Dynamic Branch & Bound Distribué

Distributed constraint satisfaction problems based on nogoods reasoning [2] seem to be the best approach in terms of theoretical time complexity. However, only few optimization protocols get the most out of valued nogoods’s power [16]. In this paper, we present a new search algorithm for distributed constraint optimization problems (DCOPs), called Distributed Dynamic Branch & Bound (DisDB&B). Agents assigns their variables sequentially and forwards valued nogoods (lower bounds) synchronously to the next unassigned agent. Combining lower bounds on inferred valued nogoods can help to speed up the search, and prune infeasible sub-problems. We show that the algorithm proposed here (DisDB&B), is optimal and guaranteed to terminate, having polynomial space complexity. Detailed experimental results show that on benchmark problems, the proposed algorithm achieve over an order of magnitude reduction on terms of messages exchanged between agents.