Comparative Analysis of Different Sketch Methods in Practical Use

Network measurement is widely used in many aspects of network management and security area. Accurate estimation with limited resource is the basic requirement for good measurement methods. Our work aims to compare the accuracy between different sketch methods under the real network environment traffic. Sketch is a compact data structure used to summarize data stream. At present, there are some based-sketch measurement methods such as Count-Min Sketch, Deltoid, Reversible Sketch and UnivMon. We first introduce the working principles of these sketch methods and their theoretically performance. Then we implement the Count-Min Sketch and Deltoid in heavy hitter detection under real network traffic. Based on the results, comparisons in accuracy and resource consumption were done respectively while the per-flow algorithm with hash buckets was taken as the standard answer. Through our experiments, we give the suitable parameter settings in practice use.