Backtracking in Distributed Constraint Networks

The adaptation of software technology to distributed en- vironments will be an important challenge in the next few years. In the scope of constraint reasoning, few works have been published on the adaptation of algorithms searching for a solution in a constraint network to distributed constraint networks. This paper presents a new search procedure for finding a solution in a distributed constraint net- work. Although based on the principle of backtracking to ensure the completeness of search, this procedure allows a high level of asyn- chronism, i.e., simultaneous search on independent parts of the net- work. Furthermore, it fits its behavior to the structure of the con- straint graph in order to minimize message passing and to avoid useless restorations when a dead-end is reached. We also present a generic distributed method for computing any variable ordering heuristic.