Constraint-Aware Deep Reinforcement Learning for vRAN Dynamic Placement

The disaggregated and virtualized Radio Access Network (vRAN) is already present in 5G but tends to have increased adoption in 6G, mainly in the context of the Open RAN (O-RAN). Despite the potential benefits, the effective success of disaggregated vRAN depends on the efficient placement of Virtualized Network Functions (VNFs), which is influenced by the demand in the Radio Units (RUs). Several factors create dynamicity in this demand, but the number of users served by each RU is one of the most impacting. This problem has already been tackled in the recent literature, however, the works oversimplify important aspects that make their proposals inappropriate for real-world adoption. In this paper, we introduce a Deep Reinforcement Learning (DRL) agent that is constraint-aware, ensuring the solutions’ feasibility. We compare our DRL solution with existing optimization models and evaluate it under different scenarios, including the presence of Mobile Edge Computing (MEC) applications that compete for computing resources. Our contributions include a novel formulation, the implementation of a publicly available DRL agent, and insights into practical application scenarios for disaggregated vRAN optimization.