Study On Network Topology Inference Based On Network Tomography Technology

Nowadays, with the development of the network, the size of the network continued to expand, that also led to the network distributed and complicated. And people have become depending on the network increasingly, network management and monitoring have become more and more important. Because of distributed network structure and the confidentiality of internal information, the network measurement is very difficult. Network tomography technology is an emerging network performance measurement techniques, through the end-to-end measurements, can get the network performance parameters, combined with statistical inference of network packet loss rate, delay and link utilization and reconstruct the network topology. This technology overcome the shortages of the traditional measurement.This paper first introduces the basis of network tomography technology, including the classification of the basic principles and techniques, and network measurement techniques which has carried out a detailed analysis of the types, principles, and the need to measure network performance parameters.Secondly, studied of the network topology inference algorithm based on network tomography technology. Including end-to-end measurement techniques and several mature topology inference algorithm, find the performance of these algorithms and the advantages and disadvantages of them. Using an inference algorithm based on the network link utilization, because the utilization rate of path is the superposition, according to the parameters to be able to quickly find the relationship between nodes, to discover the network topology, the algorithm has less computation, and the high accuracy, but when the background has changed, especially the load increases, the accuracy decreased. To solve this problem, the introduction of the concept of the correlation distance, the improved algorithm effectively.Then study the structure of multi-source network topology inference, through the decomposition algorithm decomposed the structure of N-by-N into1-by-2and2-by-2,and constructed of four basic2-by-2packet. Considering the characteristics of multi-source network, we improve the probe packets of "sandwich" train. Through the measurement,can find the network link delay and packet loss information, make a unified conversion. A topology inference algorithm was designed based on the2-by-2structure, resulting in the topology.Finally with the NS2network simulation software to build a network topology model, measurement model to measure, obtained the data at the receiving node. Proved topology inference algorithm was validated.