Research On Active Queue Management Algorithms For Congestion Control In Computer Networks

Congestion control is one of the key techniques in computer networks. The simple drop tail queue management in routers used by the traditional end-to-end congestion control mechanism can lead to many problems, such as long delay, high packet loss rate, global synchronization, fairness between flows etc. To improve the performance of end-to-end congestion control as well as provide low delay, low loss and fair service in best-effort IP networks, IETF suggested using Active Queue Management (AQM) in routers.Many AQM algorithms have been proposed in resent years. Despite the research efforts spanning several years, there are no universally acceptable solutions in this area.Some problems exist, such as the queue length oscillations in large delay networks, the effect of router buffer size on queue length-based AQM schemes, systematic parameter settings, performance evaluation and the stability of TCP/AQM system etc. We proposed a series of new methods in this dissertation to study the problems mentioned above.The dissertation presents guidelines for setting parameters of PI controller using"second-order optimization model"instead of margin of stability. The new scheme shows not only good steady-state properties but also good transient behaviors. The dissertation also proposes a multi-model fuzzy controller as active queue manager for internet routers since fuzzy logic controller does not need precise control model and has the advantage of insensitivity to control parameters. The new algorithm benefits the merits from both fuzzy controller and PI controller.The packet buffers of router are an important component of computer networks. We find the performance of AQM schemes based on queue length relying on the router buffer size. We analyze the reason using control theory and recognized saturation compensator as a potential solution. Using PI as an example, we propose a new schemecalled SCPI (PI controller with Saturation Compensator). The performance of the saturation compensator is evaluated by NS simulations. The results under a variety of network and traffic situations indicate that the proposed scheme exhibits superiorbehavior.Most of the AQM schemes have difficulty in obtaining desirable performance in large delay networks. It is difficult to restrict the negative impact of large delay becausethe network delay is very changeful. The actuating signal produced by AQM controller also reflects the information of flows average delay that traverse the router. We presenta mechanism estimating flow number and average delay in AQM router in Internet. This estimate is obtained without collecting or analyzing state information on individual flows. Then we use the estimating results and Ziegler-Nichols Method to restrict the negative impact of large delay. The performance of Ziegler-Nichols delay compensation AQM algorithm is evaluated through NS simulations. The results show that the delay estimation and compensation improve the performance of AQM scheme.Systematic parameter setting is a difficult question in AQM research. We present a generic framework which encompasses RED, P, PI, PIP, PD, SMVS, REM and AVQ as special cases by using Single neuron-based PID control. In addition, the new approach expands the current AQM controllers by providing an adaptive mechanism. Based on this framework and the adaptive mechanism, we develop adaptive PI and adaptive PIPcontroller. Simulation studies under a variety of network and traffic situations indicate that the proposed schemes exhibit more robust behavior than the prior schemes.One important area is the stability of the TCP/AQM system. Developing a new AQM mechanism or improving an existing AQM mechanism should not affect the robustness of the congestion control algorithms. Empirical studies demonstrate the efficacy of AQM, but it brings the complicacy of computer networks at the same time. We concerned ourselves with a network-wide stability analysis of AQM networks,focusing on the relationship between AQM stability and network environments.The dissertation studies thoroughly the problems of designing efficient AQMalgorithms. It proposes router buffer saturation compensator, delay estimation and compensation method, an adaptive framework and stability analysis. The results of our study show that these algorithms can be used in router design of next generation networks.