A Study on Today’s Cloud Environments for HPC Applications

With the advance of information technology - as building smaller circuits and hardware with lower energy consumption - the power of HPC (High-Performance Computing) resources increases with the target to employ complex large-scaling applications. In traditional computing, an organization has to pay high costs to build an HPC platform by purchasing hardware and maintaining it afterwards. On-premises HPC resources may not satisfy the demand of scientific applications when more computing resources for large-scaling computations are requested than own resources are available. Especially for SMEs (small and medium-sized enterprises), a temporarily-increasing computing demand is challenging. Cloud computing, is an on-demand, pay-as-you-go model, that provides us with enormous, almost unlimited and scalable computing power in an instantly-available way. Therefore, it is a valuable topic to develop HPC applications for the cloud. In this paper, we focus on developing an HPC application deployment model based on the Windows Azure cloud platform, and an MPI framework for the execution of the application in the cloud. In addition, we present a combined HPC mode using cloud and on-premises resources together. Experiments that are employed on a Windows cluster and the Azure cloud are compared and their performance is analyzed with respect to the difference of the two platforms. Moreover, we study the applied scenarios for different HPC modes using cloud resources.