Reinforcement Learning-Driven Service Placement in 6G Networks across the Compute Continuum

The advent of 6G networks promises unprecedented advancements in communication technologies, demanding innovative solutions for service placement across the Compute Continuum (CC), where computing resources are distributed across the network area, from edge to cloud. This paper explores a novel approach for service placement in 6G networks using Reinforcement Learning (RL) techniques. By leveraging the dynamic decision-making capabilities of RL, this work addresses the complexities of distributing services across heterogeneous computing resources to enhance network performance, reduce latency, and improve resource utilization. An extensive evaluation, based on a real dataset collected from commercial 4G/5G networks, was conducted under various network conditions and workloads to evaluate the effectiveness of the proposed approach. Results highlight the adaptability of our RL-driven model to dynamic network environments, demonstrated by its capacity to optimize for multiple objectives simultaneously. Our analysis also reveals that while single-objective heuristics can outperform RL in specific, limited scenarios, these struggle to handle increasing complexity. These findings highlight the viability of RL as a powerful tool for intelligent service management in next-generation communication systems, paving the way for more resilient and efficient 6G network architectures.