Study On Highly Dynamic And Heterogeneous Internet Congestion Control Algorithm

Active Queue Management (AQM) and end-to-end TCP congestion control for wired and wireless heterogeneous networks are two most interesting topics in the communication society. The dissertation starts by presenting a main/secondary merit framework for designing AQM controller, which is named YELLOW algorithm. The design rules for selecting its parameters are provided by coordinating the globally stable analysis with simulation tests. The main merit of YELLOW scheme is load factor. YELLOW scheme introduces a nonlinear queue control function as a secondary merit, which can be interpreted as an additional inner loop in control literatures. The YELLOW scheme is easy to be implemented by the design process. The performance comparison between YELLOW and other well-known AQM schemes is conducted under a wide range of traffic conditions. YELLOW algorithm exhibits a graceful performance under the highly dynamic network environment. In the following chapter, the dissertation proposes an adaptive parameter setting method for Adaptive Virtual Queue (AVQ) algorithm. The sensitivity problem for the desired utilization parameter and damping parameter can be eliminated using our adaptive method. Experiments also show that the revised scheme achieves good responsiveness and relatively high utilization compared with original AVQ algorithm.In the second part, the dissertation examines the question of stability analysis for congestion controller with time-varying delays. This can be viewed as the decentralized feedback system with time-varying delays. The dissertation treats the stability of such congestion control system with the developed Lyapunov-Krasovskii stability analysis method and the linear matrix inequality (LMI) technology. Both the local stability condition for Random Exponential Marking (REM) algorithm and global stability criterion for rate-based congestion controller are presented. Using convex optimization technical in LMI toolbox, the explicit upper-bound for time-varying delays with the robust stability condition can be derived. The relationship between network parameters and controller parameters can be solved numerically. Moreover, the dissertation presents a stability analysis for the TCP/LRED (Loss ratio based Random Early Detection)delayed feedback system by the classical Nyquist stability criterion. The stability condition reveal the effectiveness of LRED scheme, i.e., the stability condition can be decoupled from the TCP load N. To scale for the link capacity, the Dynamic Equilibrium Controller (DEC) algorithm is proposed. DEC scheme also gives a full average loss filter algorithm. The responsiveness and steady performance of the equilibrium packet loss ratio estimate can be improved by our proposed filter scheme.Finally, the dissertation studies the bandwidth estimate scheme for wired and wireless heterogeneous networks. In the first section presents a thoroughly analytic evaluation for bandwidth estimate scheme, such as the throughput formula, fairness and effectiveness. Even though its fairness and effectiveness can be proved, the bandwidth estimate scheme will have poor performance when the sender estimates the incorrect bandwidth or the random packet loss rate exceeds a few percent. To overcome the limitations, the next section presents TCP ComFF (Combination Fairness and Friendliness), which is a bandwidth estimate algorithm with the combination of loss differentiation algorithm and loss event interval measurement. The detailed implementation of TCP ComFF protocol is given by pseudocode and flowchart. By conducting extensive simulations, the dissertation compare its performance with TCP Westwood and Westwood+ and valid its inter-protocol fairness and intra-protocol friendliness.