FEEDBACK CONTROL OF CONGESTION IN DATA NETWORKS: RATE BASED ALGORITHMS


A decentralized architecture for congestion control in data networks is investigated. Within this architecture, each switching node is equipped with a controller intended to keep the buffer occupancy below a fixed level. All the controllers interact through the network. The statics and dynamics of the resulting system are investigated. The cases of a single and multiple congested nodes are analyzed. In the analysis problem, it is shown that there exist a set of controllers that renders the systems asymptotically stable in the steady state with max-min fairness property. In the design problem, both adaptive and robust approaches are investigated, and methods for selecting the controllers' gains are suggested. The approach developed has been utilized for data traffic management (ABR service) in ATM networks.