Characterization of bursty sources and capacity estimation in high speed networks

We consider the problem of characterizing bursty sources and estimating the aggregate capacity they require. The sources are described by a mean rate, peak rate, and burst length, in a network with both loss and delay. (Only loss is considered here). We begin by presenting an efficient technique for computing loss exactly in situations where the number and diversity of sources is reasonably large. We then present computational experience using this technique, and on the basis of this experience, develop an approach for characterizing the behavior of a realistically large number of I heterogeneous bursty sources on a channel. We use this characterization to explore procedures for estimating the capacity required to achieve desired performance. The techniques are efficient and can be used in the inner loop of a network design procedure and for real time network management.