On the Use of the Schwarz Information Criterion for Analysis and Segmentation of Non-Stationary LRD Traffic

Long-range dependence (LRD) and other fractal characteristics have been found in network traffic. Among all the estimators of LRD, those based on the Discrete Wavelet Transform (DWT) are preferred due to their efficiency and accuracy, but do not consider the possibility of non-constant LRD parameters. We propose the use of the Schwarz Information Criterion (SIC) combined with the DWT and the Stationary Wavelet Transform (SWT) in order to detect changes in the signal’s variance structure, and segment the traffic into pieces with homogeneous fractal characteristics. The algorithm has been tested with synthetic and real traces.