Utilizing Open MPI Middleware for HPC Clusters

Open MPI, an open source middleware implementation of Message Passing Interface (MPI), helps simplify the porting and execution of MPI-based applications and supports many High-Performance Computing (HPC) platforms, interconnects, and environments. This paper outlines the Open MPI architecture and its flexible runtime configuration options, and offers tips that can help maximize bandwidth performance. HPC clusters have a hierarchical architecture. Within this architecture, middleware and libraries bind compute nodes together to perform computations and communicate across nodes. The challenge lies in how to best distribute application processes across the clusters for maximum performance.