Evaluating Multithreaded Workloads in CMP Virtualization Environment

Abstract

Xen is popular in Server Consolidation, in which performance evaluation is very important to get the best configuration. However, existed works paid much attention to single-thread benchmarks in SMP, and few works are focused on multithreaded benchmarks in chip-multiprocessors (CMP). Therefore, we dedicate to evaluate Xen hyper visor using a new multithreaded benchmark suit named PARSEC developed for CMP architecture, in which different configurations are tested to conclude rules of running multithreaded applications in Xen virtualization environment. Experimental results show that it is trivial difference between binding DomU's vcpus to share L2 cache cores and not share L2 cache cores. So, we do not need to care the SMP or CMP architecture when running multithreaded applications.