Characterization of ATM On-Off Traffic from Cell Traffic Measurements

The optimal design of broadband switching systems depends on an understanding of the statistical nature of the broadband traffic. An important practical problem is characterization of the network traffic by fitting stochastic models to actual traffic measurements. This paper studies the problem of characterization of ATM cell traffic by maximum likelihood estimation of the commonly used Markov on-off model. The maximum likelihood estimators (MLEs) and their probability distribution function are derived. The MLEs are shown to be unbiased, consistent, and asymptotically normal. Maximum likelihood estimation is compared numerically with the method of moment matching for a simulated on-off packet trace, which indicate that the MLEs result in smaller variance and better accuracy than moment matching.