Description:

This paper, by characterizing the latency overhead of vSphere under various scenarios, proves that vSphere can readily handle applications that are sensitive to network I/O latency with very little performance overhead. Scenarios explored in this paper model real-world performance issues, including CPU contention and network bandwidth contention. Our results show that in most cases, vSphere latency overhead is only a few tens of microseconds and rises to only a few hundreds of microseconds in heavily over-committed cases. This paper also presents guidelines for running latency-sensitive applications on vSphere 5.