基于随机矩阵理论的链路流量监测

Monitoring of link traffic patterns based on random matrix theory

全网链路流量监测对流量工程和网络攻击监测都有重要意义但是一直很难实现。该文提出了一种基于随机矩阵理论(RMT)的监测方法,利用流量协方差矩阵和随机矩阵理论预测结果进行比较,从两者差异中提取流量的时间相关信息。在网络中布置少量观测点,利用协方差矩阵的最大特征值能够准确的获取高速和低速链路的流量信息,并且通过最小特征值监测到观测点是否工作。该方法节省存储资源,计算时间短,获取信息多,是一种高效的监测方法。

Monitoring of traffic patterns or links throughout the entire networks is of great importance to traffic engineering and for monitoring attacks. With increasing network size, the extraction of key information is becoming more complex and storage costs are high. Therefore, monitoring of the entire network has been very difficult to achieve. This paper presents a method based on random matrix theory （RMT） to extract time-related information from deviations between the flow cross-correlation matrix and the RMT predictions. Only a few observation points are needed to extract information about high-speed and low-speed links using the largest eigenvaloe of the cross-correlation matrix. Information about the observation points is well captured by the small eigenvalues. The monitoring method is cost-effective with small storage and short computational times giving extensive information.