Wavelet filtering of network traffic measurements

The “Caterpillar”-SSA (Principal Components of Time Series: Caterpillar Method, St. Petersburg University Press, 1997; Analysis of Time Series Structure: SSA and Related Techiques, Chapman & Hall/CRC, London/Boca Raton, FL, 2001) and statistical analysis based on the joint utilization of χ2 and ω2 tests provided the possibility to divide the whole set of components into two classes (VIII International Workshop on Advanced Computing and Analysis Techniques in Physics Research, ACAT’2002, 24–28 June 2002, Moscow, Russia, Book of abstracts, p. 176 (submitted to Physica D)). The first class includes leading components responsible for the main contribution to network traffic (Physica D 167 (2002) 72), and the second class involves residual components that can be interpreted as noise. More detailed analysis of the boundary region between these two classes may give additional information on traffic components and, thus, simplify the understanding of traffic dynamics. In this connection, we apply wavelet filtering to traffic measurements, and analyze its influence both on the characteristics of individual principal components and on the sum distributions of leading and residual components.