Reseach On Non-stationary Network Topology Tomography

Network topology recognition is an important part of network performance monitoring. It is very important for many practical applications, such as link loss rate estimation and link delay estimation, to get network topology in real time and accurately. The topology estimation method based on network tomography can estimate network performance parameters in the context of no cooperation of inernal nodes. It only needs to send and receive packets between the terminal nodes to get end-to-end data and estimate the network topology using statistical inference methods. Traditional network tomography usually assumes that the network state parameters(including topology) remains unchanged over the entire measurement. But in real network environment, network status parameters are changing, so there may be large error between the topology estimation results and the real network topologies. Aiming at these problems, this paper proposed an estimation framework for non-stationary network environment and raised a network topology tomography based on the fusion of multi-metric parameters under this framework. The main work includes the following two aspects:(1) Selection and extraction of topology parameter estimation metrics in non-stationary environment: Due to the effectiveness of cumulants to deal with nonlinear and non-Gaussian problems, we applied it to non-stationary topology tomography to measure the length of the link and the path length of the network. Therefore we proposed the construction of new metrics based on higher order cumulants of delay. Firstly, we calculated the higher-order cumulants of path delay. Then, we built the relationship equations between the higher-order cumulants of path delay and the higher-order cumulants of shared path delay based on the cumulants model to calculate the higher-order cumulants of shared path delay. It was the new metrics of topology estimation.(2) Network topology tomography based on the fusion of multi-metric parameters: Because each order cumulant contains variety of statistical informations of delay distributions, in order to take full advantage of the delay distribution informations to estimate the network topology, this paper proposed a method of network topology tomography based on the fusion of multi-metric parameters. Firstly, we proposed the concept of multi-metric based on the the nature of the new metrics and decribed its construction process and several important properties in detail. Then, due to multi-metric, we proposed a topology estimation algorithm based on the fusion of multi-metric parameters. The algorithm makes full use of informations of shared path length and estimate network topology from the bottom up through an iterative approachIn this paper, we used Matlab and NS2 to simulate the entire process. The simulation results showed that under the non-stationary network environment, the new metric contained a wealth of statistical information of the delay distribution and could reflect the time-varying characteristics of the network. Meanwhile, the simulation results demonstrated that the non-stationary network topology estimation method based on multiple metrics could estimate the topology of non-stationary network and got a better result than the result of stationary network tomography method.