Bypassing modelling: an investigation of entropy as a traffic descriptor in the Fairisle ATM network

The thermodynamic entropy of an ATM traffic stream is a useful tool for predicting cell-loss ratios and cell-delay. It was proposed that the entropy of a cell stream be estimated directly at a multiplexer; from this data, predictions can be made rapidly using algorithms which are simple enough to be executed on-the-fly. Preliminary investigations using simulated data have suggested that the method is practicable. We describe the results of an investigation of the method using real ATM traffic on the Fairisle ATM network. Fairisle is an experimental ATM LAN whose hardware design permits a high degree of experimental flexibility, including high resolution clocks for timestamping and measurement of traffic, a general purpose CPU and memory on every switch port. Fairisle is equipped with a wide range of ATM traffic sources, including compressed video, audio and LAN data. In addition traffic can be generated from a pre-recorded trace of the traffic activity. One such source is the Star Wars movie, reconstructed from a trace of the output of a DCT-based VBR video codec at Bellcore. We describe experiments involving the estimation of the entropy of both single-source and multiplexed cell streams based on the measured cell inter-arrival times. These estimates are used to calculate the decay-rate of the queue-length distribution. For multiplexed cell-streams, the decay-rate can be measured rapidly.