Design considerations and performance of a scalable version of a nonhydrostatic atmospheric model

The Naval Research Laboratory's Coupled Ocean/Atmosphere Mesoscale Prediction System (COAMPS) is being developed into a parallel, scalable model in a joint collaborative effort with Lawrence Livermore National Laboratory. The initial focus is on the atmospheric forecast model, which solves a coupled, three-dimensional set of dynamical equations using finite differences. A distributed/shared memory parallel programming paradigm is used. Distributed memory parallelism is achieved through a two-dimensional domain decomposition technique, with internodal communication accomplished using Message Passing Interface (MPI), and OpenMP is used to provide parallelism within a node. Initial performance results on both the IBM-SP and Cray-T3E are presented.