IPro: An approach for intelligent SDN monitoring

Abstract

Traffic Monitoring assists in achieving network stability by observing and quantifying its behavior. A proper traffic monitoring solution requires the accurate and timely collection of flow statistics. Many approaches have been proposed to monitor Software-Defined Networks. However, these approaches have some disadvantages. First, they are unconcerned about the trade-off between probing interval and Monitoring Accuracy (MA). Second, they lack intelligent mechanisms intended to optimize this trade-off by learning from network behavior. This paper introduces an approach, called IPro, to address these shortcomings. Our approach comprises an architecture based on the Knowledge-Defined Networking paradigm, an algorithm based on Reinforcement Learning, and an IPro prototype. In particular, IPro uses Reinforcement Learning to determine the probing interval that keeps Control Channel Overhead (CCO) and the Extra CPU Usage of the Controller (CUC) within thresholds. An extensive quantitative evaluation corroborates that IPro is an efficient approach for SDN Monitoring regarding CCO, CCU, and MA.