Study of UDP-based Internet traffic: Long-range dependence characteristics

Increasing demand for multimedia Internet applications today has shown progressive growth of the User Datagram Protocol (UDP) as the Internet transport protocol of choice for a large number of applications. However, its statistical characteristics and behaviour, specifically in terms of scaling-dependent properties are rarely studied. In this work, we firstly study the statistical characteristics of the UDP traces in terms of its long-range dependence properties as well as its marginal distribution. Secondly, based on the wavelet-based estimation method, we shall investigate the dependence structure of the wavelet coefficients in the light of the quasi-whitening concept, and lastly we shall consider a study for estimating the Hurst parameter (the degree of self-similarity) or the power law exponent for the long-range dependent processes that are present in the UDP Internet traffic. By analysing a large set of real traffic data taken from public repositories, it is evident that UDP Internet traffic reveals as long-range dependence with considerably high non-stationary processes and exhibits non-Gaussian marginal distributions. It is also interesting to see that analysis of the statistical properties of the wavelet coefficients shows that a reduction of the long dependence range to become short dependence range is impossible to be achieved by increasing the number of vanishing moments although it is done at a very coarse scale. Thus, it can be noticed that there is no significant difference on the performance of the Hurst parameter estimation for different numbers of vanishing moments for the mother wavelet.