Fast bounds for ATM quality of service parameters

We demonstrate how fast estimates of quality of service parameters can be obtained for models of VBR traffic queuing in the buffer of an ATM multiplexer. The technique rests on conservative estimates of cell-loss ratios in the buffer. These are obtained through Martingale methods; the cell-loss ratio from a buffer of size x is bounded as Prob[cell-loss]⩽e^-u-δx. An estimate of the mean cell delay can also be derived from this. We discuss briefly the algorithms used to compute the bounds. One attraction of the method is that the speed of computation is independent of the number of sources present in the traffic stream arriving at the multiplexer. This gives great advantage over estimates derived from a complete solution of the model queueing problem: these generally require the analysis of matrices whose dimension is proportional to the number of sources present. We give an application using a model of bursty traffic proposed by Finnish Telecom: a Markovian traffic model in which burst and silence are geometrically distributed subject to a fixed maximum burst length. A demonstration of a prototype software package based on the above methods is included