Virtual Machine Performance upon Intensive Computations

Abstract

The Virtual Machine (VM) technology has increased considerably within the last two decades and the performance of VMs seems closest to the host machine. This has motivated the use of VM in High Performance Computing (HPC). Running a scientific computation on VMs can severely impact the performance of applications. In addition, choosing the proper VM to run a specific application while minimizing the lost of performance is not an easy task. However, several virtualization solutions have been proposed. This paper presents the result of a comparison of the performance of the most commonly used VMs solutions: OpenVz, Linux-Vserver, LXC, XEN, KVM and VMware ESXi. The performance of these VMs are evaluated with the NAS benchmark, Lmbench, IOzone and Intel ® MPI Benchmark taking into account the consumption of resources such as CPU, memory, latency, disk and network communication. The result shows that some virtualization solutions present better performance in consumption of some of the previously mentioned resources than others. This work aims at helping a scientist in selecting the VM suitable to compute a specific task without a significant overhead on applications performance.