Summary: | 碩士 === 國防管理學院 === 資源管理研究所 === 86 === Tree-Searching algorithms may solve many problems which are indicated in tree structure. "Branch-and-bound strategy" can effectively decrease searching times and shorten problem-solving time. We can see the obvious performance if we make them parallel. Under the consideration of cost and performance, we use Parallel Virtue Machine(PVM) to reorganize current work-stations to create parallel computing environment. We propose a parallel computing model in the parallel computing environment, called master-agent(thread)-control. We applied the concepts of branch-and-bound, multi-processes, and shared-memory to design a multi-thread tree-searchingalgorithm and then experiment and analyse the performance.
We use the probability model to analyse our multi-thread tree-searching algorithm. Our analysis covers computing time, communication time and acceleration anomaly. We use the traveling salesperson prblem (TSP) as an example to experiment. We decompose the branching tree of the TSP so that threads can be assigned to solve those subtrees(subproblems) concurrently and cooperatively. We make experiment with the master-agent-control model on the PVM computing environment which is organized by 2 work-stations, 4, and 6. The experimental results reveal that our multi-thread tree-searching algorithm and master-agent(thread)-control model are effective and cost less.
|