Consistent linear speedup in parallel alpha-beta search

Alpha-beta search is an important heuristic technique used in pruning a game tree. Recently, parallelism has been introduced to further improve the efficiency of such heuristic searches. The paper presents a new parallel alpha-beta search algorithm that applies a prioritizing scheme. To verify its performance, the algorithm was implemented on sequent symmetry shared memory multiprocessor system. Experimental results show that for the best ordering uniform trees, one is able to obtain a consistent speedup of a linear order to the number of processors used. An overall speedup with random node ordering is illustrated. The paper concludes with some characteristics of the alpha-beta search that make the application of priorities suitable for its corresponding parallel algorithm.<>