Modeling and Analysis of Traffic in High Speed Networks
Soma S. Muppidi, Victor S. Frost · 1997
Recent traffic studies of high-resolution, high quality traffic measurements have revealed the phenomenon of long-range dependence in network traffic. The implication of the presence of long-range dependence in traffic is that actual network traffic exhibits more bursty behavior compared to traditional ”Poisson-like” models. Traffic modeling and performance prediction in networks with long-range dependent flows is required for the design of efficient congestion control, routing and other network management algorithms. Here a non-Markovian phase process is used to model the traffic process. The phase process captures the macro-dynamic properties of the traffic. The traffic dynamics within each phase i.e., the traffic micro-dynamics, are then described by a random process with finite mean and variance. Network performance using this model has been evaluated in the form of delay vs load curves and cell loss ratio vs buffer size. By capturing the burstiness of the network traffic in the form of traffic macro-dynamics, the model developed here predicts ATM queue performance in network flows that may be inherently long-range dependent in nature. Extensive simulations based on traces of traffic collected from a wide area ATM network has been used to validate the developed model and performance analysis methodology. Also the effects of traffic microdynamics on performance has been investigated within the frame work of the developed model. It has been observed that the mean delay is not sensitive to the specific nature of the micro-dynamics while the cell loss is affected by the traffic micro-dynamics.