Reseach On Optimal Measurement And Identification Method Of Network Topology

The Internet brings convenience to human beings,while it also brings many unpredictable threats,thus people need to manage the Internet more effectively.And managing the Internet requires accurate measurements of the network topology.Network topology measurement is an important method for identifying the network topology.The internal measurement method and the end-to-end measurement method are generally used to calculate and analyze the measurement data to identify the network topology.Accurate measurement of network topology can effectively monitor the condition of network congestion,locate network faults exactly,measure internal network performance parameters precisely,optimize the network design and so on.Network tomography method is one of the mainstream network topology measurement methods.It can identify the network topology only by end-to-end measurements does not require the cooperation of internal nodes.However,end-to-end measurements bring less internal information of the network and are greatly affected by complex background traffic,resulting in large errors and insufficient input information for topology recognition.In addition,in the general tree topology recognition,the existing network tomography methods mostly delete the wrong links by setting a fixed threshold.In fact,setting a fixed threshold is not only very difficult,but also cannot be applied to all links,which leading to less accuracy of the topology recognition.This thesis studies point at the issues above mentioned and the main work includes the following three aspects:?1?Aiming at the problem of binary tree topology identification,a binary tree topology identification method based on subtree fusion is proposed.This method only carries on end-to-end measurements on a subtree with only three leaf nodes at a time,and identify the network topology by using the shared path length measurement parameters constructed by the structural characteristics of the subtree as the input.Comparing this method with the existing method in NS2,it has been found that this method effectively reduced the error caused by background traffic introduced by end-to-end measurement without increasing the network burden,and then improved the accuracy of the binary tree topology recognition.?2?Aiming at the problem of general tree topology recognition,a general tree topology recognition method based on burst detection is proposed.Based on the topology recognition of binary tree networks,this method uses a burst detection algorithm to detect the burst status of the link sequence of the binary tree,identify and delete the wrong links,and then obtain the correct general tree network topology.Comparing this method with the existing method.The results showed that the method identified and deleted wrong links accurately without setting a fixed threshold,which improved the accuracy of general tree topology recognition.?3?Aiming at the problem of network topology similarity evaluation,a network topology similarity evaluation method based on graph matching is proposed.This method matches the node with the highest Jaccard similarity coefficient,and uses the F₁ value of the matching link as an index to evaluate the similarity of the network topology.Compared with the existing methods,this method solved the limitation of the traditional network's topological similarity evaluation indexes,which proved this method is applicable to a wider range of topological similarity evaluation.