The entropy of an arrivals process: a tool for estimating QoS parameters of ATM traffic

Classical queueing theory is applicable to telephone traffic at the call level: incoming calls to an exchange may be assumed to be independent so that the input stream can be modelled by a Poisson process; consequently, classical theory concerns itself with queues with Poissonian inputs. In the buffers of ATM multiplexers are queues of data packets waiting to be processed and the statistical characteristics of their input streams are very different. Since data packets are transmitted in bursts, their arrivals are strongly correlated and cannot be modelled well by a Poisson process. The authors give a brief exposition of their strategy and describe some of the algorithms devised to implement it.<
>