Performance Analysis of MPI Parallel Programs on Xen Virtual Machines

Recently, HPC in the Cloud has emerged as a new paradigm in the field of parallel computing. Most of cloud systems deploy virtual machines for provisioning resources. However, in a virtual machine environment, there is still no mature method to analyze performance of MPI parallel programs. In this paper, we propose a series of innovative methods for performance analysis of MPI parallel programs on Xen virtual machines, including performance data collection through instrumentation and sampling, performance bottleneck diagnosis using PAM and DBSCAN clustering algorithms, and root cause analysis of bottleneck using a rough set approach. The experimental results show that our method can not only automatically locate performs bottlenecks of MPI parallel programs in the XEN virtual machine environment, but also pinpoint the root causes of these bottlenecks automatically.