Techniques for Adaptive Estimation of E ective Bandwidth in ATM Networks

This paper examines the e ectiveness of adaptive algorithms for estimation of e ective bandwidth in ATM networks. Large deviations results for constant capacity single server queues are used to derive a basic e ective bandwidth result which is applied to a linear representation of the input process giving the e ective bandwidth as a function of the cumulant generating function of the innovations, QoS, mean arrival rate and bu er size. The algorithm is tested by simulation using the least squares lattice algorithm for estimating the linear process coe cients for a range of input processes. The e ectiveness of blocking and truncation of the input process on the algorithm is investigated.