Network Link Delay Measurement Technique

With the developing of the Internet scale at very fast speed, the distributive, no corporative and heterogeneous characteristic of Network is becoming obviously. This makes the controllability and manageability of the system-level network become more and more complex, the knowability of network performance becomes more and more difficult. So how to grasp the overall status and optimize the performance of Network is becoming a major problem to be solved for network managers. At present the measurement of Network performance is a better way to solve this problem, and it controls the network behavior and characteristics by measuring the adoption of network performance, researching and analyzing the results of measurement. However the traditional network performance measurement techniques relies on the collaboration between nodes within the network and specific network system agreement, which will affect the network's real traffic and safety issues. Therefore many research institutes at home and abroad proposed new network performance measurement techniques known as network tomography technology, which can be used to statistics and infer the link-level performance by using the result of end-to-end performance measurement. It is important for the measurement of network performance.This paper researches and analyses the network performance measurement first, focuses on the comparison of advantages and disadvantages of active and passive measurement techniques. And it proposes a measurement technique based on multicast packet detection mechanism for T-packet probe packets. Then, that is for network tomography technique to analyze network performance models and statistical analysis methods to speculate. And comparing the MLE, EM with Bayesian estimation algorithms, to the final, this paper proposes an improved algorithm of EM algorithm. Finally, it builds a system for the link delay speculation. This system obtains the multicast end-delay by the measurement technique. And it statics and analyses the various link delay by using the improved EM algorithm and Gaussian mixture model. And ultimately, The thesis achieving the effect of grasping the overall operational status of network links.In the simulation stage, the paper compares the EM and improved EM algorithm, Gaussian and improved Gaussian mixture model with the true values by using ns-2 simulation. It proves that the improved EM algorithm and Gaussian mixture model are accuracy and efficiency by anglicizing the results. This has great significance for the research of the future network measurement and speculation.