Parallel Multigrid Solvers using OpenMP/MPI Hybrid Programming Models on Multi-Core/Multi-Socket Clusters
Kengo Nakajima (The University of Tokyo)
Abstract:
OpenMP/MPI hybrid parallel programming models were implemented to 3D finite-volume based simulation code for groundwater flow problems through heterogeneous porous media using parallel iterative solvers with multigrid preconditioning. Performance and robustness of the developed code has been evaluated on the “T2K Open Supercomputer (Tokyo)” and “Cray-XT4” using up to 1,024 cores through both of weak and strong scaling computations. OpenMP/MPI hybrid parallel programming model demonstrated better performance and robustness than flat MPI with large number of cores for ill-conditioned problems with appropriate command lines for NUMA control, first touch data placement and reordering of the mesh data for contiguous “sequential” access to memory.
Keywords:
Large Scale Simulations in CS&E, Numerical Algorithms for CS&E