Intelligent Shaping: Well Shaped throughout the entire Network?

In this article we present a new class of traffic shapers that make up the shaping decision whether to delay or pass a packet by comparing a measured distribution specifying the arrival stream to a given reference distribution. Rather than simply comparing average values this shaper emits data streams with property ’better than reference distribution’ by employing a kind of nonparametric test. If all incoming data streams of the network considered are shaped according to ’better than Poisson’, network dimensioning is simplified in the sense that multiplexer performance can be predicted conservatively by assuming Poisson arrivals, hence, admission control and resource dimensioning can be based on a simple M/D/1 queueing model. The presented shaping algorithm thereby provides an analytical tool for network providers to calculate upper bounds on the load in a network fed with accordingly shaped streams only. Our simulations show that the shaper presented emits data streams with corresponding distribution function almost identical to (or better than) reference distribution independent of the input arrival process but on the other hand introduces no serious delays if the arriving stream is already ’better than reference’. Moreover the shaper is mostly robust in terms of shaping parameters and initial conditions and has the capability to remove correlations of the passing stream.