A Modular Sampling Framework for Flexible Traffic Analysis

Abstract

The paradigm of having everyone and everything connected in an ubiquitous way poses huge challenges to today's networks due to the massive traffic volumes involved. To turn treatable all network tasks requiring traffic analysis, sampling the traffic has become mandatory triggering substantial research in the area. Aiming at fostering the deployment and tuning of new sampling techniques, this paper presents a flexible sampling framework developed following a multilayer design in order to easily set up the characteristics of a sampling technique according to the measurement task to be assisted. The framework implementation relies on a comprehensive sampling taxonomy which identifies the granularity, selection scheme and selection trigger as the inner characteristics distinguishing current sampling proposals. As proof of concept of the versatility of this framework in testing the suitability of distinct sampling schemes, this work provides a comparative performance evaluation of classical and recent sampling techniques regarding the estimation accuracy, the volume of data involved in the sampling process and the computational weight in terms of CPU and memory usage.