Data-parallel concurrent constraint programming

Abstract

With the advent of cost-effective massively parallel computers, researchers conjecture that the future concurrent constraint programming system is composed of a massively parallel constraint solver as the back-end with a concurrent inference engine as the front-end (Cohen, Comm. ACM 33 (7) (1990) 52–68). This paper represents, to the best of our knowledge, the first attempt to build a concurrent constraint programming system on a massively parallel SIMD computer. A concurrent constraint programming language called Firebird is presented. Firebird can handle finite domain constraints and supports both concurrency and data-parallelism. As a result, it is suitable for implementation on both multiprocessors and SIMD computers. Concurrency arises from the stream and-parallelism of committed-choice logic programming languages. Our SIMD implementation supports or-parallelism but stream and-parallelism is implemented sequentially. In a nondeterministic derivation step, one of the domain variables is selected to create a choice point. All possible alternatives are attempted in parallel. Data-parallelism is exploited in the resulting or-parallel execution. In this paper, we first present the Firebird computation model. We then present an SIMD implementation of the Firebird language on the DECmpp massively parallel computer based on the data-parallel abstract machine (DPAM). A performance analysis of the system is also presented.