Analysis of traffic flow measurements by rate-interval curves

In this paper we propose a method for the analysis of aggregate traffic by means of empirical rate-interval curves obtained from experimental data. The approach is related to the theory of Network Calculus, and is based on an algorithm which is different from the commonly employed approach based on wavelets, although it retains some of its multiresolution features. We present results, obtained both by simulation and in the analysis of real traffic traces, which provide an assessment of the strengths and weaknesses of the proposed method. Rate-interval curve analysis provides very robust and acceptably accurate estimates of the Hurst parameter value and, even in the presence of flow irregularities, results can be proved to be correct as far as scaling properties are concerned. Further analyses concerning peaks, bursts and similar localized phenomena that may have a significant impact on the performances of a network are allowed by considering maximal rate envelopes.