Simultaneous Mesh Generation and Partitioning for Delaunay Meshes.

摘要: In this paper, we present a new approach for the parallel generation and partitioning of unstructured 3D Delaunay meshes. The couples mesh problems into single optimization problem. Traditionally, these two are solved separately, first generating (usually sequentially) then mesh, either sequentially or in parallel. traditional approach, overheads due to I/O data movement exceed 50% total execution time. Even if schemes employed, movement, synchronization, structure translation high; applications which require frequent remeshing (e.g. crack growth simulations), prohibitive. We method solving placement problem simultaneously with By eliminating unnecessary redundant cache, local, remote memory accesses, can speed up redistribution process, very large meshes, by almost an order magnitude compared approaches. Our results show that achieve nearly perfect equi-distribution elements over processors, while maintaining reasonably good separator size, all improving quality many inherent constrained generation.