A Decentralized Variable Ordering Method for Distributed Constraint Optimization

Abstract

Many different multi-agent problems, such as distributed scheduling can be formalized as distributed constraint optimization. Ordering the constraint variables is an important preprocessing step of the ADOPT algorithm [1], the state of the art method of solving distributed constraint optimization problems (DCOP). Currently ADOPT uses depth-first search (DFS) trees for that purpose. For certain classes of tasks DFS ordering does not exploit the problem structure as compared to pseudo-tree ordering [2]. Also the variables are currently ordered in a centralized manner, which requires global information about the problem structure. We present a variable ordering algorithm, which is both decentralized and makes use of pseudo-trees, thus exploiting the problem structure when possible. This allows to apply ADOPT to domains, where global information is unavailable, and find solutions more efficiently. The worst-case pseudo-tree depth resulting from our algorithm is √ 2kn, where n is the number of variables, and k is maximum block size in constraint graph. The algorithm has space complexity of O(kn) and time complexity of O(n+|E|+k 3 2 n 1 2 ), where E is the set of edges in a constraint graph.

Distributed Bayesian: A Continuous Distributed Constraint Optimization Problem Solver

“…Various methods exists for the generation of a pseudo-tree (e.g. pseudotree ordering (Chechetka & Sycara, 2005), MLSP tree generation (Maheswaran, Tambe, Bowring, Pearce, & Varakantham, 2004), and BFS construction (Chen, He, & He, 2017)). A commonly used method is to produce a depth-first search (DFS) tree from the constraint graph by a DFS traversal.…”

Synchronizing for Performance - DCOP Algorithms

“…Some DCOP algorithms organize the agents into a pseudo-tree (also known as a DFS tree) in order to gain run-time performance by the resulting enhanced concurrency (Modi et al, 2005;Yeoh et al, 2010;Chechetka and Sycara, 2006). In a DFS tree, if two agents share a constraint they are in an ancestordescendent relation.…”

“…In NCBB (Chechetka and Sycara, 2006) when agent A i receives a search message from its parent, it has a view of the SP that is backwards-consistent. A i then invokes multiple subtreeSearches (which are similar to requesting an FB-estimate) one per each child and value it has.…”

“…Section 3 presents classes of synchronization for DCOP search algorithms and classifies several well known algorithms into these classes. All categorized DCOP algorithms are treated in three groups -DCOP algorithms that use a DFS tree (Modi et al, 2005;Chechetka and Sycara, 2006;Yeoh et al, 2010); algorithms that do not use a DFS tree (Gershman et al, 2009); and algorithms that incorporate multiple search processes (Meisels, 2008;Netzer et al, 2010). The computation of expected idle time for different algorithms, as a function of their synchronization class is presented in Section 4, along with a general probabilistic model of asynchronous idle time.…”

GHENeSys, uma rede unificada e de alto nível.

“…• Capítulo 5: apresenta o estudo de caso realizado para avaliar o ambiente de modelagem. ;FOX, 1991;LIU;SYCARA, 1993;SYCARA, 1995;CHECHETKA;SYCARA, 2006 (AALST, 1994), inclusive a relação com o paradigma de objetos (SIBERTIN-BLANC, 1994).…”