The objective of our research is to come up with an end-to-end distributed congestion control mechanism for TCP that allocates the network link capacities efficiently and fairly. A network quasi-static model has been considered. Simplifying assumptions have been made. Indeed, the number of TCP connections and the routing tables have been assumed to vary slowly.; We have studied the associated static model and have defined an optimality allocation criterion that is based on proportional fairness. Starting from a characterization of the optimal allocation using an optimization problem, we have proposed, first, a numerical algorithm and then have identified a distributed congestion control algorithm. In this algorithm, each source and each network link executes a local procedure. A source controls its (average) rate using information received from the network. A link updates regularly a local variable termed congestion measure.; A simulation model has been developed to validate and evaluate the performance of our congestion control algorithm. It has been written in C++ using a simulation library called CSIM. (Abstract shortened by UMI.)... |