Abstract: High Energy physics computing is a high-performance computing application,which highly requires computing power,and CPU utilization directly affects the computational efficiency of high-energy physics.Virtualization technology has demonstrated a great advantage in achieving high utilization of resources and resource sharing.This paper gave performance testing and optimization of KVM.Firstly,we showed testing results and quantitative analysis of performance between KVM and physical machines in aspect of CPU,disk IO and network IO.Then,we demonstrated various factors in hardware level and kernel level that affect the performance of KVM through the architecture of KVM virtual machine,including EPT (Extended Page Tables) and CPU affinity which are optimized for KVM.Experiment results show the penalty of CPU performance for KVM can be decreased to about 3%.Finally,virtual cluster of high ener-gy computing was shown and the performance of virtual cluster can meet the needs of high energy physics computing.

Key words: HPC,KVM,CPU affinity,Extended page table