ABSTRACT
With the rapid development of Internet , network services research has been
developed as a hot issue , the finding of self-similarity of network traffic has a very
important influence on traffic modeling and its performances evaluation . The
traditional Poisson model is not effective for the description of the network traffic
characteristics observed .
FBM model is a model of simple and easy to solve , this paper first describes the
definition of self-similar time series , and we estimated the Hurst index of the time
series generated by the FBM model. We described the lever of self-similarity by two
different methods of estimating the Hurst index .
In network environment , there are different kinds of non-linear dynamic
characteristics in the system , showing the chaotic attractor and a fractal structure,
which is a nonlinear , dissipative , non-equilibrium complex systems. Based on the
FBM model , self-similar simulation flow data are produced, that is the number of
packet arrival sequence over time . Depending on the time interval for data processing.
Phase space reconstruction theory has been widely used in recent years , which is the
basic step of non-linear modeling , the concept of phase space reconstruction first
appeared in the field of statistics . Later it was successively introduced in dynamics
systems by Ruell , Packard and Takens , which is an important step in nonlinear
dynamical systems , in order to well reconstruct the phase space , you must select the
appropriate delay time and embedding dimension . In this paper, Takens delay
coordinate reconstruction method is used to reconstruct the original time series .
Methods to determine the delay time including the autocorrelation function method, the
mutual information method and average displacement method , they have their own
advantages and disadvantages , auto-correlation function is used in this paper to
estimate the delay time of network traffic time series , with Grassberger-Procaccia
algorithm (GP algorithm) to compute the correlation dimension . In order to filter out
the irregular component in the sequence , wavelet analysis was utilized in nonlinear
time series , for network traffic time series , we can focus on local signal structure in the
changing the course , it can well reduce the scale parameter . To achieve the above
purpose , we use Daubechies wavelet-based discrete wavelet transform . As we all
know , the wavelet method can well filter high frequency and low frequency signal
components . Remove noise part by the wavelet filtering , and showing the more
reliable values of delay time and embedding dimension , then reconstruct dynamic
system with the delay coordinate method .
After filtering out the noise component of the sequence , we use the GP algorithm
to obtain the embedding dimension in the dynamic process with a more reliable results ,
this result has also been confirmed by principal component analysis (PCA) , the
principal component analysis is widely used in multivariate time series , including the
original data conversion in linear space , where the data set may be represented by a
reduced number of features and maintain the inherent characteristics. "Caterpillar"-SSA