Science in the Cloud: Experiences Building an International Cloud for Computational Science

Cloud computing, where general purpose hardware is virtualized and made available in a ubiquitous environment, is widely used in the commercial sector. Cloud computing offers scalability and fault tolerance which is essential to operations that require a continuous presence. In contrast, scientific research tends to be bursty where large amounts of computational services are needed for limited periods. The scalability of a cloud can help address this need by making available computational nodes on demand. Clouds offer another advantage for simulations and models. Many simulations and models are computed by using tested and robust codes which may not have been written for parallel environments. When the codes are run they execute within a single core, which does not take full advantage of current multi-core processors. In a cloud the codes can be run on multiple cores simultaneously and in identical computation environments. Each instance of execution can be run with a different set of initial conditions, providing a richer set of results. This allows non-parallel codes to exhibit parallel behavior without alteration. We present our experiences building an international cloud for computational science and the results of running simulation codes within the cloud.