ADDAPT6G: Advanced Double Dueling Architecture for Proactive Tuning in 6G Networks

The continuous advancement in Beyond 5G (B5G) networks demands innovative solutions to address the increasingly complex requirements of network service distribution, considering latency, bandwidth and throughput. In particular, the non-convex nature of the associated challenges makes them difficult to tackle as a single well-defined optimal solution can not be easily identified. This paper introduces a novel framework that employs Dueling Double Deep Q-Networks (Dueling DDQN) within a distributed architecture to optimize Cloud-native Network Function (CNF) migration and service distribution. By leveraging the cutting-edge capabilities of Deep Reinforcement Learning (DRL), our approach autonomously detects, predicts and rectifies placement inefficiencies, thereby enhancing the overall network service quality. Finally, we demonstrate the practical applicability of our proposed solution through testbed experimentation using a multi-node server setup to benchmark our system's capability to efficiently mitigate performance constraints within a distributed environment.