Study On Topology Inference Method Based On Network Tomography Technology

With the development of science and technology，the scale of network is expandingunceasingly.The features of network are more obvious which include the complex ofdistribution, isomerization, and not collaborative. And these have increased the difficulty ofnetwork management; At the same time, the demand for the quality of network service is alsomuch higher, so that how to monitor and manage the performance of the network, improvethe quality of network service is the primary problem for network developers and users. Andthe measurement of network performance has become an important research field.Because the method of traditional network measurement is restricted by many factorsand the network performance can only be measured in he authorized network.. Networktomography is a new technology of network measurement,which does not need to understandthe relevant information of network measurement, using the data received by the receivingnode,and using the method of statistical inference to infer the performance of the measurednetwork.. It has the advantages of traditional network measurement which does not have,such as no need for the cooperation of internal node in the autonomous system, not depend onspecific network protocol, and this is one of the important reasons for its development.This paper first summerizes the network measurement technology and networktomography technology. And then it introduces the main content of network tomographytechnology including the performance inference of packet loss rate, delay and topology. Itintroduces the network topology seriously including the measuring method, the calculationmethod of the nodes’ correlation and topology inference algorithm.Then, this paper introduces the topology algorithm based on the Manhattan distance andHGLT algorithm. The former judged the brother nodes by calculating the Manhattan distanceof two nodes, but it can only infer the topology of the two fork tree; And the later can inferthe multi-branches fork tree, but its inference accuracy can be improved. Because the formercan only infer the topology of the two fork tree, this paper proposed the MDA algorithm toinfer the topology of multi-branches tree. It firstly calculates the Manhattan distance of thenodes to judge the brother nodes, and then it uses the successful transmission rate of the probepackets to judge the brother node which is also the brother of the other brother nodes alreadyjudged; And the above operations are repeated until the whole network topology is inferred.In addition, this paper also proposes the SDA algorithm to infer the topology of thenetwork using the time delay. The network topology can be inferred by calculating theaverage delay of the two short data received by the same node.Finally, using NS2simulation software and building the simulation envirment, it verifiesthe reasonableness and efficience of the MDA and SDA algorithms. The simulation results verifies that compare with HGLT algorithm,MDA algorithm can obviously improve thetopology inference accuracy of the multi-branches tree; And SDA algorithm can accuratelyinfer the network topology of the multi-branches tree.