A Portable Tool for Running MPI Applications in the Cloud

Cloud computing founds on the pay-per-use paradigm, and offers a simple way to acquire any kind of resources through an as-a-service approach. As a consequence, the cloud is currently seen as a viable and inexpensive alternative to customary parallel/distributed computing solutions, ranging from small clusters to GRIDs, which were widely used in the past. In this paper we present a portable cloud tool, cMe (cloud MPI enabler), able to run on any cloud provider, which builds up dedicated cloud-based MPI-enabled clusters. The objective is to help the end-users to choose dynamically both the cloud provider and the amount of the resources to be used. The cloud application performs simple job scheduling, based on the same principles adopted in the typical HPC environments (i.e., different jobs never share computing resources), but also enables the dynamical acquisition of resources. For user convenience, the application is provided with a simple web interface that helps the resource acquisition process.