Fuzzy-AR Modeling for Dynamic Effective Bandwidth Estimation in High-Speed Networks

In this paper a fuzzy autoregressive (AR) model described in [1] is used to model and predict highspeed network traffic. This model approximates a complex nonlinear time-variant process by combining linear local autoregressive processes using a fuzzy clustering algorithm. We propose a method to estimate the traffic effective bandwidth at regular intervals, assuming the network traffic can be described by AR models obtained with the fuzzy modeling. For this aim, the linear coefficients of the fuzzy-AR modeling are used assuming network traffic is stationary at short time intervals. Using this traffic modeling, an effective bandwidth based rate allocation method is suggested to provide efficient traffic management. Finally, simulations with real traffic traces demonstrate the validity and performance of the prediction and the effective bandwidth based rate allocation method. Key-Words: Fuzzy modeling, prediction, network traffic, rate allocation, byte losses.