Design, Analysis, and Implementation of a Parallel Tree Search Algorithm

The alpha-beta algorithm for searching decision trees is adapted to allow parallel activity in different parts of a tree during the search. The algorithm has been implemented in a procedural simulation language (GASP IV). The simulation environment provides the illusion of multiple software processes and multiple hardware processors. A number of preliminary experiments have been done to gather statistics on run time, nodes scored, and nodes visited. Results indicate that a substantial reduction in time of search occurs because of the use of parallelism. An analytic expression for the storage requirements of the algorithm is derived. The analysis provides an example of the classical tradeoff between time and storage.