Cloud benchmarking and performance analysis of an HPC application in Amazon EC2

Computing has revolutionized the way computational resources are delivered by making them available on-demand and at a low cost. Consistent with the democratization of technology in this era of globalization, it has made computing power and data storage more affordable and accessible than ever before. The existence of hundreds of cloud providers today attests to the massive success of cloud computing. We performed a feasibility study in the category of HPC in the cloud. It consisted of porting a CFD simulation from a Cray XC40 supercomputer (Beskow) to Amazon Cloud using the tool ParallelCluster and running a set of benchmarks to help us understand the results. We find the performance of our target application is 1.17 times better in the cloud, and we show strong scaling up to 2304 processes. This dataset shows the results in both the Amazon Web Services Cloud environment and Beskow for:the OSU benchmarks for all the collective MPI routines (in the form of plots and a pdf file with tables)the NAS benchmarks in shared memory and for up to 8 nodesParaview solution files for the problem we tackleComparison plots between the performance on Beskow and the AWS EC2 clusterStrong scaling results