一种新的网络业务流的多重分形小波模型

针对当前Internet上日益复杂的网络业务流和视频流，文章提出了一种新的多重分形小波模型．与普通的多重分形小波模型(Multifractal Wavelet Model，MWM)不同的是，该文提出的新模型在各个时间尺度上对小波系数都依据源数据尺度系数的边缘分布作了修正，这样确保新模型能在不同的时间尺度上拟合源数据的分布．决定网络业务流的排队分析特性的是一个关键时间尺度(Critical Time Scale，CTS)，CTS是随着缓冲区的容量和节点传输速率的变化而变化的，该文提出的模型能描述几乎各个时间尺度的业务流特性，因此能适应各种不同情况的缓冲区的排队分析．同时，新模型也继承了传统MWM的一些优良特性，比如能描述业务流的多重分形特征以及能确保最终结果始终是非负等等．最后通过对视频业务流和网络业务流的仿真实验与排队分析验证了该模型的有效性．

A New Multifractal Wavelet Model for Network Traffic

In this paper, we propose a new multifractal wavelet model for the increasing network traffic and video traffic on the Internet. Differed from the normal multifractal wavelet model (MWM), the proposed model pay more attention to the marginal distributions of all the time scales. For new MWM, we develop a novel shaping algorithm to shape the scaling coefficients from the coarsest scale to the finest scale, so new MWM could match the marginal distributions at almost all the time scales very well. As the buffer overflow probability is determined by a critical time scale (CTS) which varying with the buffer size and link rate, the proposed model can match the marginal distributions at a wide range of time scales, thus it can be used to different buffers queueing analysis. At the same time, the proposed model also inherits the good properties of the normal MWM, such as the multifractal properties, positive-value guaranteed and low computational complex. Finally, we use the proposed model to generate synthetic network traffic and video traffic, and compare their QQ plots with normal MWM's. QQ plots show that the proposed model could match the time-scale marginal distributions much better. Moreover, by trace-driven queueing analysis, we show the effectiveness of the proposed model.