The main purpose of this thesis is to compare the performance overhead of the virtualization infrastructures KVM and VMWare. All the experiments are carried out by using the Red Hat Enterprise Linux(RHEL) Operating System version 6.1. The study focuses on the performance of disk I/O operations, memory operations and CPU operations. The benchmarking tools used are Iozone for disk I/O, Ram Speed for memory and UnixBench for CPU.
First a set of benchmarking tests are carried out by using a Bare Metal installation of RHEL 6.1 on a Dell Poweredge R710 server. Next the exact same set of benchmark tests are run after installing RHEL 6.1 on a single virtual machine running on KVM on the same server. Finally VMWare ESXi 5.0 is installed on the server and RHEL 6.1 is installed on a single virtual VMWare machine. In this way the performance overhead of the two virtualization infrastructures KVM and VMWare is measured and compared. Each benchmarking test is run in each of the three cases sufficiently many times to produce statistically significant results.
The VMWare I/O disk performance is mostly from 20 to 30% better than KVM, with a few exceptions. And generally the VMWare I/O performance is 10-15% less than the Bare Metal performance. The memory performance overhead is relatively smaller. KVM performs better than VMWare for block sizes of 4MB and less, while the results show the opposite for block sizes larger than 4MB. When testing pure ALU usage, there is almost no virtualization overhead. There was some overhead for the other UnixBench CPU tests and in all these cases VMWare was performing better than KVM. Our general conclusion is that the virtualization overhead is less for VMWare than for KVM.