| Multimedia applications have become increasingly popular in the Internet. TCP is the dominant Internet congestion control protocol, but it does not serve all applications well. Thus, many new congestion control protocols have been proposed recently, in particular for multimedia applications.; To ensure that these flows share bandwidth fairly with TCP flows, TCP-friendliness is proposed as a criterion for designing new protocols. Currently, the TCP-friendliness criterion is defined based on the assumption that all flows experience the same static congestion signal. However, the bandwidth sharing and congestion signal is a result of the dynamic behavior of all participating flows. The claim of this thesis is that the bandwidth sharing behavior among competing flows should be studied in a dynamical environment.; To understand a dynamic phenomenon one needs a theoretical model that adequately describes the behavior of the system being studied. In this dissertation, we propose a state-space model to study the dynamics of the bandwidth competition, in particular among AIMD-based TCP-friendly flows. It characterizes a dynamic system by a set of related state variables, which can change with time in a manner that is predictable provided that the external influences acting on the system are known. We use the model to describe the stability of bandwidth competitions, which is characterized as convergence to a dynamically oscillating limit cycle in the state space. This stability description clearly distinguishes transient and long-term fairness.; Along with the state-space modeling, we build an adaptive AIMD-based congestion control protocol that exposes its parameters to applications. This dissertation presents some example uses of this adaptive protocol to verify the results derived from the state-space model. As an example, we adjust the AIMD parameters to achieve a uniform fairness that is independent of round-trip-times. |
