Transputer-based implementation and evaluation of parallel Prolog interpreter

Abstract The aim of parallel symbolic computing is to provide inference speeds beyond those attainable in sequential models by using parallel machines. The paper describes the implementation issues and analyses the performance results of a parallel Prolog interpreter running on a message-passing multinode machine. The execution model adopted exploits the AND/OR-parallelism inherent in logic programs, while supporting sequential Prolog semantics. The interpreter is written in Occam and runs on a transputer parallel system with up to four fully connected nodes. Speed-up evaluation shows the actual degree of parallelism that the interpreter can recover and points out the costs of parallelization. The pairing of the interpreter with routing support makes it suitable for scalable parallel systems.