Abstract: Inference of network internal link characteristics has become an increasingly important issue for operating and evaluating networks.Existing methods are usually focusing on the cases with fixed topology of networks，and are not applicable to networks where the topologies are unknown.This paper deals with the estimation of network link delay distribution by using the network tomography based on the Genetic Programming（GP） and the Pseudo Likelihood Estimation（PLE）.The Importance Sampling（IS） method is utilized to improve estimation of link delay distribution.Authors also implement the proposed method in a simulated network and validate its accuracy and efficiency.

Key words: network tomography, delay estimation, pseudo likelihood estimation, genetic programming, importance sampling

摘要： 网络内部链路性能推测对网络操作与评估至关重要，现有估计方法通常针对固定拓扑网络，无法应用于动态路由情形下的未知拓扑网络。提出了一种基于伪似然估计（PLE）和遗传程序设计（GP）的网络延迟断层扫描方法估计网络内部链路延迟分布，并利用重要抽样（IS）技术进一步改进链路延迟分布估计。最后利用仿真实验验证了该方法的有效性和准确性。

关键词: 网络断层扫描, 延迟估计, 伪似然估计, 遗传程序设计, 重要抽样