The Study And Applications In Non-stationary Network Of Unicast Network Loss Tomography

Internet has evolved from a simple small-scale network to the complexlarge-scale network with multi-platforms and multi-terminals.The heterogeneous andno-corporative structure of the internet makes the traditional network measurementswhich requires the cooperation of internal node facing great challenge.How toaccurately and timely estimate network performance characteristics in more and morecomplex internet environment have become one of the forefront scientific problemswhich are foccused by the academic community all over the world.In order to describethe performance of network accurately,such as link loss rate,and improve networkmanagement and control,a new estimation approach named Network Tomography wasproposed and development rapidly.Network tomography is a new technique to infer the network internalperformance or logic topology only by end to end measurements.The early research ofNetwork loss tomography has a lot of problems,for example,there are correlationdefects in the package group,meanwhile, they only concern about the link loss rate ofstationary network used mainly some algorithm of statistics,and can not obtaintime-varying characteristics of link parameters.In this paper,in order to strengthen the correlation between probes and indeedimprove inference accuracy,we propose a new probe-sending methods to get theinternal loss rates.Previous studies showed that the network packet loss characteristicsis continuous,based on this theory, we adjusted the package group in the unicasttomography.NS2simulations demonstrate the feasibility of this tomography model.Our estimation method to solve the packet loss rate of non-stationary network isbased on the theory of numerical analysis,in this way,we improve the sliding windowmodel for estimating time-varying packet loss rate.The sliding window modelassumes in a relatively short time window,the time-varying curves of link loss rate aredescribed by a k times continuous differentiable function.The k-th order TaylorSerieses of this function are estimated using network tomography approach.Then,based on the estimates of each window,the time-varying link loss rates of entiremeasurement period are obtained by integrating the estimates of all time windows.Weuse mathematical methods such as data fitting algorithm to improve thismodel.Finally,we study the packet loss rate of non-stationary network by the new probe-sending method and the improved sliding window model.The simulation resultsshow that the improved packet loss tomography model of non-stationary network canapproximate the actual time-varying packet loss rate.