Research On Performance Analysis And Model Of Network Traffic

With the rapid growth of communication technology, the broadband network becomesmore and more complicated. To manage and maintain the network more efficiently, we need toevaluate the network performance based on which the network configuration can be optimized.Along with the discovery of self-similarity in network traffic, a large number of reseach in thisarea have been emerging and providing powerful tools for the analysis of self-similar networktraffic.Most of the current reseaches on self-similarity focuse on the model of long-rangedependence and the estimation of Hurst parameter. The Hurst parameter represents the degreeof self-similarity. However, reaseach results showed that small scale traffic indicated localsingularity, which required a multifractal model to characterize. Thus, the local H lder indexand multifractal spectrum are used to evaluate the multifractal characteristic of network traffic.In this thesis, we firstly introduce the concepts of self-similarity, long range dependence,Hurst parameter and heavy-tails distribution. Based on these theories, the self-similarity of realnetwork traffic is verified. We adopt classical R/S method to estimate the Hurst parameter andverify that the real network traffic has a strong burstness. Secondly, to prove the heavy-tailscharacteristic in network traffic, a heavy-tails distribution (Alpha-stable distribution) isintroduced. By fitting the real traffic with the Alpha-stable distribution, we show that the taildistribution of the real network traffic is much heavier than the Gauss distribution. Moreover,the characteristic index of Alpha-stable distribution also valids that the network traffic hasself-simiilarity.Finally, multiplicative model is used to model multifractal network traffic, and in this basis,a new multifractal model (which is called as Variable Scale parameter Cauchy MultiplicativeModel, V.S.C.M) is proposed. Through simulations on the local H lder index and the multifractal spectrum, it proved that the proposed model is multifractal, and fit the originalnetwork traffic accurately.