On the Value of Coordination in Distributed Decision Making

We discuss settings where several “agents” combine efforts to solve problems. This is a well-known setting in distributed artificial intelligence. Our work addresses theoretical questions in this model which are motivated by the work of Deng and Papadimitriou [11]. We consider optimization problems, in particular load balancing and virtual circuit routing, in which the input is divided among the agents. An underlying directed graph, whose nodes are the agents, defines the constraints on the information each agent may have about the portion of the input held by other agents. The questions we discuss are: Given a bound on the maximum out-degree in this graph, which is the best graph? What is the quality of the solution obtained as a function of the maximum out-degree? ∗An earlier version of this paper appeared in Proceedings of the 34th IEEE Symposium on Foundations of Computer Science, Palo Alto, 1993, under the title “on the value of information in coordination games.” †Supported in part by NSF Grant CCR-9309456. Address: Department of Information and Computer Science, University of California, Irvine, CA 92717. Email: [email protected] . ‡Address: Lab. for Computer Science, MIT, 545 Technology Square, Cambridge, MA 02139. Work done while at ICSI, Berkeley, and supported by a Rothschild post-doctoral fellowship. Email: [email protected] .