Relieve the Control Plane Burdens in Heavy Hitter Detection with Selective Report

Abstract

Sketch algorithms are apt at detecting Heavy Hitters accurately at real time with low memory overhead. However, traditional wisdom of network measurement mainly concentrates on the two key properties: accuracy and memory-efficiency, which only targets the data plane. This paper aims to achieve timeliness and communication-efficiency in Heavy Hitter detection by properly coordinating the interplay of the data plane and the control plane, thus further relieve the control plane burdens. Results show we maintain the accuracy in Heavy Hitter detection while reduce the delay from microseconds to nearly zero and communication overhead by at most 70%.