Static Parallel Arc Consistency in Constraint Satisfaction

Constraint satisfaction problems (CSPs) are ubiquitous in artificial intelligence; versions arise in areas such as vision, design, Boolean satisfiability, cryptarithmetic, and database retrieval. Most researchers have solved CSPs on sequential computers; relatively few have addressed the use of parallel computers for these problems. Among those who have investigated parallel approaches, several authors have solved CSPs using parallel tree search algorithms, while others have pre-processed constraint networks using parallel consistency algorithms. No one, however, has measured the specific work performed by the individual processors using arc consistency techniques to pre-process a constraint network. In this paper we introduce two Static Parallel Arc Consistency algorithms (SPAC-1 and SPAC-2), which ensure arc consistency of a finite domain binary constraint network, and which are designed for any general-purpose parallel processing computer. Through simulation, we measure work performed by each processor and compare it with work performed by existing sequential and parallel algorithms. Results show that our parallel arc consistency algorithm can be used to pre-process a constraint network with good speedup and utilization.