This paper presents an experimental evaluation of a suite of dynamic partitioning/load-balancing techniques for adaptive grid hierarchies that underlie parallel structured adaptive mesh refinement applications. Partitioners evaluated include those included in popular software libraries such as GrACE, Vampire, and ParMetis. The overall goal of the research presented in the paper is an application-centric characterization of the partitioners as a function of the number of processors, problem size, and partitioning granularity.
展开▼
