Applications of Game Theory to Multi-Agent Coordination Problems in Communication Networks
Ramaswamy Pillai, Vinod
MetadataShow full item record
Recent years there has been a growing interest in the study of distributed control mechanisms for use in communication networks. A fundamental assumption in these models is that the participants in the network are willing to cooperate with the system. However, there are many instances where the incentives to cooperate is missing. Then, the agents may seek to achieve their own private interests by behaving strategically. Often, such selfish choices lead to inefficient equilibrium state of the system, commonly known as the tragedy of commons in Economics terminology. Now, one may ask the following question: how can the system be led to the socially optimal state in spite of selfish behaviors of its participants? The traditional control design framework fails to provide an answer as it does not take into account of selfish and strategic behavior of the agents. The use of game theoretical methods to achieve coordination in such network systems is appealing, as it naturally captures the idea of rational agents taking locally optimal decisions. In this thesis, we explore several instances of coordination problems in communication networks that can be analyzed using game theoretical methods. We study one coordination problem each, from each layer of TCP/IP reference model - the network model used in the current Internet architecture. First, we consider societal agents taking decisions on whether to obtain content legally or illegally, and tie their behavior to questions of performance of content distribution networks. We show that revenue sharing with peers promote performance and revenue extraction from content distribution networks. Next, we consider a transport layer problem where applications compete against each other to meet their performance objectives by selfishly picking congestion controllers. We establish that tolling schemes that incentivize applications to choose one of several different virtual networks catering to particular needs yields higher system value. Hence, we propose the adoption of such virtual networks. We address a network layer question in third problem. How do the sources in a wireless network split their traffic over the available set of paths to attain the lowest possible number of transmissions per unit time? We develop a two level distributed controller that attains the optimal traffic split. Finally, we study mobile applications competing for channel access in a cellular network. We show that the mechanism where base station conducting sequence of second price auctions and providing channel access to the winner achieves the benefits of the state of art solution, Largest Queue First policy.