Control in Parallel Constraint Logic Programming

Although constraint satisfaction problems can be specied declaratively, the cost of actual solving depends heavily on the order of computation. The ideal combination of declarative problem description and automatic satisfaction is di cult; because determination of the optimal order of computation in advance is impossible in general. Our new proposal, Parallel Constraint Logic Programming(PCLP), aims to solve this dilemma. In PCLP, computation is described in terms of parallel transformation of constraints. This approach can explore natural parallelism, and is suitable for dynamic control of the computation. Our prototype system PARCS, based on this framework, can handle constraints on Herbrand universe and nite domains by using SLD-resolution and consistency techniques such as forward checking. To control the computation dynamically, priority is associated with each process according to run-time information. In addition, learning mechanisms adjust the parameters of the scheduling policy so that scheduling becomes near optimal.