基于全源NT的链路时延分布推断技术  

Research on Delay Inference Technology Based on the Full Source Network Tomography

作　　者：段琪[1] 王备战[2] 蔡皖东[1] 

机构地区：[1]西北工业大学计算机学院,陕西西安710072 [2]厦门大学软件学院,福建厦门361005

出　　处：《厦门大学学报（自然科学版）》2011年第4期707-713,共7页Journal of Xiamen University：Natural Science

基　　金：国家高技术研究发展计划(863)项目(2009AA01Z424);教育部博士点基金项目(200806990030)

摘　　要：互联网链路性能具有非对称性,但是目前NT技术采用单源和多源测量,只能推断从源节点到目的节点一个路径方向上的链路性能,因此提出了推断功能更强的全源NT测量方法并解决了关键技术.证明基于包对测量方法和交叉汇合测量方法,严格全源网络结构的链路时延分布是可辨识的;提出采用期望最大化(EM)算法的链路时延分布的极大似然估计方法;最后通过模型仿真和网络仿真对推断方法的有效性进行了验证.The link performance of Internet is asymmetrical.At present,only the one-way link performance from source to destination can be inferred by the network tomography technology based on single source and multiple source measurements.Therefore,the full source measurement pattern is proposed in this paper.The link delay distribution of strict full source NT network structure is proved identifiable by using the cross joining probe method and the back-to-back pair probe method.Furthermore,the sufficient condition of the measurement sub-network selection which makes the link identifiable is proposed.The measurement traffic and the computational complexity could be reduced observably with the minimum measurement sub-network set.At last,the maximum likelihood estimation of link delay distribution computed by the EM algorithm is derived and the effectiveness is validated by the model simulation and network simulation results.

关 键 词：全源NT 时延分布 网络推断 EM算法 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]