EXPLOITATION OF FINE-GRAIN PARALLELISM IN LOGIC LANGUAGES ON MASSIVELY-PARALLEL ARCHITECTURES

Logic programming has been drawing increasing attention as a parallel programming paradigm due to its ease of programming. However, an important question remains: Can logic programs achieve scalable performance on general-purpose massively parallel architectures? The problem is to determine how to efficiently incorporate parallelism in logic programs within the architectural features of such machines. We frame the question by identifying important issues that are crucial for scalable performance. We have systematically developed the logical organization of a parallel execution model based upon data-driven principles of execution. In this paper, we present some details of the model, particularly its fine-grain parallelism support and the self-scheduled inference as the basis for its distributed implementation.