The Internet traffic data have been found to possess extreme variability and bursty structures in a wide range of time-scales, so that there is no definite duration of busy or silent periods. But there is a self-similarity for which it is possible to characterize the data. The self-similar nature was first proposed by Leland et a1 [l] and subsequently established by others in a flood of research works on the subject [2]-[5]. It was then a new concept against the long believed idea of Poisson traffic. The traditional Poison model, a short ranged process, assumed the variation of data flow to be finite but the observations on Internet traffic proved otherwise. It is this large variance that leads to the self-similar nature of the data almost at all scales of resolution. Such a feature is always associated with a fractal structure of the data. The fractal characteristics can exist both in temporal and spatial scales. This was indicated by Willinger and Paxson [6], as due to the extreme variability and long range dependence in the process. Presently, one of the main research interests in the field of Internet traffic is that of prediction of data which will help a network manager to render a satisfactory quality of service. Before preparing a model of prediction, one of the important tasks is to determine its statistics. Any model to predict the future values will have to preserve these characteristics.
On the Self-Similar Nature of ATM Network Traffic