Efficient Backtracking in And-Parallel Implementations of Non-deterministic Languages

We consider the problem of eeciently supporting backtracking in independent and-parallel non-deterministic systems. We consider this problem in the context of logic programming, although the solution proposed is suuciently general to be applicable to any non-deterministic language or system. EEcient implementation of backtracking is a diicult task, the primary reason being that backtracking is an inherently sequential activity. Due to this complexity, most existing and-parallel systems either do not support backtracking over parallel calls or they perform very poorly on non-deterministic benchmarks. We propose an implementation model for eeciently supporting backtracking. Our model employs various optimizations, as well as a novel memory organization scheme in which processors are allowed to traverse each others' stacks to achieve this eeciency. The solution developed has been implemented in the ACE Parallel Prolog system. The performance of our implemented system is analyzed on a variety of non-deterministic benchmarks. The results achieved are remarkable: speedups achieved during forward execution are not lost in heavy backtracking activities and frequently super-linear speedups are obtained thanks to a semi-intelligent backtracking scheme.