Study On Topology Inference Based On Tomography And Network Coding

Aided by end-to-end measurement methods and statistical inference theory, network tomography(NT) is able to estimate internal performances and topology structure of a network. Having an accurate knowledge of topology structure is a significant prerequisite for monitoring and assessing the other performance parameters of the whole network, thus topology inference based on network tomography has been an important direction in the field of network measurement(NM). The emergency of network coding(NC) breaks the restrictions imposed by traditional routing technology on nodes behavior. The coding process of nodes can reflect in some extent the state of nodes or links when encoded packets flowing through them. Network coding enhances the dependence of packets on topology structure, which can exactly provide a solid basis for further exploiting network topology tomography methods. Selecting an appropriate end-to-end measurement approach is the principal procedure to solve the problem of network tomography. The selections include probing sources and receivers, metric function, data processing approaches of the internal nodes(which refers to the simple “store-and-forward” or network coding), etc. In view of topology inference problem, the following researches are conducted in this paper:1. In view of the existent deficiency of adopting single parameter metric, an active topology inference algorithm is put forward which combines both delay jitter and packet loss rate. Simulation models are established based on OPNET to acquire network simulation data, which can be used to analyze and calculate correlations between nodes, thus reconstructing the candidate network.2. In binary tree topology, to ensure that any leaf nodes which are interested in network structure can estimate the whole topology exactly through its own observations exclusively, a network coding based active topology inference algorithm(NCATI) are proposed in this paper. Complicated statistical calculations are avoided in the proposed algorithm, which lowers the difficulty of inferring topology.