Flow Characteristics Estimation Based On Multilayer Virtual Active Counter Sharing in Data Center Network

Abstract

Per flow characteristics estimation has a variety of applications in the fields of traffic engineering and routing distribution, etc. Some recent works have adopted different methods to estimate the flow size, and the counter architecture used can be suitable for compact SRAM. However, the existing counter architecture still has limitations. Either it still requires a considerable amount of SRAM space, or the processing speed is reduced in order to ensure accuracy in a tight space. At the same time, the scalability is insufficient when performing multi-characteristic estimation. It is not suitable for flow characteristics other than the flow size. In this paper, we design a scalable active counter architecture called Multilayer-VAC, which can not only achieve better counting efficiency, but also can be used to count multiple characteristics of traffic. Experiments conducted with real network trajectories show that this counter architecture can produce good estimation of various sizes of traffic characteristic values in a compact memory space.