Multi-node Of The Atm Network Congestion Control Algorithm Research

The problem of congestion control is more complicated because there are so many heavy uncertainties in ATM networks. On the one hand, there is saturated nonlinearity of transmit rate, process capacity of node because of the physical constraint of networks equipment. On the other hand, there are serious measure errors and disturbance in the networks on account of stochastic delay, stochastically accessed users and prior burst traffic in high-speed networks. Most of the existing congestion control methods on rate-based feedback didn' t consider such the uncertainties. At the same time, the disturbance in the queue model is increased because of the coupling between the multiple-node switch. It requires more demand on the stability and transient performance of the controller. In a word, the control purpose is how to deal with the disturbance and high- priority traffic, the proposed controllers can guarantee the stability of closed-loop system and reduce the magnitude and persistent time of the congestion. It can also decrease the ratio of the loss cells and maximize the utilization of network resource with fairness requirement.The thesis first builds up the network fluid multiple-node model, and takes the link delay and the measure errors into account, and also considers the high -speed traffic and the coupling between the switch as disturbance. And then a predictive congestion control scheme is proposed. The simulation results show that the algorithm enhances the utilization of the resource and guarantees the stability of close-loop control system and the fairness of the bandwidth allocation. Then adaptive predictive congestion control algorithm is designed to estimate the parameters of the model which is variant because of the time-varying stochastic delay, accessed users. The simulation results demonstrate that it can rapidly reflect the stochastically access of the users and reach the stability when the number of users is finite. Finally, the thesis designs neural-based congestion control algorithm. A self-tuning algorithm is used to identify the linear queuing model and a BP neural network is used to detect the unknown nonlinear dynamics. The simulation results demonstrated the algorithm is stable, convergent and fair, and robust with respect to variants of stochastic delay and stochastically accessed users.