A Reinforcement Learning Framework for Knowledge-Defined Networking

The deployment of 6G networks is expected to bring fundamental improvements to network architectures and a focus on incorporating artificial intelligence (AI) technologies. The paper explores the concept of knowledge-defined networking (KDN), where the intelligence of the network resides in the knowledge plane (KP), resulting from a combination of software-defined networking (SDN), network telemetry, and machine learning (ML) algorithms. The paper highlights the use of programming protocol-independent packet processors (P4), a technology which enables SDN networks, and emphasizes the importance of in-band network telemetry (INT) for providing real-time network information. The paper also draws a connection between P4-SDN network architecture and reinforcement Learning (RL), exhibiting how network components and existing techniques can be mapped onto RL principles. The potential of AI-driven network orchestration and the interpretation of networks as AI-based systems are also discussed.