Network Friendly Congestion Control: Framework, Protocol Design and Evaluation

The increase in bandwidth, the number of internet users and the variety of internet applications, all point towards the need for adaptive and flexible internet protocols. Alternatively, they also emphasize on the necessity for specialized protocols that cater to application requirements. Current Transport protocols such as TCP and UDP are not flexible enough to be used by applications which differ in their performance expectations. UDP, for example, is inelastic in nature and therefore can cause exacerbation of congestion in an already congested network. TCP on the other hand was designed to provide proportional fairness, such that a ``flow'' is the unit which is allocated resources. This not only results in all flows having an equal share at the bandwidth irrespective of the application requirements, but also give way to a new trend of applications that are able to take advantage of the ``flow-oriented'' fairness. Delay-insensitive applications, such as P2P file sharing, generate substantial amounts of traffic and compete with other applications on an equal footing while using TCP. Further, to optimize throughput, such applications open multiple connections. This results in an unfair and potentially poor service for applications having stringent performance objectives. The main part of this thesis proposes NF-TCP, a TCP variant for P2P and similar background delay-insensitive applications. NF-TCP aims to be submissive to delay-sensitive applications under congestion. It is designed to be network friendly based on a fluid flow model for intermediate queues and uses explicit congestion notification (ECN) for early detection of congestion. Moreover, it exploits the measure of the available bandwidth to be able to aggressively utilize spare capacity. It can be observed that the traditional congestion control protocols comprise of tightly coupled mechanisms, and are designed to address a specific problem or scenario. They either lack the flexibility in meeting varying requirements or utilizing different types of available network support. This thesis proposes a congestion control framework that consists of pluggable components and utilizes it to guide the NF-TCP design. It identifies congestion-detection, flow-control and bandwidth-estimation as the main components of the congestion control protocol, which can be loosely or strictly decoupled from each other under a modular framework. This thesis implemented NF-TCP on Linux and ns-2. The evaluations of the NF-TCP Linux implementation on ns-2 show that NF-TCP outperforms other network friendly approaches (e.g., LEDBAT, TCP-LP and RAPID). NF-TCP achieves high utilization, fair bandwidth allocation among NF-TCP flows and maintains a small average queue. The evaluations further demonstrate that with NF-TCP, the available bandwidth can be efficiently utilized for supporting both delay-sensitive and insensitive traffic in a wide range of scenarios.