High-level versus low-level DO-loop parallelization: Results for one testcase of a multi-block solver on a shared memory parallel vector computer
Within the NICE project, supported by the Dutch Foundation HPCN, the multi-block solver for Navier-Stokes equations ENSOLV is being parallelized. Results of the block-loop parallelization of ENSOLV on a 16 processor NEC SX-4, a shared memory parallel vector computer, are presented and compared to those of low-level DO-loop parallelization implemented earlier. The conclusion is that the block loop parallelization gives better speed-up, requires more memory, and gives overall less execution costs.
Unable to display preview. Download preview PDF.
- 1.A. Kassies, M. Amato, J.C. Kok and S. Bosse, Software Design of ENSOLV, A flow solver for 3D Euler/Navier-Stokes Equations in arbitrary multi-block domains, NLR Contract Report CR 93190 L, 1993.Google Scholar
- 2.A.R. Sukul, Predesign of ENSOLV Parallelization on the NEC SX-4, NLR Technical Report TR 96726 L, 1996.Google Scholar
- 3.M. Laban, Parallelized ENSOLV User Requirements, NLR Technical Report TR 96353 L, 1996.Google Scholar
- 4.P. Wijnandts and M.E.S. Vogels, Coarse-grain parallelization of ENSOLV on the NEC SX-4, NLR Technical Report TR 98050 L, 1998.Google Scholar