Using SIMD parallelism to support rule-based systems

A parallel algorithm for the implementation of Rete networks on a parallel array processor is proposed. Although special-purpose architectures have been proposed for directly implementing production systems, the availability of such machines is very limited. Proposed implementation algorithms, which take advantage of more widely available parallel machines, are more practical. Such an approach was taken in a previous work where a number of workstations connected via a local area network were considered. Here an SIMD (single instruction, multiple data) array processor is used. The Rete network is functionally divided into two subnetworks: the discrimination network and the join network. It is the join network where most of the processing load is carried out. The operation of the join network is the main objective of the SIMD implementation. The proposed algorithm emphasizes data parallelism as opposed to functional parallelism. In this type of parallelism, the data are partitioned into sets of approximately equal size and distributed among several processors. The data are then operated upon simultaneously by similar operators. The underlying architecture is described