| In broadband communication networks where a wide diversity of users require their own quality of services, the complexity of network control problems calls for a decentralized networking approach rather than a centralized one. With decentralized control, networks are designed to operate by multiple performance objectives of users (system controllers) rather than by a single objective imposed by a network provider. Moreover, this approach is nowadays gaining a lot of interest as the universal cooperation of users may not be guaranteed in heterogeneous internetworks, e.g., the Internet. The first part of this dissertation formulates a decentralized bandwidth allocation model. Each user is assumed to have its own performance objective that is simply assumed to be strictly concave, and its own strategy space which is defined by its maximum and minimum demands. We address several important properties of a Nash equilibrium, as well as its uniqueness, and convergence of some distributed algorithms. As a case study, the following chapter treats the generalized power criterion, presenting a simple algorithm to compute the equilibrium, and suggesting a pricing scheme for a load balancing in multiple links. The next chapter contains two general topology network models: In parallel link networks, users are allowed to perform both flow control and routing, and by formulating a general topology network model, the problem of decentralized virtual path capacity allocation in ATM networks is investigated. We show that a special case of the latter model corresponds to the constrained version of the bottleneck flow control. The next part reports study on the constrained max-min fair bandwidth allocation, interpreting the fair rate vector as a limit of the equilibrium of the decentralized bandwidth allocation model. This study suggests an efficient algorithm for computing the fair rate vector, and proposes a new framework for the available bit rate ATM service such that the max-min optimality can be achieved and maintained by end-systems. Finally, an analytical result on Braess' paradox, a phenomenon that may occur in a competitive routing environment, is presented with simulations, and an effective resource pricing scheme is mentioned. |
