Maintaining Hierarchical Distributed Consistency

Recent work on bringing classical CSP techniques to distributed environments has focused either on backtrack search or on local consistency. We consider the “natural” issue of combining the two techniques. A new distributed algorithm, called MHDC (Maintaining Hierarchical Distributed Consistency), is presented which incorporates distributed consistency into asynchronous backtracking. One of its main characteristics is to consider consistency maintenance as a hierarchical task. Enforcing the hierarchies of consistency and performing search can then be done with a high degree of asynchronism. This gives the agents more flexibility and freedom in the way they can contribute to search, and increases parallelism. As expected, the experimental results show that substantial gains in computational power can result from combining distributed search and distributed local consistency algorithms.