并行广度优先搜索算法分析-analysis of parallel breadth-first search algorithm.docxVIP

摘要随着计算机科学的不断发展，图的搜索技术已经渗透到语言学、逻辑学、物 理、化学、电子、通信、数学等诸多学科领域，特别是网络技术的快速发展和并 行计算机的出现，并行处理进入了前所未有的繁荣期，图的遍历也显得越来越重 要。广度优先搜索算法是图论的一个基本问题，也是一个热点问题。广度优先搜 索算法的并行化已经成为目前亟待解决的问题。本文首先介绍了广度优先搜索并行算法的研究现状。然后，在 Cilk++运行时 系统上，实现了用“bag”的数据结构代替共享队列的一种优化方法，该算法是基于 层同步思想的一个细粒度并行算法，能够更容易的并行代码。其次，在分布式系 统上，实现了基于邻接矩阵一维划分的并行广度优先搜索算法。最后，在分析和 研究现有的广度优先搜索并行算法的基础上，结合现有的并行计算环境，提出了 一种新的广度优先搜索并行算法。该算法是在分布式系统上提出来的，结合了层 同步的思想和邻接矩阵二维划分的思想。通过大量实验证明，我们提出的新算法 在时间复杂度上达到了更好的性能，在分布式系统上具有更好的适用性。关键字：广度优先搜索算法分布式系统层同步邻接矩阵二维划分AbstractWith the development of computer science, the search technology of graphs has emerged in linguistics, logic, physics, chemistry, electronics, communication, mathematics and some other fields of science. Especially with the fast progress of internet technology and the appearance of parallel computer systems , parallel manage has in an unprecedented boom period and the problem of traversing graphs is more and more important. Breadth-First search algorithm is a basic problem of graph theory and also a hot problem. The parallelism of Breadth-First search algorithm has been an urgent problem.First of all, this paper introduces the status about parallel breadth-first search algorithms. Secondly, we replace the shared queue with a new “bag” data structure which is more amenable for code parallelization with the cilk++ run-time model, and this is a fine-grained parallel algorithms based on layer synchronization theory. Thirdly, we achieve a parallel breadth-first search algorithms based on one-dimensional decomposition of the adjacency matrix corresponding to the graph. Finally, analyzing the methods present of parallel breadth-first search and combining with the current parallel system conditions we have, this paper proposes a new method of parallel breadth-first algorithm. This algorithm combines the level-synchronous theory and two-dimension partition thought of the adjacency matrix. Plenty of experiments prove that this new algorithm reaches a better performance of time complexity, and mor