Control Estimation Error of Sampling Method for Passive Measurement

Abstract

Sampling is increasingly utilized by passive measurement systems to save the resources consumption. However, the widely adopted static linear sampling selects packets with the same sampling rate (probability) for both large flows and small flows, which leads to intolerably high relative error for small flows. In order to bound the relative error for both small and large flows, we have proposed an adaptive nonlinear sampling method for passive measurement, which dynamically tunes the sampling rate according to the counter value. We have provided the unbiased estimation of the actual number of events n and have demonstrated that the relative error is radic[(1-1/n)a/2] for both large flows and small flows, where a is a constant parameter, and the counter size is bounded by a logarithmic function, log(1+an)/log(1+a). The theoretical and experimental results have shown that the proposed adaptive sampling method obtain a better tradeoff between relative error and memory consumption than existing sampling methods.