A parallel inference model for logic programming

In this paper, we describe a parallel inference model for logic programming on general-purpose multicomputers. In the model, input clauses are partitioned into subsets, and resolution is conducted on each subset concurrently. The partitions are dynamically adjusted via clause migration as inference proceeds. This allows each processor to work on virtually the whole clause set while a shorter resolution cycle is achieved. In the context of AND/OR tree space search, the parallel model explores another dimension of parallelism in addition to AND/OR parallelism. It implicitly forces multiple processors to jointly search the same path that leads to a refutation. Problem-solving heuristics can be incorporated in the parallel model systematically to determine clause partitions and guide inference. With a distance measure derived from problem-solving heuristics, a partition that has the best combination of clause subsets and a small rate of clause migration can be obtained using existing clustering algorithms. Clause migration decisions can also be made based on the distance measure. Last, we point out that all the mechanisms of the parallel model can be efficiently supported by a connection graph. The graph also simplifies the implementation of the subsumption strategy.