Parallel processing of deductive databases on a commercial multiprocessor

Abstract

A processing strategy for large knowledge bases, which features large granularity of computation because it works with relations, has been proposed. The performance behavior of this strategy was tested on a parallel processor; specifically, it was implemented on a Sequent Symmetry S81. The data-partitioning parallelization approach was used. Experimental results show that real problems have unbalanced trees, therefore increasing the difficulty in the use of the available parallelism. The balanced parallelism can be artificially increased by partitioning the extensional database. This allows not only a better load balancing in the multiprocessor, but also faster join and union operations, which greatly affect performance.<>