Fractal Traffic Model for Internet Traffic Engineering

Abstract

Traffic models have been at the core of teletraffic research for the last several decades because they have traditionally served as fundamental tools for network performance analysis. With the recent discovery of fractal characteristics of Internet traffic, the need for fractal traffic models that play beyond this traditional role has grown significantly. This is attributed to the fact that despite tremendous interests being focused on a wide variety issues dealing with this fractal nature of Internet traffic, there still remain much new knowledge yet to be gained for its implications for Internet traffic engineering. This requires the fractal traffic models that can quantitatively link the model parameters to the fractal statistics of Internet traffic so that a parameterizable, physical structure can be associated with the Traffic under study, rather than providing mainly qualitative and abstract explanations. To meet this daunting challenge, a library of traffic models based on Fractal Point Processes (FPPs) are described and analyzed. Compared to other fractal Traffic models, the FPPs have several key advantages. First, they yield a simple and effective traffic parameterization method that determines the model parameters from the first- and second-order statistics. This result allows for quantitatively understanding how model parameters are related to and control the range of time scales over which fractal behavior is dominant. Second, they reveal how session-level fractal dynamics such as session arrivals, duration, and volume affect packet-level fractal dynamics in a quantitative manner. Moreover, they are versatile and thus are able to capture a broad range of different fractal behaviors. Additional advantage of these models includes low modeling complexity, yielding high computational efficiency for running large-scale simulations. Demonstration of these benefits are provided based on simulation and characterization of two Web traces.