An Empirical Analysis Of Cloud Platforms For High Performance Computing

High-performance computing (HPC) tasks can benefit from the scalable and adaptable resources cloud computing provides. But clouds come with additional overhead due to virtualization, which has to be considered. Using the HPCC benchmark suite, this study empirically compares the performance of three popular open-source cloud systems for HPC: OpenStack, OpenNebula, and Nimbus. The essential HPC components—processor, memory, storage, and network—are tested. Across all benchmarks, OpenStack's performance was the closest to that of native hardware, indicating the least amount of virtualization overhead. In all compute and memory tests, OpenStack's results varied between 6 and 11% from native ones. Performance on the network and storage was proportionate to native. Because it has the least negative effect on performance, the results point to OpenStack as the best open-source platform for putting HPC applications on cloud infrastructure.