Generalized Predictive Congestion Control Algorithm And Simulation Studies

ATM is a connection-oriented, packet switching, and statistically multiplexing technology. However, With many burst traffic simultaneously arriving at a node, the queue length may become larger and buffer overflow in a moment, or high-speed link is emerged into slower one, there will be in congestion. Once networks congested, the transfer delay grows up and the ratio of cell loss increases rapidly, and the sustained congestion will bring on the breakdown of the network. In a word, it is the first task to control congestion efficiently to increase resource utilization and improve the quality of services.ATM Forum selects rate-based feedback control as the standard algorithm. It merely provides some constructive sketch without detailed plan. The disadvantage of the experimental algorithms is that they all don' t guarantee the fairness and cause the oscillation in queue of nodes or allowed cell rate of users, and lacks of theoretical performance evaluation. Linear control theory based design doesn' t completely consider the uncertainties caused by time-varying transfer delay, saturated non-linearity and dynamical variety of users. These factors not only restrict the use of formal control algorithms but also result in large-scale vibration. All these factors make it hard to decide the system order, which brings the difficulty in implementing model-based approach.The thesis first builds up the network fluid model of single bottleneck node. It only takes the link delay into account, and thinks others (queuing and switching delay)and uncertainties as disturbance. And generalized predictive congestion control algorithm is presented, which ensures global stability of closed-loop system, and then adaptive generalized predictive congestion control algorithm is proposed, which improves the robustness to dynamical variety of users. The simulation results show that the algorithms enhance the transient performance, and increase robustness against uncertainty, and improve networks utilization, and realize fairness of bandwidth allocation in steady state.