Author(s): Tung T. Vu, Alvin W. Z. Chew, Adrian Wing-Keung Law

Linked Author(s): Wing Keung, Adrian Law

Keywords: Computational hydrodynamic simulations, PGAS-UPC, MPI, OpenMP, high performance computing

Abstract: In this study, an approach for computational hydrodynamics (CHD) simulations with high performance computing using the parallel architecture, PGAS-UPC, is developed. UPC hybrids the advantages of MPI's scalability and OpenMP's direct memory access, and could minimize the possible stagnation in computational speedup with an increasing number of CPUs deployed for the simulations. For evaluation, the computational performance of UPC-CFD is compared with that of OpenMP and MPI. Two incompressible CHD cases, namely (a) Blasius boundary layer flow, and (b) Poiseuille's flow, are tested. The accuracy of the proposed UPC-CHD model for both CHD cases is first verified by comparing with the respective analytical predictions. The computational efficiency of UPC-CHD is then demonstrated by comparing with the respective versions of MPI and OpenMP at their basic designs, in a SGI UV-2000 server with 100 CPUs. The positive results from the computational performance analysis have demonstrated that UPC-CFD is a viable alternative for future large-scale CHD simulations

DOI:

Year: 2017