Research On Congestion Control Algorithm For High Speed Communication Networks

With the rapid development of Internet and the improvement of different users for quality of service of networks congestion has been the "bottleneck problem" of the development of communication networks. Congestion control for high-speed communication networks has become the most interesting topic both in communication and control society. This dissertation focuses on the algorithm design of congestion control and its stability analysis for ATM and TCP networks.The problem of traffic management of ABR service in the ATM networks is considered firstly. An improved control algorithm is proposed based on discrete model, which is based on the structure of follow-up control systems. The effect of time delay is considered in the system and the method of parameters setting is given further, which can guarantee the stability of the closed-loop system. This scheme realizes the congestion avoidance and the fair allocation of available bandwidth in the dynamic environment and improves the performance of the existing algorithm.To improve system response and simplify algorithm, a fast queue-tracing controller is presented in the paper. It consists of two parts: tracing controller and feed-forward controller that can realize the tra(?)ing the desired queue length with no error and effective restraint for disturbance respectively. Through the theoretic analysis, the parameter region is got which can guarantee both the controller and close-loop system. And the dynamic Max-Min fairness for different connections under the bursty VBR background traffic is realized. The algorithm needs less control parameters and suits for the condition with large variation of round trip time.The characteristic of large propagation time delay and high bandwidth delay product in wide area networks is considered in the third part of the paper. The queue length is regarded as a scale to evaluate the QoS, and the available bandwidth for ABR traffic is regarded as unknown but bounded disturbance. The dual Smith predictor is designed. The stability of system is proved from the view both of time and frequency domain. To reduce the effect of mismatch of model and saturation of controller, the feedback filter and assistant controller are designed further. The scheme can operate under the environments of larger round trip time and variation of time delay. The response caused by the oscillation of available bandwidth is restrained fully. Moreover, the bandwidth information is not necessary.Next, to overcome the limitation in the networks modeling of the congestion control, the Fuzzy PID controller is designed based on the ER feedback mechanism. It need not model the networks. According to the network dynamic performance and considering the nonlinear factors in the system, the control rules are made and the PID parameters are gained through optimization. The scheme combines the merits of traditional PID control and Fuzzy inference. Simulations are made under various network environments and system performance is analyzed. The results show the better adaptation and robustness.Finally, the dissertation studies the active queue management algorithm for HSTCP. The characteristic of HSTCP network and difference of dynamic window adjustment between high speed TCP and traditional TCP are analyzed first. A two-order linear time-delay model for HSTCP is gained through local linearization. Then, PI controller is designed and its stability analysis is given analytically. The proposed algorithm is suit for both standard TCP and HSTCP and it is named SPI(scalable proportional integral). The method of choice of adaptive control parameters is given further. The simulations are made using NS2 simulator and the detailed performance comparison with Drop-tail algorithm is given. The results verify the effectiveness of SPI controller under two heterogeneous types of flows. The simulations show that the algorithm has high throughput and better fairness and friendliness under heterogeneous network environments.