关键词: 网络断层扫描, 时延分布, 最大期望（EM）算法, 移植算法, 时延估计

Abstract: The link performance inference is crucial to network quality assessment, however usually the present assessment methods can only infer the simple network with definite layer and can’t be applied to the large scale network. This paper proposes a maximum likelihood estimation based on incomplete data to estimate the delay distribution of the inside network. This method divides the tree-like network topology into different two-layer binary subtrees and estimates every chain’s delay of every subtree. And then the link delays are divided into every link through the transplantation algorithm and every subtree is done in this way with this method one by one, thus the link delays of the whole network are obtained. The feasibility and accuracy of the algorithm are verified through NS2 simulation.

Key words: network tomography, delay distribution, Expectation Maximization（EM） algorithm, grafting estimation, delay estimation