Network Congestion Control, Active Queue Management Algorithm

With the development of computer Internet, The number of users and the network applications is increasing dramatically, which resulted in more and more serious network congestion. So the further development of the Internet is seriously hindered. Thus, the congestion control of the Internet is an important problem. Therefore, the network congestion control is the main way to improve the network performance and reform the service quality. The investigation of the network congestion control is important not only in the theory but also in the application.Active Queue Management (AQM) for the routers have been discussed widely. This thesis reviews the state of art in Active Queue Management (AQM), and two new AQM algorithms are proposed based on control theory. Simulation results are given to prove the performance of the new AQM algorithms. The contributions of this thesis are as follows:(1) A neural network based on the PID (PIDNN) is proposed as active queue management algorithm. The adjustment of neural network parameters is implemented by using gradient algorithm as learning the rules. So that the probability of packet loss can achieve adaptation. The simulations result shows the queue stability and drop ratio of the proposed PIDNN algorithm are better than PID algorithm.(2) On the basis of the existing AQM algorithms, we propose a novel active queue management scheme based on RED of the rate and loss ratio, called RLRED. The algorithm measures the total sending rate and the latest packet loss ratio, and uses them as a complement to queue length for adaptively adjusting the packet drop probability. The simulation results show that the algorithm proposed not only can maintain a fast response time, also can significantly improve the stability of the queue, reducing the queue delay, and other performance index don't deteriorate.(3) We discussed the fairness about active queue management and designed a new fairness index. Finally, four AQM algorithms are compared based on simulation experiments in the performance of fairness. The four algorithms are RED, PI, PIDNN and RLRED, respectively.