包延迟测量中泊松采样低估效应的仿真研究  

Investigating Underestimation Effect of Poisson Sampling on Packet Delay Dynamics Based on Simulation

作　　者：王恺[1] 李忠诚[1] 杨峰[1] 吴起[1] 毕经平[1] 

机构地区：[1]中国科学院计算技术研究所信息网络室,北京100080

出　　处：《系统仿真学报》2006年第5期1227-1232,1238,共7页Journal of System Simulation

基　　金：国家自然科学基金项目大规模互联网网络行为测量和分析方法(#90104006);异构网络服务质量控制算法研究(#60273021)

摘　　要：采样是网络测量的基础,泊松采样是一种无偏的采样方法的观点被人们普遍接受,也使其在主动测量领域得到了广泛应用。用户流包性能测量已成为主动测量领域的一个重要问题,然而泊松采样的无偏性仅针对网络状态,目前尚无研究工作研究其对于用户流的包性能的测量效果。利用真实的Internet流量数据进行了仿真试验,系统定量地研究了该问题得到如下结论:泊松采样系统低估了用户流包延迟性能;低估程度不受泊松采样频率的影响且随着网络链路利用率的升高而下降。定性分析和解释了泊松采样测量用户流包延迟性能时的低估现象。Sampling technique is elemental and important for network measurement. It is traditionally believed that Poisson sampling is unbiased, which makes it being widely used in network active measurement area. And it is well known that the unbiased estimation of Poisson sampling is true when network state is considered. As it is coming to be an important problem to measure the packet performance of user flow, there are not empirical efforts to investigate the performance of Poisson sampling on packet performance measurement from user＇s view. By using various actual Internet traffic datasets in simulation experiment, the difference of packet delay statistics between Poisson sampling results and user flow was systemically investigated, Poisson sampling systematically underestimates the statistics of packet delay, and the error degree decreases with the increasing of system utilization and has no relation with the frequency of Poisson sampling, Then the reason of above simulation results was analyzed.

关 键 词：主动测量 泊松采样 用户流 包延迟 利用率 网络测量 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]