多发送端拓扑结构中网络链路延迟推断方法  

Network link delay inference in multiple-source topology

作　　者：许鑫[1] 何泾沙[2] 石恒华[1] 

机构地区：[1]北京工业大学计算机学院,北京100124 [2]北京工业大学软件学院,北京100124

出　　处：《中南大学学报（自然科学版）》2010年第3期1058-1064,共7页Journal of Central South University:Science and Technology

基　　金：北京市自然科学基金资助项目(KZ200610005003);北京市教委科技发展面上项目(KM200510005008)

摘　　要：为检测多发送端拓扑结构中网络内部延迟情况,在充分利用路径延迟数据的基础上,提出一种简单易行的网络链路延迟分布推断方法。在满足网络平稳性、网络链路延迟的时间独立性和空间独立性的假设下,将复杂的多发送端拓扑结构的网络分解成多个简单的单发送端拓扑结构的分解单元,采用最大似然估计法并按照分解单元所含链路个数的升序推断各分解单元中的网络链路延迟分布,使得分解单元中的数据流共享链路延迟分布的真实值和估计值之间的差异逐渐减小。研究结果表明:采用该方法能有效推断出复杂的多发送端拓扑结构中网络链路延迟分布情况,与最小方差权值平均方法相比,具有较高的精度。Considering making good use of path delay data,an easily-accomplished method for inferring network link delay distributions was proposed to monitor network internal delay in multiple-source topology. Under the assumptions that the network was stationary,network link delays were temporally and spatially independent,the complex multiple-source network was decomposed into simple multiple single-source decomposition units. The ascending order of the number of links in every decomposition unit was regarded as the inference order for all decomposition units. Maximum likelihood estimation was adopted to infer network link delay distributions for each decomposition unit. The difference between the true and estimated value of network shared data-flow link delay distributions gradually decreased in these inference procedures. Simulation results demonstrate that this method can effectively infer network link delay distributions in complex multiple-source topology,and obtain higher accuracy than the method of minimum variance weighted average.

关 键 词：延迟 推断 拓扑 最大似然估计 多发送端 

分 类 号：TP393.06[自动化与计算机技术—计算机应用技术]