基于离散模型的网络链路延迟分布推断方法  

Network link delay distributions inference based on discrete model

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：《华中科技大学学报（自然科学版）》2009年第S1期275-279,共5页Journal of Huazhong University of Science and Technology(Natural Science Edition)

基　　金：北京市自然科学基金重点资助项目(KZ200610005003)

摘　　要：为了在网络内部节点处于非协作的情况下能较精确地检测出网络内部延迟情况,提出一种基于离散延迟模型的网络链路级延迟分布推断方法.采用路径端到端多播测量法,针对不同的网络路径级延迟分布,使用分情况量化处理方式来选取离散延迟模型中的量化单元大小和量化范围,并在此基础上,通过采用最大似然估计和期望最大化算法对网络路径级延迟分布情况进行分析,来推断出网络链路级延迟分布情况.该方法不但可以合理地选取延迟分布的初始化值,而且可以减少估计值计算次数.仿真结果表明该方法能够较精确地推断出网络内部链路级延迟分布情况.A method for inferring network link-level delay distributions based on discrete delay model was proposed to accurately detect network internal delay characteristics without the cooperation of network elements.The method was based on end-to-end measurements of multicast traffic.According to different network path-level delay distributions,the bin sizes and the bin intervals in the discrete delay model were selected by using different binning patterns.Network path-level delay distributions were analyzed by maximum likelihood estimation and expectation maximization algorithm to infer network link-level delay distributions.It not only can rationally initialize delay distributions,but also can reduce computational times for estimation.Simulation results show the proposed method can accurately infer network link-level delay distributions.

关 键 词：网络测量 延迟 推断 离散模型 最大似然估计 

分 类 号：TP393.06[自动化与计算机技术—计算机应用技术]