Load balancing sequences of unstructured adaptive grids

Publisher Website
Google Scholar
Abstract
Mesh adaption is a powerful tool for efficient unstructured grid computations but causes load imbalance on multiprocessor systems. To address this problem, we have developed PLUM, an automatic portableframework for performing adaptive large-scale numerical computations in a message-passing environment.This paper makes several important additions to ourprevious work. First, a new remapping cost model ispresented and empirically validated on an SP2. Next,our load balancing strategy is applied to sequences ofdynamically adapted unstructured grids. Results indicate that our framework is effective on many processors for both steady and unsteady problems with several levels of adaption. Additionally, we demonstratethat a coarse starting mesh produces high quality loadbalancing, at a fraction of the cost required for a fineinitial mesh. Finally, we show that the data remappingoverhead can be significantly reduced by applying ourheuristic processor reassignment algorithm.

This publication has 8 references indexed in Scilit: