DISTRIBUTED SEARCHES - A BASIS FOR COMPARISON

This paper attempts to explore the impact of greatly enhanced computing power on the alpha-beta algorithm, notorious for its high processing requirements. On a limited set of test positions, it explores, under some limitations, the question of how best to distribute the power of multiple processes. The main matters probed at least in tendency are the vital questions of when to split the search among processes and, once splitting has been decided upon, to which process to split. Stress is laid on the apparent limitation of speed-up, for which a severe law of diminishing returns soon sets in under any reasonable conditions, even when the best feasible splitting strategy is utilized. It is shown, again in tendency, that fertile further exploration essentially requires architectures with intrinsically low communication and context-switching overheads. A uniform hardware/software platform Network C-Linda, is presented as essentially free of distortion under carefully stated conditions.