Research On Several Algorithms Of Network Congestion Control Based On Control Theory

Due to the unbalance distributions of network resources and traffics, the occurrence of congestion is an intrinsic characteristic of Internet. With the rapid growth of network scale and the emerging of the variety of applications, network congestion problems is growing, which becomes the main factor to affect the network performance. So network congestion control algorithms have been the hotspots in the network research field.Congestion control algorithms can be broadly classified into two categories:source algorithm and link algorithm. Source algorithm, such as TCP, is executed in the end hosts or edge devices to adjust sending rate in response to feedback. Link algorithm, such as active queue management, is executed in network devices to detect congestion and send feedback information to the host. It is hard for network modeling and congestion control algorithm designing by the dynamic network environment and uncertainty of the network delay. Otherwise, with the development of the wired/wireless network and wireless multihop network, which introduce some new characteristic into the Internet, many limitations of the traditional congestion control mechanisms emerged. To address these problems, with the help of the control theory, some congestion control algorithm is proposed in order to improve the stability and robustness of the algorithms. Several congestion control algorithm is also designed in the thesis, which is suitable for the novel network structure. The main results obtained in this thesis are as follow:1. A class of active queue management algorithms is proposed to execute in the dynamic network environment based on H∞performance index. In the premise of ignoring the frequency characteristics of the network, a H∞PI controller is designed, whose coefficient can be determined by the linear matrix inequality. Considering the shortage of the H∞PI controller, an output feedback controller is designed by H∞control method of the linear time-delayed system. The robust stabilization of the proposed controller is proved, whose coefficient also can be determined by the linear matrix inequality. Finally, simulation results indicate a clear advantage in controlled network performance of the proposed controller, which also has high robust stabilization under diverse network conditions.2. In view of the uncertainty of the communication delay, a class of active queue management algorithms are proposed based on delay-varying. Considering the retransmission timeout mechanism in TCP, a delay-dependent H∞state feedback controller by observer is proposed via the small-signal linearization. Due to the local stability of the small-signal linearization, the nonlinear network model is set up by the linear function between the RTT and queue length. A nonlinear output feedback controller is designed via backstepping technique, the value range of the parameters are also discussed. A nonlinear TCP window size observer is proposed. It can be proved that the observer state converges to the real sate asymptotically, when packet dropping or marking ratio falls between 0 and 1. Ns simulation results indicate that the proposed algorithms maintain better network performance.3. Considering the shortage of the conventional TCP algorithms in wired/wireless network, a fuzzy selective rate congestion control algorithm is presented, which is based on the fuzzy logic control theory. Measuring the queue length and the data rate in the router, the proposed congestion control algorithm combines the strengths of AIAD and AIMD, acting conservatively on wireless loss and aggressively on congestion. Simulation results indicate that our algorithm performs well in heterogeneous wired/wireless links, adapting effectively to the dynamics of the network.4. As the performance of TCP is still poor in some way, such as queue delay, a TCP window controller is presented based on Kalman filter. A TCP windows controller is proposed by the proportional control theory. Via measuring the queue length and RTT of the network, the non-causal TCP traffic controller is approximated by the kalman filter theory. Ns simulation results indicate a clear advantage in TCP windows prediction of the proposed controller, which even provides high link throughput and low queue delay in a dynamic network condition.5. Due to the "hop-to-hop" characteristic of the wireless multihop network, the congestion control algorithm based on data rate is proposed by the distributed dynamic system theory and digraph theory. First of all, the congestion control algorithm based on Min-Rate is designed for wireless Ad hoc network, and it can be proved that the send rate for all nodes converges to the minimal available bandwidth by the proposed Min-Rate algorithm. Second, the congestion control algorithm based on leader is designed for wireless sensor network. The aggregation node is considered as the leader, and determines the available bandwidth for every TCP connection. Finally, the congestion control algorithm is proposed for the wireless mesh network, combining both the algorithms of the wireless Ad hoc network and wireless sensor network.