An Experimental Study of Adaptive Application-Sensitive Partitioning Strategies for SAMR Applications

Dynamic adaptive mesh re nement methods for the numerical solution to partial di erential equations yield highly advantageous ratios for cost/accuracy when compared to methods based upon static uniform approximations. Parallel/distributed implementations of these techniques, however, present signi cant challenges in dynamic data-distribution and load-balancing. This is because the choice of the \best" partitioning technique and associated partitioning parameters depends on the nature of the application and its run-time state. This paper presents an experimental study of an adaptive application-sensitive meta-partitioner for structured adaptive mesh re nement (SAMR) applications that dynamically selects and con gures partitioning strategies at run-time based on system parameters and the current application state. The selection is based on a run-time classi cation of application state and an application-centric classi cation of the partitioners. Experimental results show that adaptive partitioning can improve application performance.