Research Of Predictive Functional Control For Active Queue Management

With the development of the Internet and more applications, networkcongestion becomes a very important issue, especialy for the active queuemanagement (AQM) method. In the paper, a new active queue managementmethod in dynamic TCP networks was proposed—PFC/AQM (Predictivefunctional control/Active Queue Management) controller. It supports explicitcongestion notification. The ability of the controller in handling system delayalong with its simplicity and low computational load makes PFC a privilegedAQM method in the high speed networks. In the control of dynamically-varyingcomputer networks，Fast response, low queue fluctuations,high link utilizationare important and desired property. The paper mainly aim at designing an AQMmethod, which performs well in most of the features such as responsesensitively, queue management, link utilization.Firstly, introduces some concepts and principles of network congestion andthe active queue management, as well as predictive control. Then analyzes thereasons of network congestion, and researches a few classic algorithms.Secondly, according to the dynamics of TCP, improve the model ofTCP/AQM networks, establishes a small signal linearized fluid-flow model ofthe TCP/AQM networks. Furtherly,in order to resolve the problems of highcomputational load and large queue fluctuations of the existed algorithms,combinating with predictive control theory, a new active queue managementmethod—PFC/AQM is proposed.Thirdly, analyzes the performance of the system, by using its closed-looptransfer function.And furtherly we research the effect of factors such as RTT,bottle-neck capacity on the system. When the system parameters are varyingmost often, its performance will be degraded. Therefore, we extend thecontroller to an adaptive predictive function controller—APFC. Finally, simulates the PFC/AQM controller on the NS2simulation platform,and analysis of experimental to prove the validity of the developed controller. Ithas a great improvement in response, queue length management and linkutilization.