Using Cloud Constructs and Predictive Analysis to Enable Pre-Failure Process Migration in HPC Systems
Commercial cloud offerings rely heavily on virtualization technologies and redundancy to provide reliable customized compute environments tailored to a specific customer needs. These environments are well suited to software development work, web hosting, and even some embarrassingly parallel types of applications that have traditionally been run on High Performance Compute (HPC) platforms. The current interconnects and individual resource reliability in these environments, however, don't lend themselves to the kind of tightly coupled MPI applications that typify today's large scale scientific applications being run on HPC platforms which are specifically designed for these application's requirements (e.g. high reliability, fast processors, high bandwidth low latency interconnects, etc.).