Research Network Congestion Control Method Based On Optimization Algorithms

As the network population is getting more and more, it's easy to cause resource shortage, so network congestion happens and the performance gets worse. For these reasons, it is important to avoid congestion, and improve its performance at the same time.The scholars in different fields have done many research and improvement of the congestion control strategies based on TCP windows continuously. The researches show that those strategies can't control congestion congestion effectively. So the congestion control strategies on the network middle node have arised at the historic moment. After continuous developing, the most effective one is active queue management algorithm(AQM). Furthermore, the AQM algorithms based on various kinds of optimization algorithms show some obvious advantages, such as queue stability, robustness to model uncertainty and disturbance outside. This paper proposes several AQM algorithms based on optimization theory:(1) In order to make the conventional AQM algorithms adaptive to the change of network parameters, a optimized algorithm based on the self-organizing feature mapping network is proposed. This algorithm can optimize the controller's parameters online when the network parameters change. The NS-2 simulation results show that the algorithm is efficient to guarantee the stability and robust stability to network dynamic change.(2) Because of the state delay and input delay in the network, a controller with delay compensating for time delay system is necessary. The parameters of the optimization controller depend on optimization goals set by users. They can choose high utilization, low jitter, high throughput and so on. We transform the target problems into convex optimization problems to make it convenient and easy to be realized using LMI.(3) Aim at the discreteness of the network events, a discrete AQM controller is designed. The network system model consists of state delay and control signals delay as well as uncertainty. The sufficient and necessary conditions for the system's stability is presented with the strict theoretical analysis. Simulation results show that this algorithm can maintain good robustness and stability when network parameters change, and also shows good performance in many bottleneck links.