LFSN自相似网络流量模型的相关结构研究

Dependence Structure Study of LFSN-based self-similar network traffic model

在基于一般线性分形稳定噪声(LFSN)的网络流量模型中,需要对模型与实际流量样本的相关结构进行检验,以验证模型有效性。LFSN作为一种非高斯过程,一般不存在二阶矩,使基于二阶矩的方法不再适用,给检验带来一定的困难。本文考察了样本标准自相关函数(NACF)、共变(Covariation)和共差(Codifference)这三种度量,对相关结构进行实际验证,验证后二者的有效性,提出将一般化共差作为非高斯情况下相关性度量的一种统计手段。

Validation of dependence structure is required for the general LFSN （ Linear Fractional Stable Noise） self-similar network traffic, which does not posses its second order moment, because it is a non-Gaussian self-similar stable increment stochastic process. In this paper, three alternative measurement of dependence structure are taken into account, which includes Sample Auto Correlation Function （SACF） , Covariation and Codifferenee. By experimental comparison, we validate our simulation algorithm and propose the use of generalized codifferenee as a statistical instrument under non-Gaussian situation.