Incentive engineering for distributed non-cooperative networks

My thesis studies incentive engineering for distributed non-cooperative networks. The need for new reliable distributed resource allocation methods considering selfish users emerged, as the Internet has evolved into a vast distributed system, comprised of independent and potentially selfish users who value the benefit they derive from the network much more than the efficiency of the network as a whole. With this motivation, I have abandoned the assumption of cooperative users in communication networks. I have worked on applying prices and other incentives to modeling, congestion control and optimization of resource distribution in distributed networks of selfish and non-cooperative users. My thesis cosisted of modeling of users' behaviors in pursuit of maximum self-benefit, development of distributed resource allocation algorithms, and analysis of the proposed mechanisms for various networks.;First, I proposed a generic user access control mechanism for the emerging Internet offering a premium class of services that are billed based on usage.;Second, using the generic user access control mechanism, I developed a user access rate control mechanism for a single trunk with Erlang loss dynamics. The proposed rate control mechanism is a distributed Multiplicative-Increase/Multiplicative-Decrease (MIMD) access-rate control algorithm leading to Nash equilibria. The stability of the Nash equilibria is proved by using Lyapunov stability theory.;Third, I developed a user access rate control mechanism for a differentiated services (diffserv) network with a single node. The proposed user access rate control algorithm is an extension of the current TCP's additive increase and multiplicative decrease (AIMD) congestion control to accommodate different user demands. I showed that a higher service class guarantees higher priority to receive network resources when the modified TCP AIMD congestion control is used.;Last, avoiding the assumption of voluntary congestion control in TCP networks, I proposed a selfish window congestion control framework. When end-users pursue their own maximum benefits, it is doubted that the network can achieve its global congestion control objective. A distributed window control algorithm is proposed to drive the selfish users toward a social optimum point. The proposed window control algorithm has desirable properties such as achieving fairness, maximizing total utility, and providing disincentives to altering the protocol.