Analysis And Design Of Distributed Congestion Control Algorithm Based On Queue Management

With the rapid development of Internet, the scale of network tends to widen, and its structure becomes more complex. Also the bandwidth, numbers of users as well as types of network transactions grow fast. How to efficiently manage the resource of network, avoid and control network congestion, and guarantee the quality of service (QoS) have been a hot research issue in communication networks. TCP congestion control mechanisms operate at the source side of network. However, TCP congestion control mechanism's effectiveness is limited since the congestion is detected after the change in packet drops or data transmission delays. So it is hard to satisfy the QoS merely on the basis of TCP congestion control. Since the intermediate nodes (router or switch) have more information about network status, the link algorithm becomes a key congestion control scheme to assist TCP. As a link algorithm, active queue management (AQM) aims to provide an early notification of congestion degree to end-hosts by dropping or marking packets. It can achieve low transmission delay, small queue length jitter and high link utilization by managing the queue length stable at a low level in routers.The dissertation conducts further researches on AQM, improves some AQM algorithms and proposes some novel AQM methods. Simulation results in NS2 demonstrate the effectiveness of our proposed schemes. The main contents and contributions are list below:(1) Random Exponential Marking (REM) could not detect congestion effectively, and suffers from poor adaptability to various network conditions. This dissertation analyzes the control property of REM and proposes two improved REM algorithms—enhanced price-based REM (EPREM) and dynamic REM (DREM). EPREM introduces the deviation of traffic rate into price deriving a PID control form. EPREM can achieve faster convergence speed, and improve the responsiveness of the TCP/AQM system. DREM introduces two new variables, the queue factor and load factor, to divide the regulating procedure of queue length into four cases. Also, the key parameter of DREM is adjusted online to assist the'additive increase multiplicative decrease (AIMD)'strategy-based TCP congestion control mechanism. Simulation results demonstrate that DREM improves the adaptability and robustness for active queue management.(2) On account of the problem that traditional AQM schemes lack adequate capability of adaptation to dynamic networks, this dissertation presents a novel double model congestion control method (DMC). DMC synthesizes the advantages of fuzzy logic and REM controlling modes by using the segment control strategy. Specifically, when the error of queue length is larger than the expected threshold, fuzzy controller is employed to improve the responsiveness and adaptability, otherwise, REM is used to get stable queue length. The simulation results in NS2 demonstrate that DMC is a robust AQM scheme with good flexibility.(3) In light of the difficulty of modeling the TCP/AQM system, a neuron controller with fuzzy self-tuning gain (FN-AQM) is proposed for active queue management. Queue length and traffic rate are both employed as congestion indicators which detect both current and incipient congestion states. Combining the advantages of neuron control and fuzzy control strategies, the end-to-end mark probability is calculated by single neuron controller, in which the weights are adjusted on-line by supervisory Hebb learning rule. Additionally, fuzzy logic control is used to tune the gain of the neuron controller dynamically aiming at improving the network performance. The proposed algorithm exhibits good adaptability and self-learning ability, being simple in form and easy to implement.(4) On the problem of traditional AQM schemes'poor robustness against dynamic network conditions, a robust active queue management scheme named PID-SMC is proposed via sliding mode variable structure control. Integral separation PID sliding surface and enhanced reaching law are designed to improve the dynamic performance and decrease the high-frequency chattering. The simulation results demonstrate the effectiveness of PID-SMC.(5) To solve the problem that most AQM schemes suffer from big queue jitters and low link utilization in wide area networks with large delays, a novel congestion control scheme named ISC is proposed. Smith predictor is employed to compensate for the round trip time (RTT) delay, and internal model control is used to design an executable PI-type feedback controller. Simulation experiments indicate that ISC is able to keep the queue length stable and maintain high link utilization. Based on ISC, we employ a modified Smith predictor with disturbance rejection component and propose a two-degree-of-freedom congestion control algorithm named TRBM. The feedback controller is designed by internal model control theory for set-point tracking. The disturbance rejection controller is derived from frequency-domain analysis to reject external disturbance. Simulation results in NS2 demonstrate that TRBM can effectively overcome the negative influences cased by RTT delay and external disturbance.Finally, we draw our conclusions and propose some further research directions.