| Over the years the Internet has become a critical part of modern world. Prudent congestion control mechanisms have been primarily responsible for the stability and robustness of the Internet. However, these lessons about the necessity of congestion control mechanisms were learned after a series of failures in the formative years of the Internet. As a result of these failures, considerable research efforts have been spent at understanding congestion control and many solutions to avoid and control congestion in the Internet have been proposed. Of the proposed solutions, end-system based congestion avoidance and control is now an integral part of the most widely used transport protocol; TCP. However, subsequent research which provided complementary network based solutions for managing congestion are yet to be deployed on the Internet, for a variety of reasons.; In this thesis we re-evaluate the function placement of the building blocks of the Internet congestion control architecture. Specifically, we attempt to bridge the gap between the deployment costs and desirable congestion control features. Towards this end, this thesis proposes a series of deployable end-and-edge based solutions which combine almost all the beneficial properties of existing congestion control solutions. An essential step for achieving such deployable solution lies in decoupling the congestion control tasks from their placement in the network. As such, this thesis proposes end-and-edge based architectures which help its de-construct AQM schemes, thus disassociating congestion control tasks from their placement.; In this thesis we proposes an end-system based solution, Randomized TCP, which can alleviate the network and end-to-end performance degradations which arise out of use of TCP on simple FIFO queueing. Uncooperative Congestion Control is a proposed network-based architecture which de-couples management of selfish behavior of flows in the Internet from router based AQM policies. Specifically, this framework helps prevent traffic volume based denial of service, enforces congestion response conformance and provides an architecture for implementing simple differentiated services. Finally, this thesis also proposes virtual AQM, an end-or-edge based solution for managing bottleneck queues in the Internet. The virtual AQM framework can help us send early congestion indications and also reduce steady state queue size and latency. |
