Transparent Parallelization of Constraint Programs on Computer Clusters

The availability of multi-processor machines and computer clusters offer significant opportunities for constraint programming. They also present a fundamental challenge: how to exploit parallelism transparently to speed up constraint programs. Our recent research showed how to parallelize constraint programs transparently on parallel computers. This paper generalizes the approach for cluster of computers and provides preliminary experimental evidence of its benefits. Indeed, on some difficult job-shop scheduling problems, the transparent parallelization provides dramatic, superlinear speed-ups ranging from 61 to 42,090 on 16 processors for depth-first search and from 9 to 48 for limited discrepancy search.