Abstract
This paper considers various Markov-modulated processes which have been used in the literature to model internet traffic. In particular, such processes have been used to replicate the long-range dependent nature of internet traffic and the justification for doing so is that long-range dependence (LRD) has important effects on delays and buffer occupancy in traffic models. However, in combination with an extremely simple queuing model, the traffic generation models have their parameters tuned to match real network traffic traces. The queuing performance is then measured both in terms of the probability of the buffer exceeding a given size and in terms of mean queue length (or mean delay).