A simple approximation of multimedia traffic with long term correlation

A simple yet accurate model is needed to approximate correlated multimedia traffic sources such as that seen on typical file server networks. We consider a superposition model based on HI-LO Markov chains and propose a novel model parameterization which uses the normalized DC component of the power spectrum of the traffic. Using this model as input to an ATM queue, we apply a mean value analytic approach to find exact and upper bound results for the mean queue size. This approach provides a simpler, more intuitive development for mean queue size than previous results, and is extendable to other birth-death arrival processes. It is shown that, for this general model of a correlated arrival process, the normalized DC power component acts as a scaling factor on previous results for the mean size of the same queue with uncorrelated arrivals. This characteristic is illustrated for Kingman´s heavy traffic approximation for a G/G/1 queue, the mean queue size for an M/D/1 queue, and the Kingman-Brunelle upper bound for the G/M/m queue. Finally, using traffic measurements from a file server network, we provide heuristics for approximating the normalized DC power component of the sampled data