In this research project, we are developing mesh partitioning techniques that consider the static and dynamic features of distributed systems and adaptive applications. Currently, we have have the initial version of PART that considers the static system and application features. In particular, PART exploits the heterogeneity in processor and network performance as well as the element types within an application. Our initial experiments demonstrate that PART can aid in reducing the parallel execution time by 20% as compared to conventional mesh partitioning techniques. Currently, we are working on a parallel version of PART and starting to consider dynamic features of the distributed system and application.
This research supported by a continuing NYI grant from National Science Foundation and AlliedSignal, Inc. The funding has been used to support 3 Ph.D. students and 2 undergraduate students. This work has generated 3 journal papers and 6 conference papers.