Parallel implementation of an unstructured grid-based Navier-Stokes solver

The unstructured grid, Navier-Stokes flow solver, USM3Dns, belonging to the TetrUSS software system has been parallelized for efficient operation both on shared memory systems (e.g. Origin 2000) and on a cluster of workstations connected by a local area network. A single block tetrahedral, unstructured grid is partitioned into a user specified number of contiguous partitions (blocks) each containing nearly equal number of grid cells. The grid partitioning is accomplished by either an indigenously developed partitioning algorithm or using available Metis partitioning program. The communication between partitions is accomplished using Message Passing Interface (MPI) calls suitably embedded into the code. By innovative use of specialized data structure and by using non-blocking communication, nearly linear ‘speed-ups’ have been demonstrated. Sample calculations on an ONERA M6 wing and a F16 configuration show the efficiency and accuracy of the parallel implementation.