3D Navier-Stokes Flow Analysis for Shared and Distributed Memory MIMD Computers.

摘要

Multiple instruction multiple data (MIMD) parallel computers were investigated as a means of obtaining the immense computer power required to analyze the flow about a complete aircraft configuration. A computational fluid dynamics computer program for solving the Navier-Stokes equations, with coupled nonequilibrium chemistry, was parallelized on three MIMD computer: the Intel Touchstone distributed memory computer, the Cray Y/MP supercomputer, and the Silicon Graphics IRIS 4-D/380 shared memory computer. The program used an LU-SGS implicit algorithm. The parallelization was performed using permanent domain decomposition. The parallel efficiency and the main limitations to efficient parallelization were evaluated for a series of real life engineering problems. On the Intel Touchstone, the parallel efficiency decreased slowly as the number of processors was increased. The main deterrent to parallel performance was the latency time for interprocessor communication. On the Cray Y/MP, the parallel efficiency dropped rapidly as the number of processors was increased. The main deterrent to parallel performance was the reduction in vector length during domain decomposition. 3D Navier-Stokes, Viscous-Flow, Compressible-Flow, Hypersonic Parallel, Shared Memory, Distributed Memory.