Research On Topology Inference Based On Network Tomography Technology

Today's Internet is a huge distributed network.As the society continues to grow,its scale continues to expand.It also poses greater challenges for network management,optimization,and fault diagnosis.If the network topology can be identified,all these problems will be simplified.There are two main methods for topology inference of fixed networks:one is the traditional network inference method,which analyzes and processes the related topology information of nodes inside the network to infer the network topology.This method depends on the collaboration of internal nodes in the network.In addition,the communication load increases in the information collection process,which cannot be ignored;the other method is network tomography,which adopts active measurement or passive reception at the network boundary without cooperation from internal nodes.The way to obtain useful information within the network,and then use statistical methods for topology inference.In this paper,the network tomography technology is introduced in detail from the aspects of system model,parameter measurement method and logical topology inference algorithm.Among them,the network inference algorithm based on network tomography technology is mainly explored,and its outstanding features and existing problems are illustrated,and improvements are made.Firstly,for the problem of high complexity of topology estimation based on maximum likelihood(especially when the network scale is large),an improved fast topology estimation method with regular terms based on maximum likelihood is proposed,which effectively reduces the computational complexity.Proof and simulation experiments were demonstrated.In addition,the existing hierarchical topology estimation algorithm has a performance degradation when the variance of node correlation is large.To this end,a network topology inference algorithm based on merged hierarchical clustering(MHT)is proposed,which uses bottom-up consolidation.In each hierarchical clustering procedure,each cluster only uses data related to the largest relevant node pair.Compared with the traditional hierarchical topology estimation algorithm,the computational complexity is reduced and the parameter estimation accuracy of the algorithm is improved.