Coarse-Grain Parallelization of a Multi-Block Navier-Stokes Solver on a Shared211 Memory Parallel Vector Computer

摘要

The coarse-grained, or block-loop parallelization of the multi-block Navier-211u001eStokes flow solver ENSOLV on a NEC SX-4, a shared memory parallel vector 211u001ecomputer, is discussed. The peformance of the parallel code was tested by running 211u001ethe code on ten benchmark cases, provided by the ENSOLV group. The performance is 211u001emeasured in terms of speed-up, memory usage and execution cost. The results of 211u001ethe benchmark cases are presented. The results are compared to those of the low-211u001elevel DO-loop parallelization implemented earlier. The conclusion based on the 211u001ecomparison of the results, is that for all benchmark cases, except the single 211u001eblock, the block-loop parallelization gives better performance in terms of speed-211u001eup. Although block-loop parallelization requires more memory, it gives overall 211u001eless execution cost.