AGC Sketch: An effective and accurate per-flow measurement to adapt flow size distribution

Abstract

Network measurements are indispensable for efficient network management. As a probabilistic data structure, sketch is widely used in network measurements. To adapt to the skewed and dynamic distribution of network traffic, most of the existing sketch-based schemes separate large and small flows either by asymmetric structures or multilayer structures. However, the former approach requires further accuracy of measurement due to increased errors of small flows. And the latter approach limits the throughput due to the increased hash calculations for large flows. To this end, a novel sketch, called AGC Sketch, is proposed in this paper. By storing the large flows and small flows separately while maintaining the small flows, it can not only adapt to the skewed network traffic to achieve efficient memory utilization, but also accurately estimate the size of flows with high throughput. Moreover, AGC Sketch is implemented on CPU and OVS platform. The experimental results show that AGC Sketch greatly reduces the error by 90% for flow size estimation compared with the state-of-the-art schemes.