Experiments in Distributing and Co-ordinating Knowledge

Abstract

We have built a distributed chess program running on a network of workstations. The program consists of independent advisor processes each using different chess knowledge to evaluate the position; each advisor suggests a move to a co-ordinating process, which uses a selection policy to choose the move to be played. The program uses the knowledge embedded in a known sequential program, except that it is organised differently in different advisors. Experiments are reported with several different knowledge-distribution and move-selection policies. An encouraging result is that some programs built are significantly stronger than their sequential prototypes when using as few as seven workstations.