APPLICATIONS OF THE CONTRACT NET FRAMEWORK: SEARCH

We discuss the implementation of heuristic search algorithms in a distributed problem solver whose processors interact according to the contract net protocol. Task distribution is viewed as a local mutual selection process based on a two-way transfer of information between processors with tasks to be executed and processors with knowledge-sources capable of executing those tasks. As an example of the approach, we consider the N Queens problem. We then derive measures of the speedup that can be expected from the application of a distributed processor to search problems that involve regular trees, and discuss the effect of coupling between processors on speedup. Bounds are developed for the number of processors that are required to achieve maximum speedup.