On the Use of Principle Component Analysis for the Hurst Parameter Estimation of Long-Range Dependent Network Traffic
Long-range dependency and self-similarity are the major characteristics of the Internet traffic. The degree of self-similarity is measured by the Hurst parameter (H). Various methods have been proposed to estimate H. One of the recent methods is an eigen domain estimator which is based on Principle Component Analysis (PCA); a popular signal processing tool. The PCA-based Method (PCAbM) uses the progression of the eigenvalues which are extracted from the autocorrelation matrix. For a self-similar process, this progression obeys a power-law relationship from which H can be estimated. In this paper, we compare PCAbM with some of the well-known estimation methods, namely; periodogram-based, wavelet-based estimation methods and show that PCAbM is reliable only when the process is long-range dependent (LRD), i.e., H is greater than 0.5. We also apply PCAbM and the other estimators to real network traces.
KeywordsPrinciple Component Analysis Fractional Brownian Motion Hurst Parameter Regression Plot Autocorrelation Matrix
Unable to display preview. Download preview PDF.
- 4.Crovella, M., Bestavros, A.: Self-similarity in World Wide Web Traffic: Evidence and Possible Causes. SIGMETRICS, Philadelphia, USA (1996)Google Scholar
- 10.Ozkurt, T.E., Akgul, T., Baykut, S.: Principle Component Analysis of the Fractional Brownian Motion for 0 < H < 0.5. In: Proc. of IEEE ICASSP, Toulosse, France (2006)Google Scholar
- 11.Lau, W.C., Eramilli, A., Wang, J.L., Willinger, W.: Self-Similar Traffic Generation: The Random Midpoint Displacement Algorithm and Its Properties. In: Proc. of ICC, Seattle, USA (1995)Google Scholar
- 17.UNC DIRT Laboratory Internet traces, http://www-dirt.cs.unc.edu/ts/
- 18.Erol, M., Akgul, T., Baykut, S., Oktug, S.: Analysis of the Hurst Parameter Estimation Methods for Network Traffic Containing Periodicity. Elsevier Computer Networks (submitted, 2006)Google Scholar