突发业务的多重分形建模及其参数估计

Multifractal Traffic Modeling and Its Parameter Estimation

本文通过对大量突发业务数据的分析指出，局域网和广域网上的视频和数据业务的Ｈｕｒｓｔ系数随参数估计所用矩的不同而存在波动，但在很长的时间尺度下具有多重分形的特征．基于业务的排队仿真表明，采用分数布朗运动得到的等效带宽对具有多重分形特征的业务并不足够精确．因此，分析突发业务的多重分形谱有助于理解相同Ｈｕｒｓｔ系数的业务源在统计复用时可能出现的不同结果，并对业务的突发度给出更为精细的分析．为描述业务的多重分形特性，我们给出基于多窗口小波分析的多重分形参数估计，用来估计时变的Ｈｌｄｅｒ指数．仿真表明，此方法比基于粗粒化直接估计得到估计量的方差大为减小．

In this paper,we use parsimonious LAN and WAN traffic traces to examine the effectiveness of multifractal traffic modeling.We find that Hurst index may vary when we use different moment estimation while the multifractal property remains on a long range time scale.Traffic based simulation reveals that using fractional Brownian Motion model is not accurate enough to capture the queueing behavior of real trace exhibiting multifractal phenomena.In order to describe multifractal property,we propose a multi window wavelet transform method for the purpose of estimating the time varying scaling Hlder exponent.Numerical results are also given to show that the proposed method gives lower estimation variance than directly using coarse grain method.