Evaluating performance of sequential programs in virtual machine environments using design of experiment

Sequential or single-threaded programming is already a complicated task. The shift to multithreaded programming only increases complexity, cost and code of the program and moreover increases the effort required for debugging, to avoid deadlocks, race conditions and other problems associated with the parallel programs. In this paper we present a comprehensive evaluation methodology to benchmark the performance of sequential programs by running them in several virtualized environments by the technique called virtualization and checking the throughput in those environments by means of a method called experimental design or design of experiment. The benchmark in this paper is running on Linux guests on virtualization technology (VT-x) enabled platforms. The results show that the how the performance of sequential program in both Xen and VMware changes with respect to the factors viz. type of hypervisor, size of RAM and the number of virtual CPUs affecting it.