Performance evaluation — Test configurations and results

Abstract

In this paper experimental designs for investigating performance parameters and their corresponding experimental results on the EXPLOIT testbed facilities are given. In an exemplary way the performance evaluation of a switch and an ATM network is described. These designs and results can be used in two ways: firstly in determining the network performance of already existing equipment and networks and secondly in designing new equipment by modelling the described scenarios within simulations. As important network performance parameters under study have been identified cell delay, cell loss and cell delay variation. These will be investigated together with some techniques for assessing those. Initially the question, whether the occurring delays in successive switches are independent, is tackled. If the independence assumption holds true or the correlation can be quantified, then out of measurements at a single switch the delay behaviour after having passed a number of switches can be determined, thus extrapolation from a small network to a larger one becomes possible. Afterwards the fairness of the switch design with respect to multiplexing traffic sources while being in an overload state are discussed. For being able to predict delays in the future the suitability of “traditional” queueing models will be checked. Out of this, forecasts on cell losses can be gained as well. For this investigation specific traffic mixes have been chosen, which are representative of the predicted traffic for future ATM networks. Also some results on the end-to-end cell delay variation are given.