Cooperative search vs classical algorithms

We have presented, in a previous work ([15]), a cooperative parallel search for solving the constraint satisfaction problem. We run independently solvers based on Forward-Checking with Nogood Recording. The solvers exchange nogoods via a process (”the manager of nogoods”) which regulates the exchanges. Solvers exploit the nogoods they receive to limit the size of their search tree. Experimentally, we have shown the interest of our approach from a parallel viewpoint, namely we have obtained linear or superlinear speed-up. However, we haven’t studied its behavior with respect to classical algorithms. In this paper, we first improve the exploitation of received nogoods. Then, we provide experimental comparisons between the cooperative method and some state-of-the-art algorithms. In particular, we observe that the cooperative search with at least four solvers (even in some cases from two solvers) is faster than classical algorithms like FC or MAC.