We use the theory of effective bandwidth for bandwidth allocation in high speed digital networks (ATM), and experiment with simulated Markov traces and actual VBR traces. To approximate the effective bandwidth, we use the four traffic descriptors: average rate, asymptotic variance, peak rate and average burst duration. In our experiments, we find the theory yields good approximations of the queue length distributions for simulated Markov traces and a video conference trace. However, our estimations for asymptotic variance do not converge for some VBR video traces. This implies that some VBR traces might be long-range dependent. Based for the fractional Brownian motion (FBM) model, we revise the theory of effective bandwidth and modify the associated traffic descriptors for such VBR traces. Our experiments show that the FBM model is good in heavy traffic, but does not fit well in light traffic.
展开▼