In recent years, the computing world has shifted to cloud computing, which provides on-demand access to network and computing resources with cost-effectiveness and scalability. The flexibility and scalability of cloud computing is largely due to the powerful hypervisor for managing the physical platform and accessing resources. Since hypervisors are the most important part of the virtualized system, analyzing its performance is essential to optimize the performance of cloud applications. While providing a variety of features and services, hypervisors have proprietary licenses and may require specialized expertise to operate. In this paper, we studied hypervisors’ performance on virtual machine management under various workloads, and we focus on KVM which utilizes the standard Linux kernel and hardware-assisted virtualization. We conducted experiments to evaluate the performance merits of the hypervisor by varying workload configurations. Our experimental results show that KVM is effective for CPU with a performance overhead of 8% or less, while for memory-intensive workloads, the observed bandwidth of VM is greater than that of the native server. However, for network-intensive workloads, KVM is effective only when the socket size is sufficient and creates an overhead of around 18% when the socket size is smaller than the message size.