Stochastic Sampling for Internet Traffic Measurement

The increasing complexity of the Internet demands continued improvements to measurement techniques and data analysis methods to aid our understanding of network operation. The availability of accurate measurement data is necessary in many areas ranging from attack detection, novel pricing schemes, buffer dimensioning and switch design to general network management. In this paper, we develop a theory for accurate and unbiased Internet traffic measurement using the tools of Poisson random sampling. We show how this approach helps in storing, managing, and aggregating data from different sources with independent clocks and sampling rates. We present results that show that stochastic sampling maintains important information about network measurements that would be lost when using conventional uniform sampling.