Characterization of domain-based partitioners for parallel SAMR applications

Abstract

Dynamic adaptive mesh refinement methods for the numerical solution to partial differential equations yield highly advantageous ratios for cost/accuracy as compared to methods based upon static uniform approximations. Distributed implementations of these techniques have the potential for enabling realistic simulations of complex systems. These implementations however, present significant challenges in dynamic data-distribution and load balancing. This paper presents an experimental characterization of dynamic partitioning/load-balancing techniques for adaptive grid hierarchies. Techniques studied include newly proposed as well as existing approaches. The overall motivation is the formulation of policies required to driveadynamically adaptive meta-partitioner for SAMR grid hierarchies capable of selecting the most appropriate partitioning strategy at runtime, based on current application and system state. We believe that such a partitioner can significantly decrease application execution time. Keywords: Parallel/distributed algorithms# Dynamic partitioning/load-balancing# Adaptive metapartitioner # Performance characterization# Run-time adaptation# Structured adaptive mesh refinement. 1