Parallel tetrahedral mesh adaptation with dynamic load balancing

The ability to dynamically adapt an unstructured-grid is a powerful tool fo r efficiently solving computational problems with evolving physical feature s. In this paper, we report on our experience parallelizing an edge-based a daptation scheme, called 3D_TAG, using message passing. Results show excell ent speedup when a realistic helicopter rotor mesh is randomly refined. How ever, performance deteriorates when the mesh is refined using a solution-ba sed error indicator since mesh adaptation for practical problems occurs in a localized region, creating a severe load imbalance. To address this probl em, we have developed PLUM, a global dynamic load balancing framework for a daptive numerical computations. Even though PLUM primarily balances process or workloads for the solution phase, it reduces the load imbalance problem within mesh adaptation by repartitioning the mesh after targeting edges for refinement but before the actual subdivision. This dramatically improves t he performance of parallel 3D_TAG since refinement occurs in a more load ba lanced fashion. We also present optimal and heuristic algorithms that, when applied to the default mapping of a parallel repartitioner, significantly reduce the data redistribution overhead. Finally, portability is examined b y comparing performance on three state-of-the-art parallel machines, (C) 20 00 Elsevier Science B.V. All rights reserved.