Loss Performance Inference Algorithm Based On Network Tomography

As the Internet grows in size and diversity,its internal performance becomes ever more difficult to measure.Any one organization has administrative access to only a small fraction of the network's internal nodes,whereas commercial factors often prevent organizations from sharing internal performance data.The most challengeable thing may be the measurement of Internet.Because of its distributive,no corporative and heterogeneous characteristic,accurate measurement of such a network is very difficult.One promising technology named network tomography has been proposed as an efficient approach to analyze network performance.This approach uses end-to-end measurement to infer the inner network parameters.It neither needs to deploy a measurement node inside the network nor requires cooperative node.Network tomography can not affect the network traffic,and also can not cause the security problem.In this paper,we considere the problem of characterizing link-level loss behavior through end-to-end measurements.First we elaborate in detail the derivative process of multicast-based Expectation Maximum-Maximum Likelihood Estimate Algorithm (EM-MLE) and Pseudo Maximum Likelihood Estimate Algorithm(PMLE).Base on the EM-MLE and PMLE,we presented an approach to link loss inference from end-to-end data;our approach has the advantages of high efficiency.Base on PMLE,a new algorithm(μ-PMLE) is proposed to infer the inside link loss,through the introduction of correction factor,μ-PMLE algorithm has higher accuracy.In this paper,we also introduce the multiple points measurement,and describe a fast Minimum variance weighted average algorithm(MVWA).We research the unicast-based loss inference algorithm,in this paper;we offer a unicast-based flexible Expectation Maximum algorithm.At last,we describe the simulated environment based on NS2.According network traffic model,we construct the background and detection traffic,and analyze each characteristic.Simultaneously used the massive experimental result,we determine same important parameters.We used simulation to analyze all algorithms in this paper, summer up the advantages and disadvantages of various algorithms.