When a search problem is solved by a distributed network of agents, it is desirable that this search exploits asynchronism as much as possible. This increases parallelism, allows agents to maintain privacy and makes the process robust against timing variations. Recent work on bringing constraint satisfaction techniques to distributed environments has focused either on backtrack search or on local consistency. A high degree of asynchronism can be reached when the two techniques are combined. A generic protocol for maintaining consistency during distributed search is presented. It allows: i) distributed consistency to be enforced concurrently at all levels of the search trees and ii) decisions about instantiations and agent reordering to be taken asynchronously by the agents. We show how its completeness can be ensured during backtracking with polynomial space requirements when a large class of reordering policies is used. A comparative experimental evaluation of several instances of this protocol is presented.
[1]
Leslie Lamport,et al.
Distributed snapshots: determining global states of distributed systems
,
1985,
TOCS.
[2]
Marius-Calin Silaghi,et al.
Asynchronous Search with Aggregations
,
2000,
AAAI/IAAI.
[3]
Ying Zhang,et al.
Parallel and distributed algorithms for finite constraint satisfaction problems
,
1991,
Proceedings of the Third IEEE Symposium on Parallel and Distributed Processing.
[4]
Yves Deville,et al.
Analysis of Distributed Arc-Consistency Algorithms
,
1997
.
[5]
Devendra Kumar,et al.
A Class of Termination Detection Algorithms For Distributed Computation
,
1985,
FSTTCS.
[6]
Gerhard Weiss,et al.
Multiagent Systems
,
1999
.
[7]
Makoto Yokoo,et al.
The Distributed Constraint Satisfaction Problem: Formalization and Algorithms
,
1998,
IEEE Trans. Knowl. Data Eng..
[8]
Eugene C. Freuder,et al.
Contradicting Conventional Wisdom in Constraint Satisfaction
,
1994,
ECAI.