ip骨干网流量成分分析分析word论文.docx

摘要
网络流量的建模与预测在网络管理中扮演了非常重要的角色。在过去对流量的分析和处理中,研究者们发现网络流量可以分解为代表其长期增长趋势的趋势分量、代表其周期性变化的周期分量,以及随机波动的分量。这种流量分解的方法被广泛应用于流量的长期预测、流量的周期性波动分析等领域。
在本文中,我们试图对某中国网络运营商的 IP 骨干网络流量进行成分分析,寻找其变化规律。本文所针对的数据采集自一个 IP 骨干网,数据包括 12 个骨干节点的日峰均值,为期 523 天。我们采用了两种网络流量的分解方法,一种假设趋势分量和周期分量呈现线性组合,另一种假设它们呈现加性组合。基于这两种分解方法,我们可以分别获得网络的长期增长趋势分量以及以星期为单位的周期分量。我们对不同节点的流量进行了分解,研究对不同节点而言较为合适的流量分解方法;同时,我们研究各节点的趋势分量和周期分量的特征,比较其特征的异同。
我们的研究结果表明,在 IP 骨干网络节点的趋势分量方面,基于指数的拟合模型较为适合网络的长期趋势建模,而基于 ARIMA 的模型较为适合网络的中短期的趋势建模;在 IP 骨干网络节点的周期分量方面,所有节点均呈现了以七天为单位的周期性特征,但是其振荡的幅度与各节点的流量特征有密切的关系。
关键词:流量分析,时间序列模型,IP 骨干网
Abstract
Network traffic modeling and prediction play a very important role work management. In the past on the flow analysis and processing, the researchers found that the ponent, ponent work traffic can be posed as the representative of its long-term growth trend on behalf of itsperiodic variation, and random ponents. Long term prediction, flowanalysis of the cyclical fluctuation of the flow fields of the position is widely used in flow.
In this thesis, we attempt to IP work traffic to a work operators ponent analysis, to find out the regularity of changes. The datacollected from an IP work, data include daily peak value of 12backbone nodes, for a period of 523 days. We use the position method of two kinds work traffic, a hypothesis that the ponent and ponent appears to be a bination, another hypothesis they bination. These two kinds of position based methods, we canlong- term growth ponent work respectively in weekly ponents. We pose the different node flow, the more appropriate for different node flow position method; at the same time, the ponent and ponent features of our study of each pares the characteristics.
Our results show that, the ponent IP work node, the long-term trend index is more suitable for modeling fitting model based work, and the ARIMA based model is more suitable for work in the short-term trend model; in the ponent IP work node, all nodes arepresented in seven days characteristics