DMCA
A Parallelization of State-of-the-Art Graph Bisection Algorithms on the Grid
BibTeX
@MISC{Dun_aparallelization,
author = {Nan Dun and Kenjiro Taura and Akinori Yonezawa},
title = {A Parallelization of State-of-the-Art Graph Bisection Algorithms on the Grid},
year = {}
}
Share
 |
 |
 |
 |
||
OpenURL
Abstract
We have parallelized Reactive Randomized Tabu Search (RRTS) graph partition algorithm and adapted it to the grid environment. Speedup is achieved by parallelizing scoring phase of RRTS. We also introduced two techniques: multilevel scoring and parallel initialization. By using multilevel scoring, tabu lengths are highly tuned to adapt large graphs. And by parallel initialization, dispersed computing efforts are centralized to output more consistent graph partitions. Experiments show that our approach significantly outperformed state-of-the-art heuristics by partitioning large graphs with higher quality and efficiency. 1.
