Estimation of traffic parameters in high-speed data networks

Fractional Brownian Motion (FBM) is a compact mathematical model for the burstiness and long-range correlation observed in the aggregate traffic of high-speed data networks. The model is described by the mean rate m, the 'peakedness' a, and the Hurst parameter H. For practical applicability, these parameters have to be estimated from traffic measurements and applied in engineering algorithms to ensure that adequate network capacity is provided. We examine the accuracy of parameter estimation by the variance-time method and the wavelet method, for different frequencies of measurements. For capacity engineering, the two methods offer about the same accuracy (with a slight edge for the wavelet method). For traffic with 0.5 less than or equal to H less than or equal to 0.8, measurements at 1-minute intervals provide adequate accuracy for engineering, while intervals of the order of 100 milliseconds may be needed for H > 0.85. For either method of parameter estimation, standard switch measurements at 15-minute intervals are inadequate for FBM traffic models, and could result in provisioning errors of about 10%.