Cross-layer Based Study On Congestion Control In Networks

With the explosively increasing requirement of information, people are hunting for networks combining real traffic, e.g., VoIP, online video, with data traffic, e.g., Web browsing, files transferring, which claim high speed, high quality and high reliability on information exchange and files transport. Therefore, it's important and challenging to study congestion and congestion control in networks, which is also an attractive research topic in recent two decades. In this thesis, several congestion control schemes are designed with regards to different types of networks.First, a new robust AQM controller with a simplified parameter regulation method is designed based on H∞optimal theory. Taking the closed-loop performance and robustness into consideration, the uniform expression of the controller is developed through theoretical analysis. With these, analysis on stability and robustness of the controller is conducted. Further, a classical PID type controller, named R-PID, is taken as an example to show the controller's robustness and parameter regulation. A traffic monitor is designed through heuristic and empirical approach. Further, a cross-layer-based adaptive parameters regulation method is implemented at router, making a trade-off between the performance and robustness accordingly. Simulation results show that R-PID has the significant robustness and outperforms the traditional AQM controllers on performance of queue length keeping, queuing delay and bottleneck link utilization. Moreover, through the adaptive parameter adjustment, the controller can make a trade-off between the robustness and performance with regard to network dynamic.Second, a fair resource allocation model, FairRA, with congestion control scheme is presented for WLAN accessed network. The Max-min fairness principle is extended to cater for the requirements of fairly scheduling at base station in WLAN, and then a queue-length-based scheduling algorithm is developed based on these extended principles. The link utilization of WLAN is calculated with the MAC layer information provided by cross-layer method, and then a congestion control scheme is designed based on available WLAN capacity. Integrating scheduling methods and congestion control scheme, the resource allocation model can provide fairness among the competing users in a great extent. Moreover, the model is a distributed system as it balances functions of congestion control among end points and base station. Simulation results show that FairRA provides a smooth sending rate, goodput, and a considerable fairness as well.Finally, focusing on performance enhancement of CCID-3 in DCCP, the thesis proposes two improvements on the equation of sending rate computation of CCID-3. One is to substitut the RTT estimated in transport layer with theoretical one, which is analyzed with IEEE 802.11 DCF protocol under the assumption that there is no retransmission at MAC layer. The other is modification on the packet loss rate computation, which can speculate the loss event from the ACK sequence number at sender by considering the ACK loss as a loss event and assigning the same weight as the packet loss. The weights of loss events are also modified to fit for the new computation method for the loss rate. The simulation results show that the modified CCID-3 has great improvement on throughput.