Parallel Computation of Solution Adaptive Cartesian Grids with SFC
Jinwook Lee, Stephen Ruffin · 18th AIAA Computational Fluid Dynamics Conference · 2007
The efficient parallel computation of unstructured grid flow solver requires an adequate grid decomposition strategy because of its complex spatial data structure. The difficulties of even and block-contiguous partitioning in frequently adapting unstructured Cartesian grid are overcome by implementing the 3D Hilbert space-filling-curve (SFC). Grids constructed by SFC in a parallel environment promise shorter inter-CPU communication time while maintaining perfect load balancing between CPUs. The load imbalance due to local solution adaption is simply apportioned by re-segmenting the curve into even pieces. The detailed structure of 3D Hilbert SFC and the parallel computing efficiency results based on this grid partition method are also presented.