Employing Data Plan Programmability and Hardware Acceleration for Network Operations and Maintenance

The ever increasing necessity for modern computer networks to accommodate high-bandwidth and low-latency applications has led to increased complexity and dynamicity in modern network environments. The advent of the concept of data plane programmability and the relevant enabling tools, offered network operators the opportunity to customize and control the behavior of network devices at a granular level. In our research, we have focused our efforts on exploring the P4 language, a domain-specific language for programming network data planes, and NVIDIA Data Processing Units (DPUs), which are specialized hardware accelerators for network functions, to achieve our research goals of performing real-time network operations and maintenance. In this dissertation we show our work that covered various topics, starting from using P4 capabilities to extract and collect important telemetry data from network devices in multiple network environments, and then leveraging the collected data for various implementation. This includes our work on a multi-level closed-loop self configuring network in real-time fashion. We have also explored using the collected data to train AI models for the prediction of link failure in wireless environments, and combine that with power offered by NVIDIA DPUs for the detection of Distributed Denial of Service attacks on data-center networks, and to achieve end-to-end post-quantum encryption over optical links.