Based On End-to-end Measurement Of Network Link Characteristics Parameter Estimation

Network Tomography can infer diversified network status parameters including topology without the cooperation of internal nodes. Network link parameters estimation is a major component of network tomography research. Network link delay parameters can describe the characteristics of the network link delay distributions, we can obtain the delay distributions by them, and they can be very useful to many purposes such as network troubleshooting, traffic control, multimedia, network design and so on. The existing methods for estimating the distribution of continuous delay can be devided into two categories:the first category method in advance assume that the dealy meet certain distribution, and then estimate the distributions' parameters by end-to-end measurements. the second category method, in advance estimate link delay parameters, and then use this parameter to estimate the delay distributions,without assumption of the delay meet some kind of distribution, but requires the collaboration between the internal nodes of network. This paper is aiming at the shortage of the two categories methods, and based on the second category methods, studied the method of network link delay parameters estimation and the method of delay distributions estimation using network link delay parameters.At first, this paper proposed a method, which completely relies on unicast end-to-end measurements, to estimate delay cumulant. Contributions are:(1) studied the model of the cumulant estimation, raised the binomial model of the delay cumulant estimation, and proposed the method to create model for the whole network based on the set of abstruct binomial network from the whole network. On this basis, proposed a measurement that using unicast back to back packet to measure the whole network and proposed a method of link delay cumulant estimation based on the unicast back to back measurement. (2) After analyzed all kinds of particular unicast packets which were used to simulate the multicast, proposed a method of link delay cumulant estimation based on the group of unicast probes measurement. (3) We studied the EM algorithm and the application of EM algorithm on delay estimation.Therefore, use the EM algorithm to estimate first order delay cumulant. (4) The above methods' accuracy has been validated by simulation on NS2.Based on the methods of link delay cumulants estimation which has been proposed in previous research, this paper further proposed a method of network link delay distribution function estimation based on the unicast end to end measurement. Contributions are:(1) studied the relationships among Edgeworth series, cumulants, probability density function and distribution function, raised a method that using the cumulants and Edgeworth series to approximate the distribution. (2) With the methods of link delay cumulants estimation, proposed a method of network link delay distribution function estimation based on the unicast end to end measurement. (3) With the result of pre-simulation, later simulation validated the method's accuracy.