(A) mode driven dataflow model for AND/OR parallelism of logic programs

Abstract

An important objective in logic programming research is to improved the performance of logic programming systems. A promising approach, in this respect, is to introduce parallelism into the program evaluation mechanism. Many of these models have, however, been based on the conventional approach of organizing concurrent components of computations as independent communicating processes. Some of them have the data-driven computation organizations. In this thesis, we investigate the application of the data-driven organization to the evaluation of lgic programs. Several kinds of parallelism can be found in logic programs. It is natural to associate the logic programming with a dataflow architecture to cope with the high parallelism. Since input/output relationships of arguments are not fixed in the logic programs, for the conventional dataflow models to exploit AND-parallelism, they require a run-time support to control the dynamic token flow which induces severe overhead, or the join operation whose cost is not cheap. Until now, there has been no evidence that the dataflow model has been successfully employed to exploit the AND-parallelism for pure logic programs. In this thesis, we propose D-Model, which is a mode driven dataflow model for AND/OR-parallelism of logic programs. In this model dataflow graphs supporting AND/OR-parallelism for logic programs can be produced through a graph generation procedure based on the mode which can be predicted at compile-time. Although the dataflow graph in the model does not provide an optimal AND-parallelism, there is no run-time overhead such as dynamic arcs between the literal nodes and no join operation between the literals. D-Model adopts the dataflow call-by-value mechanism and reduces the amount of multiset. The OR-parallelism is also supported. The mode can be generated at compile-time through a refined static data dependency analysis, which is a variation of the static data dependency analysis proposed by Chang, et. a...