Algorithm for Load-balancing Adaptive Scienti c Simulations

Adaptive scientiic simulations require that periodic repartitioning occur dynamically throughout the course of the simulation. The computed repartitionings should minimize both the inter-processor communications incurred during the iterative mesh-based computation and the data redistribution costs required to balance the load. Recently developed schemes for computing repartitionings provide the user with only a limited control of the tradeoos among these objectives. This paper describes a n e w Uniied Repartitioning Algorithm that can gracefully tradeoo one objective for the other dependent upon a user-deened parameter describing the relative costs of these objectives. We s h o w that the Uniied Repartitioning Algorithm is able to minimize the precise overheads associated with repartitioning as well as or better than other repartitioning schemes for a variety of problems, regardless of the relative costs of performing inter-processor communication and data redistribution. Our experimental results show that the Uniied Repartitioning Algorithm is extremely fast and scalable to very large problems.