The Entropy of Cell Streams as a Traffic Descriptor in ATM Networks

Abstract

We examine the properties of a promising new traffic descriptor for ATM networks, namely the entropy of cell streams. The entropy is a measure of the disorganization among cells within a traffic stream; alternatively we can say that entropy captures the amount of randomness in cell scattering. We study the entropy of ON-OFF sources with respect to the typical queue parameters of interest: average queue size, queue variance and equivalent buffers. The equivalent buffer is defined as the minimum buffer size needed to achieve a specific loss probability. We demonstrate that the average queue size and the variance of queue size are monotonically decreasing with increasing entropy in streams with the same fixed load. We find that the inverse of the entropy is closely linear, to within a good approximation, to the equivalent buffer. This simple relation demonstrates the appeal of the entropy estimator. In addition to a measure of cell scattering, our results suggest another interpretation of entropy as a measure of smoothness. Traffic streams with higher entropy (i.e. smoother) have less buffering needs in terms of average, variance and equivalent buffers. Based on this observation we introduce a traffic shaping mechanism whose goal is to boost the entropy of a stream.