End-to-End Internet Congestion Control

With the rapid development of network users and all kinds of applications, Internet traffic is becoming extrodinary busier nowadays. Congestion phenomenan occurs easily due to the restriction of Internet resource bottlenecks and the bursty nature of Internet traffic. The current Internet is based on connectless IP architecture, which only provides best-effort service and cannot guarantee the Qualityof Services (QoS). Thus high layer protocols such as TCP are utilized to guarantee the correct transmission of data packets. Especially because TCP flows occupy nearly 90 percent traffic of current Internet, TCP congestion control plays a very important role in maintaining dynamic performance of Internet as well as its stability, robustness.The backgrounds and causes of Internet congestion phenomenon are introduced in the beginning of this thesis. Chapter one decribes recent research progress in the area of Internet congestion control and analyzes several problems existing in this research topic. This thesis also describes and compares some significant TCP congestion control algorisms and Active Queue Management (AQM) strategies in detail, and introdeces some novel research methods and results in the stability analysis of Internet congestion control with the perspective of classic control system, modern control system and nonlinear instability phenomena. On the basis of previous research works, the study in this thesis is listed as follows:1. Adopts network simulator 2 to simulate the performance of various TCP versions considering both single source single link and multiple sources single link. Aiming at the instability phenomenon from large transmission delay and bandwidth, a state feedback TCP Vegas algorithm is proposed in order to achieve better stability, which consults the AQM idea in the network router. An improved Round Trip Time (RTT) calculation in TCP Vegas is presented to overcome the influence of short-time disturbing traffic and CBR flows.2. Presents a novel fuzzy logic method for congestion control in TCP networks. States of queue are transformed into linguistic values and packet drop probability is derived from predefined fuzzy inference engine. Simulation results manifest its effectiveness.3. Analyzes the dynamic characters of current Internet transmission control protocol (TCP) with RED gateway by concentrating on single link network topology. Explains that the RED algorism is susceptible to the variation of parameters using method of root locus on the foundation of continuous congestion control model based on previous research. Investigates nonlinear behaviors of TCP with RED gateways such as bifurcation, chaos considering both linear and nonlinear drop functions. Analyzes and deduces the stability conditions of adaptive RED and proportional derivative RED algorisms controlling bifurcations and chaos. Simulation results explicitly compare outcomes of above two algorisms.4. Introduces an appropriate Lyapunov function aiming at the time-dependent model of deterministic continuous flow control system. Global stability conditions of flow control system are brought forward without time delay and with bounded time varying delays, which solves the difficulty of parameter configuration of flow controller in the network routers very well. The stability results are also extended to a general network topology in the presence of many non-elastic flows.