On parameter estimation of a simple real-time flow aggregation model

There exists a clear need for a comprehensive framework for accurately analysing and realistically modelling the key traffic statistics that determine network performance. Recently, a novel traffic model, sinusoid with uniform noise (SUN), has been proposed, which outperforms other models in that it can simultaneously achieve tractability, parsimony, accuracy (in predicting network performance), and efficiency (in real-time capability). In this paper, we design, evaluate and compare several estimation approaches, including variance-based estimation (Var), minimum mean-square-error-based estimation (MMSE), MMSE with the constraint of variance (Var + MMSE), MMSE of autocorrelation function with the constraint of variance (Var + AutoCor + MMSE), and variance of secondary demand-based estimation (Secondary Variance), to determining the key parameters in the SUN model. Integrated with the SUN model, all the proposed methods are able to capture the basic behaviour of the aggregation reservation system and closely approximate the system performance. In addition, we find that: (1) the Var is very simple to operate and provides both upper and lower performance bounds. It can be integrated into other methods to provide very accurate approximation to the aggregation's performance and thus obtain an accurate solution; (2) Var + AutoCor + MMSE is superior to other proposed methods in the accuracy to determine system performance; and (3) Var + MMSE and Var + AutoCor + MMSE differ from the other three methods in that both adopt an experimental analysis method, which helps to improve the prediction accuracy while reducing computation complexity. Copyright (C) 2005 John Wiley & Sons, Ltd.