An efficiency prediction method for ATM multiplexers

This paper describes a methodology for conservatively predicting the traffic carrying efficiency of an ATM multiplexer that is operating in compliance with realistic performance objectives. Our approach uses a, multi-phased treatment of this important problem. In the first phase, a synthetic model of bursty traffic sources is used to construct a set of loading curves that predict worst-case statistical multiplexing efficiencies as a function of certain parameters which characterize both the traffic sources and the multiplexing system under consideration. In the second phase, a set of real traffic sources is matched with these same parameters through appropriate traffic measurements, and a prediction of achievable statistical multiplexing efficiency is obtained by the appropriate use of these loading curves. This technique has been successfully tested on real traffic source data obtained from Local Area Networks, and which has been shown to possess the "self-similar" temporal characteristic that is known to present significant challenges for statistical multiplexing. This methodology was developed with the aid of a high-performance, parallel simulator that is used both for the construction of representative loading curves and for demonstrating the conservative nature of the predicted multiplexing efficiencies that result when this methodology is applied to samples of real traffic. When augmented with a suitable set of routine traffic measurements, it is anticipated that the methods described here can play a significant role in practical ATM network dimensioning processes.