Study On Network Topology Inference Method Based On Unicast Double-parameters

In the modern network management, optimization, information war and new network service, network security plays a more and more important role. How to simply and safely manage and optimize the network has become a study focus of many researchers. As network topology measurement is prerequisite to network security and management, a study on this technology is imperative.Due to low traditional network measurement efficiency and accuracy, and increasingly high appeal of modern computer to network security, popularization of traditional network measurement methods is limited to a great extent. Network tomography(NT) is a new network measurement technology, which sources from medical insight. NT technology adopts end-to-end measurement method to reversely calculate property parameter of each network part, and makes use of probability and statistics to deduce the network topological structure, without intermediate nodes collaboration. Because only one group of target nodes is selected to measure the network performance and there is no need for intermediate nodes, concerns about network security during measurement process does not exist.However, the current NT technology has a narrow range of application. As the current technology can only calculate the simple tree-like topological structure, and a majority of actual network is network structure, how to apply the current technology into large topology structure is a difficult problem faced by the NT researchers. This paper targets at low accuracy of single parameter-based topology inference algorithm to introduce the network topology inference algorithm which is based on two parameters. Aiming at the shortcoming of previous methods that the noise impact is neglected, this paper proposes stochastic approximation algorithm and Kaczmarz algorithm to deal with the noise arising from the measurement process, which greatly improves the accuracy and working efficiency of network topology inference. See the followings for details:The first two chapters of this paper discuss the historical background and study value of this topic and explain in details the method and principle of traditional network measurement technology. Based on shortcomings of previous network measurement methods, a NT technology is introduced and its basic idea, model and key technology are also simply described.The third chapter introduces processing algorithms targeted for noise network — stochastic approximation algorithm and Kaczmarz algorithm. The experiments demonstrate that the results treated have higher accuracy than those untreated.The fourth chapter is the core of the paper. This chapter discusses in details network topology inference technology based on NT, highlights summary of currently single-parameter-based measurement methods and topology inference technology, and proposes two-parameter-combined topology inference technology after summing up shortcomings of the current methods.The fifth chapter is simulation experiment, where Network Simulator Version2(NS2) is used to validate accuracy of the two-parameter network topology inference algorithm and the SAK algorithm.