基于分层抽样的高速网络吞吐率测量  

Throughput measurement technique for high-speed network based on stratified sampling

作　　者：张峰[1] 雷振明[1] 

机构地区：[1]北京邮电大学ATM技术研究中心,北京100876

出　　处：《吉林大学学报（信息科学版）》2004年第6期557-563,共7页Journal of Jilin University（Information Science Edition）

基　　金：国家重大自然科学基金资助项目(69896240)

摘　　要：在测量精度要求较高时,随机抽样流量测量的样本容量仍然很大,仍会造成一定程度的资源负担和测量开销上的问题。针对前面这种简单抽样策略的局限性,提出了一种基于报文分层抽样的高速网络吞吐率测量技术,并对分层抽样参数的选取及其理论进行了探讨;从网络吞吐率测量的角度对分层抽样与简单随机抽样的测量性能进行比较。结果表明,在相同样本容量的情况下,分层抽样测量精度几乎平均是简单随机抽样精度的9倍,且算法复杂度仅为O(n),有效解决了高速网络测量环境中测量效益不高的问题。这种基于分层抽样的测量技术还可以用于其他网络流量参数的测量。Random sampling traffic measurement still encounters the large sample size under the high measurement precision requirement, and results in high overhead on memory and processing to some extent. Contrasting to the limitation of this previous simple sampling strategy, a new technique based on stratified sampling is proposed to measure the throughput of high-speed network, and a thorough research is made on the decisions of parameters and its theories; the detailed examinations are performed on precision of measurements in the respect of network throughput, which indicates that the precision of stratified sampling is averagely 8 times higher than that of the simple random sampling under the same sample size, and its complexity is O(n), so it resolves well the low efficiency of measurement in high-speed network environment. This technique is also flexible for the measurement of other network traffic parameters.

关 键 词：报文抽样 分层抽样 随机抽样 吞吐率 

分 类 号：TP393.07[自动化与计算机技术—计算机应用技术]