Achieving scalability in parallel tabled logic programs

Tabling or memoing is a technique where one stores intermediate answers to a problem so that they can be reused in further calls. Tabling is of interest to logic programming because it addresses some of the most significant weaknesses of Prolog. Namely, it can guarantee termination for programs with the bounded term-size property. Tabled programs exhibit a more complex execution mechanism than traditional Prolog´s left-to-right search with backtracking. The reason is that Prolog programs are highly recursive and generate multiple answers. This rather involved execution mechanism requires a more complex implementation than traditional Prolog. The declarative nature of tabled logic programming suggests that it might be amenable to parallel execution. On the other hand, the complexity of the tabling mechanism, and the existence of a shared resource, the table, argues that parallelism might be limited, and that performance for real applications might never scale. In this work we prove that parallel tabling is indeed scalable for real applications by experimenting the OPTYap parallel tabled system on a scalable shared-memory machine.