点击次数：

论文类型：期刊论文

发表时间：2011-09-01

发表刊物：ICIC Express Letters

收录刊物：EI、Scopus

卷号：5

期号：9 B

页面范围：3511-3516

ISSN号：1881803X

摘要：Parallel Dynamic Programming, though a popular and robust strategy for solving the large scale single shortest path problems, is sometimes inconvenient to be used as the number of node is restricted by execution time and the parallel machines are relatively difficult to use, manage, maintain and may not be accessible to most re- searchers. In this paper, we propose a parallel dynamic programming method based on GPU-acceleration, which maps parallel DP algorithm to thread block on consumer-level graphics cards. The analytical results demonstrate that the proposed method increases the number of node, speeds up its execution and provides a feasible parallel DP solution for ordinary users. ? 2011 ISSN 1881-803X.