The Design Of Packet Loss Detection And Analysis System For Three-tier Network Base On Tomography Technology

In recent years, with the development and popularization of information technology, computer network is increasing its scale and business applications. The heterogeneity and complexity of the network are also increasing. Especially for large-scale three-tier LAN in the government, military, corporation and institution, these phenomena become more highlighted which urgently needs excellent performance detection system.Traditional network performance testing tool always requires cooperation among the network routers and other nodes, and is usually implemented based on internal nodes. Thus, it can not describe the overall performance of the network. In recent years, tomography technology is relatively mature in the areas of modern medicine and geographic, which has already been widely used in network performance testing. The basic principles of tomography are to use network end-to-end performance detection result and detection path matrix to infer the internal link performance of the network. However, three-tier network has huge scale, thus the number of end nodes is large and direct end-to-end measurement in the whole network faces the problem of high detection cost and high computational complexity. To solve these problems, through the research work on traditional network measurement and network tomography techniques, trying to combine the features of three-tier network topology and the characteristics of applications, a three-tier network packet loss detection and analysis system based on network tomography is designed in this thesis, and its key technologies are studied.Firstly, a three-tier network packet loss detection and analysis system is designed, which is based on the three-tier network topology and the characteristics of business applications. Since each of the upper three-tier network management domain contains multiple lower administrative domains, the system can detect the use of various management domains loss of parallelism, use a combination of distributed and centralized approach to inspect and analyze the lost packets at different levels of hierarchy network in a collaborated way, and combine the results of different level, which has been gradually expanded and refined view of network packet loss.Secondly, a method of calculating of packet-loss rate based on shortest link sequence and a reduction algorithm of shortest link sequence step-by-step are studied. Due to the limiting size of detection path, there exist some problems in an existing network tomography computing equation, which can not find the exact link level packet loss rate. Therefore, it can be realized to calculate the packet-loss rate based on shortest link sequence, and the results of the stepwise merger statute gradually expanded view of packet loss is detected, the progressive refinement packet loss rate is calculated. Experimental data shows that the model can be applied to packet loss detection with smaller overhead.Thirdly, since the shortest link sequence can not meet the needs of the current packet loss detection accuracy, an algorithm for constructing dynamic probing path set is proposed. The main idea of the algorithm is to index candidate probing path in increasing order by its information gaining, select a fixed number of round-end detection path, and refine the shortest link sequence and the corresponding loss of information based on the detection results. When the number of links to be solved is reduced to a certain limit, pre-selected algorithm is directed started to construct probing path set and send probes into the network, thus get the packet loss rates of all links. Experimental results show that the algorithm can refine the packet loss information of link sequence with smaller detecting overhead in a very short time, and thus satisfy the fineness requirements of detection and analysis.At last, some experiments are designed to verify the above system and algorithms.