In [1, 2] we applied a nonlinear analysis to traffic measurements obtained at the input of a medium size local area network. The reliable values of the time lag and embedding dimension provided the application of a layered neural network for identification and reconstruction of the underlying dynamical system. The trained neural network reproduced the statistical distribution of real data, which well fits the log-normal form. The detailed analysis of traffic measurements [3] has shown that the reason of this distribution may be a simple aggregation of real data. The Principal Components Analysis of traffic series demonstrated that a few first components already form the fundamental part of network traffic, while the residual components play a role of small irregular variations that can be interpreted as a stochastic noise [4]. This result has been confirmed by application of the wavelet filtering and Fourier analysis to both the original traffic measurements and individual principal components of original and filtered data [5]. The log-normal distribution of traffic measurements and a multiplicative character of traffic series confirmed the applicability of the scheme, developed by A. Kolmogorov [6] for the homogeneous fragmentation of grains, also to the network traffic.

PDF (1.42 Mb)