Study On The Analysis Of Traffic Fractal Dimension Based-on FARIMA

Through the research of Local Area Network (LAN) and Wide Area Network (WAN) traffic, Leland and Klivansky found the self-similar nature of Network traffic, but traditional network traffic models can not describe this nature well. Self-similar is very important to system performance such as Cell Loss Ratio (CLR), network delay, and so on, so study of self-similar models is increasingly significant nowadays. One difficulty of study is to estimate traffic fractal dimension.The paper according to fractal autoregressive integrated moving average (FARIMA) process models network self-similar traffic, estimates fractal dimension through fractal analysis methods directly and indirectly, and emphasizes on rescaled range (R/S) analysis. The paper develops R/S analysis which was referred by Hurst, and introduces R/S concept in broad sense, then analysis and discusses its computing condition, scale range and system error. While adopting R/S analysis to study traffic data, initial scale and maximum window size are main factors which influence the computational precision, for nonideal fractal can be approximated to beeline well only in some part range, which determined to initial scale and maximum window size.The paper discusses the computing condition and precision of variance-time analysis and variation algorithm based-on fractal dimension definition. During the research, we find that variance-time analysis also has some calculative scale range, namely limit of M sequence, so we must select proper range to get precise estimate result. With emulation analysis, during the long range, since amplitude of fluctuate gap of traffic data further less than amplitude of matrix width, estimate result of variation algorithm is on the high side of actual value, and this algorithm is not very proper to analysis traffic time series.