Reliable paths prediction with intelligent data plane monitoring enabled reinforcement learning in SD-IoT

Abstract

Legacy routing with conventional distributed networking in the Internet-of-Things (IoT) networks needs to work on efficiently handling service requests, leading to compromised Quality of Service (QoS) for user demands. In this research, we introduce Reliable routing based on Reinforcement learning in SD-IoT Networks (RRSN), an intelligent routing technique leveraging Software-Defined Networking (SDN) in IoT networks. RRSN utilizes a hybrid mechanism that combines Machine Learning (ML) and Reinforcement Learning (RL) within SDN's centralized control to optimize QoS-aware reliable routing. In addition, network telemetry is utilized for intelligent network monitoring in the data plane to get the underlying network's status efficiently. The proposed approach consists of three main modules: a Support Vector Regression (SVR) machine learning algorithm to predict link reliability, a traffic classifier for QoS-based traffic classification, and a reliability-aware reinforcement learning algorithm to compute optimal routing paths for IoT sensors. SVR algorithm has a high prediction accuracy, is lightweight, less susceptible to noisy data, and its computational complexity is independent of the dimensions of the input space. RRSN leverages intelligent network monitoring with RL, global view, and centralized SDN control to compute and implement optimal routes in data plane switches efficiently. We conduct extensive Proof-Of-Concept (POC) experimental evaluations, comparing RRSN with state-of-the-art models, including POC experiments in real Internet topologies. The results demonstrate that RRSN outperforms existing schemes in network performance metrics such as throughput, jitter, packet loss ratio, and delay, showcasing its effectiveness in improved QoS for IoT networks.