Reliable and Trustworthy IoT Networks

The concept of the Internet of Things (IoT) has evolved significantly due to advancements like 5G communication, edge computing, and the proliferation of IoT devices. These breakthroughs have expanded the IoT market and accelerated adoption, especially due to the 2020 global pandemic, which pushed industries toward digital transformation using IoT solutions. However, IoT’s growth brings communication challenges with a multitude of devices and services demanding efficient and secure networks. Network monitoring is crucial to track and analyze traffic, enhancing reliability and trustworthiness, yet current methods struggle with IoT’s dynamic nature. This thesis introduces novel network monitoring frameworks to enhance IoT network reliability and trustworthiness, focusing on advanced traffic analysis and incident response in dynamic, low-latency IoT networks. Key technologies, such as machine learning (ML) models and programmable data planes, are incorporated for intelligent traffic monitoring and analysis. Four framework designs demonstrate how ML-based traffic analysis can boost IoT network reliability and trustworthiness, exemplified by attack defense services. These designs decentralize advanced analysis by shifting ML capabilities from the cloud to IoT network edges, offering flexible service configuration, faster analysis, incident response, and local privacy preservation. The first design emphasizes efficient traffic processing and ML-based analysis via big data engines in end-to-end network monitoring. The second design reduces latency with in-band feature extraction using programmable switches on IoT edge. The third design integrates traffic feature processing and ML inference within IoT gateways, enabling real-time threat defense and runtime reconfigurable service. To address privacy, the fourth design employs federated learning for local ML-based analysis, sharing only local models with network operators and service providers. Prototyping and evaluation in simulated networks and hardware devices validate the efficiency of these designs, with high analysis accuracy and millisecond-scale to microsecond-scale traffic handling speed. This thesis underscores the benefits of programmable data planes in flexible traffic monitoring and how ML elevates advanced traffic analysis across IoT network deployments. This approach safeguards traffic security and privacy, reduces operational frequency, and enhances network infrastructure sustainability, with opportunities for further exploration in use cases and system optimization for production deployment.