Asymmetric distributed constraints satisfaction problems

Roie Zivan, Amnon Meisels

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

1 Scopus citations

Abstract

Distributed constraint satisfaction problems (DisCSPs) with asymmetric constraints reflect the fact that agents may wish to retain their constraints private. Brito and Meseguer proposed a model for asymmetric constraints which they term Partially Known Constraints (PKC). In the PKC model each binary constraint is divided between the two constraining agents. In order to solve the resulting DisCSP with asymmetric constraints, a two phase asynchronous backtracking algorithm was proposed [?]. In the first phase an asynchronous backtracking algorithm is performed, in which only the constraints held by the lower priority agents are examined. When a solution is reached, a second phase is performed in which the consistency of the solution is checked again, according to the constraints held by the higher priority agents in each binary constraint. The present paper proposes a one phase asynchronous backtracking algorithm which solves DisCSPs with asymmetric constraints. In the proposed asynchronous backtracking for asymmetric constraints algorithm (ABT_ASC) agents send their proposed assignments to all their neighbors in the constraints graph. Agents assign their local variables according to the priority order as in ABT but check the constraints also against the assignment of lower priority agents. When an agent detects a conflict between its own assignment and the assignment of an agent with a lower priority than itself, it sends a Nogood to the lower priority agent but keeps its assignment. Agents which receive a Nogood from higher priority agents, perform the same operations as if they have produced this Nogood themselves. As in ABT [?], they remove their current assignment from their current-domain, store the eliminating N ogood and reassign their variable. The ABT_ASC algorithm is evaluated experimentaly on randomly generated DisCSPs and is shown to outperform the 2-phase ABT with respect to two different distributed performance measures. ABT_ASC performs fewer non-concurrent constraints checks by a factor of 6, for the harder problem instances. The load on the network is very similar for both algorithms, counting the total number of messages sent by both algorithms.

Original languageEnglish
Title of host publicationLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Pages875
Number of pages1
DOIs
StatePublished - 2005
Externally publishedYes
Event11th International Conference on Principles and Practice of Constraint Programming - CP 2005 - Sitges, Spain
Duration: 1 Oct 20055 Oct 2005

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume3709 LNCS
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

Conference11th International Conference on Principles and Practice of Constraint Programming - CP 2005
Country/TerritorySpain
CitySitges
Period1/10/055/10/05

Fingerprint

Dive into the research topics of 'Asymmetric distributed constraints satisfaction problems'. Together they form a unique fingerprint.

Cite this