Congestion pricing as scalable, efficient and stable congestion control for future IP networks

This dissertation focuses on the design of distributed congestion control algorithms for TCP/IP networks that are more powerful by utilizing the theory of Congestion Pricing as a mathematical framework. Currently implemented congestion control algorithms have several drawbacks that lead to sub-optimal usage and unfair distribution of network resources. Further, new applications have significantly changed the demand for network performance and quality of service. As will be shown, the use of wireless links and large link capacities can cause instability of the algorithms in use today. Congestion Pricing is a strategy based on economics and optimization theory: A congestion measure (shadow price) is computed at each network node and fed back to the source. The sources adapt their rates according to utility functions and aggregate pricing information. It can be shown that this will lead to a social optimum for the entire network while maintaining both low queue sizes and high utilization. By choosing the user’s utility function, different classes of service can be implemented without additional network support. Thus, the increasing demands on the network can be met without changing the Internet’s fundamental principle of keeping the network nodes simple, and thus retaining its flexibility and scalability. At the same time, efficiency will also be significantly improved. In this dissertation, Congestion Pricing will be applied to the Transmission Control Protocol (TCP). First, an implementation is developed that makes use of the full pricing information. Then, in an effort to make the new TCP source compatible with the existing network and TCP receivers, the pricing information is reduced to a single bit. This reduction of information introduces new challenges that are addressed by a “Single Bit Resource Marking (SBRM)” proposal developed by the author of this dissertation. Its performance is evaluated by comparison with other proposals and current congestion control algorithms. While the efficiency and scalability problems are solved by the application of Congestion Pricing, a control theoretic model is further developed to examine the linear stability of the proposed algorithms. Current congestion control algorithms can become unstable in realistic scenarios, which significantly harms network performance. SBRM, in contrast, is stable over a wider range of network scenarios, and the impact on performance parameters is lower in cases of instability. Lastly, multimedia applications are also addressed in this dissertation. They usually cannot change their transmission rate. Therefore, a distributed Call Admission Control using Congestion Pricing is developed. Even without special network support, it will be effective and highly efficient. Since the Call Admission Control works in a distributed manner to make it scalable, it can be implemented in network border gateways or in the sources themselves.