Internet congestion control algorithms : a stochastic approach

Consider an Internet traffic source sending packets into a single link connected to (an)other source(s). The source's "window" is the maximum number of packets discharged without waiting for acknowledgement, at any point in time. Such a window concept is the cornerstone of a number of congestion control algorithms, such as Tahoe, Reno, NewReno, Vegas and Sierra (partly rate-based). All these are briefly treated. We then develop quantitative models for the performance (throughput) levels of all these algorithms. It is shown that Sierra is by far, analytically, the most superior among all these algorithms. A goal of this chapter is to introduce the neophyte in congestion management to the discipline and then be in a position to contribute meaningfully to the literature. As such, all concepts introduced are motivated and developed from the ground up.