Abstract: Packet sampling technique is a key technique in high-speed network traffic measurement and network supervise. In this paper, by introducing into five parameters which are stratified characters, the number of strata L, stratified boundary, sample size assignment of each stratum and sampling strategy inside stratum, and through reconfiguring them and analyzing their theory simply，we improve the technology of stratified sampling. After then we use simple linear estimation method to inferring original information, further more by using Φ discrepancy examine method, we carry on the accuracy performance comparison between improved stratified sampling technique and other sampling methods form packet size distribution aspect .The result indicates that the accuracy performance of improved stratified sampling technique is well-above other sampling methods in packet size distribution aspect, and improves measured precision greatly.

摘要： 报文抽样技术是高速网络流量测量和管理中使用的一项关键技术。本文通过引进分层特征、层数L、分层边界、各层样本量分配、层内抽样策略5个分层抽样参数，并对其进行重新配置和简单理论探讨，实现对分层抽样技术的改进。同时文章使用简单线性估计推断原始流数据，并借助于Φ偏差检验方法，对改进的分层抽样技术和其它抽样技术在测量网络报文长度分布方面进行准确性性能比较。结果表明，改进的分层抽样技术在测量报文长度分布方面的准确性性能远高于其它抽样方式，提高了测量的精度。