An Empirical Study of VMM Overhead, Configuration Performance and Scalability

Virtual Machines have been a key enabling technology for Cloud Computing. Compared with the physical machine, VMs aim to enhance resource utilization via on-demand and dynamic resource allocation. However, with VMs, new performance issues are introduced such as hyper visor overhead and scheduling effectiveness, the impacts of configuration on system performance, and mutual influences among VM instances. This empirical study is motivated by 5 research problems including performance comparison between physical and virtualized environments, hyper visor overhead analysis, scalability analysis, and the impacts of VM management strategies. Using VMWare ESXi as a representative VM product, we have set up two experiment environments of different scales. A series of experiments were designed and exercised using data-intensive and computing intensive benchmarks. The results show the performance patterns of VM systems under different usage scenarios and system configurations. The study helps to analyze and understand Cloud performance characteristics from the users´ perspectives.