An effective and accurate flow size measurement using funnel-shaped sketch

Abstract

Accurate measurement of network traffic is an essential part of current network management tasks. Sketch is a kind of probabilistic data structure widely accepted in network measurement. However, it is challenging for classic sketches to reduce the storage cost with little loss of accuracy for highly skewed network traffic. Therefore, most sketch-based schemes store elephant flows and mouse flows separately to deal with the skewed network traffic. But they are not usually suitable for estimating the size of mouse flows and thus may lose some information of traffic. To this end, a novel sketch, called Funnel Sketch (FS), is proposed in this paper. FS utilizes a funnel-shaped architecture to separately store the elephant flows and mouse flows, while maintaining the mouse flows as possible. Thus, FS can not only adapt to skewed network traffic to improve memory efficiency, but also accurately estimate the size of mouse and elephant flows with efficient memory consumption. Moreover, FS is implemented on CPU and OVS platform to evaluate its performance. The experimental results show that FS greatly reduces the error by 90% for flow size estimation and elephant flow detection compared with the state-of-the-art sketch-based schemes.