AdaptMon: Adaptive Polling-Based Network Monitoring

Network monitoring promotes stable network operation by giving Software-Defined Networking (SDN) controllers data on the dynamically changing underlying infrastructure, such as traffic flows, bandwidth usage, network latency, error rates, and device health and status. Dynamically changing SDNs pose challenges for periodic monitoring and computationally intensive adaptive monitoring. New protocols are needed that can obtain monitoring data quickly and adapt to changes flexibly. We propose AdaptMon, an adaptive periodic polling monitoring algorithm. AdaptMon leverages network error frequency within a flexible allocation scheme based on Linear Programming (LP), ensuring polynomial complexity, avoiding over-polling one switch while maintaining adaptivity. The object of this paper is to express the monitoring fairness and proportionality trade-off as an LP and to investigate the feasibility of using this solver in a real-time system. Our findings show that AdaptMon reacts to changing network conditions and performs error detection around 40% faster than traditional periodic polling. By prototyping AdaptMon we demonstrate that deploying the AdaptMon LP in an SDN is feasible. We identify the following perspectives: network error awareness is crucial to reduce monitoring delay; formulating adaptive periodic monitoring as an LP balances flexibility with computational efficienty.