Equation-based congestion control for unicast and multicast data streams

We believe that the emergence of congestion control mechanisms for relatively-smooth congestion control for unicast and multicast traffic can play a key role in preventing the degradation of end-to-end congestion control in the public Internet, by providing a viable alternative for multimedia flows that would otherwise be tempted to avoid end-to-end congestion control altogether. The design of good congestion control mechanisms is a hard problem, even more so for multicast environments where scalability issues are much more of a concern than for unicast. In this dissertation, equation-based congestion control is presented as an alternative form of congestion control to the well-known TCP protocol. We focus on areas of equation-based congestion control which were not yet well understood and for which no adequate solutions existed. Starting from a unicast congestion control mechanism which in contrast to TCP provides smooth rate changes, we extend equation-based congestion control in several ways. We investigate how it can work together with applications which can only operate in a very limited region of available bandwidth and whose rate can thus not be adapted to the network conditions in the usual way. Such a congestion control mechanism can also complement conventional equation-based congestion control in regimes where available bandwidth is too low for further rate reduction. When extending unicast congestion control to multicast, it is of paramount importance to ensure that changes in the network conditions anywhere in the multicast tree are reported back to the sender as quickly as possible to allow the sender to adjust the rate accordingly. A scalable feedback mechanism that allows timely congestion feedback in the face of potentially very large receiver sets is one of the contributions of this dissertation. But also other components of a congestion control protocol, such as the rate increase/decrease policy or the slow-start mechanism, need to be adjusted to be able to use them in a multicast environment. Our resulting multicast congestion control protocol was implemented in a simulation environment for extensive protocol testing and turned into a library for the use in real-world applications. In addition, a simple video transmission tool was built for test purposes that uses this congestion control library.