Branch and Bound Algorithm for Parallel Many-Core Architecture

Abstract

In recent years, computer environment using multiple processors such as multi-core and many-core device attracts attention due to the limit of performance improvement per processor. In this paper, we propose a new algorithm for the combinatorial optimization problem using a parallel search method called LazySMP to efficiently use many-core processors. LazySMP is a method based on the iterative deepening depth-first search, which is used for board searching of chess and shogi software. In this method, the search results are saved in a table that all processes can share, and the results are used in the search of other processes to shorten the search time. In the proposed method, Lazy SMP is applied to the branch and bound method. Specifically, it performs a branch and bound method that iteratively deepens in all threads and save a part of the result of some nodes in the shared hash table. Then, when it performs the subsequent searches, the hash table is referred to instead of researching the nodes. Our aim is to make efficient use of many-core processors. We make computer experiments with the traveling salesman problem as the benchmark in order to verify the performance of the proposed method.