QoS-Aware Dynamic CU Selection in O-RAN

This technical report presents a framework for dynamic service function chain (SFC) provisioning in Open Radio Access Networks (O-RAN) utilizing graph-based reinforcement learning. The authors address the inefficiencies of static mappings between logical functions and physical locations in conventional RAN deployments, particularly under varying traffic and resource conditions. By formulating the problem as a Markov decision process and employing a graph neural network-assisted deep reinforcement learning approach named GRL-DyP, the report details how the proposed agent selects optimal routes and O-CU locations to minimize energy consumption while adhering to quality-of-service (QoS) constraints. The evaluation conducted on a dataset of 24-hour traffic traces from Montreal demonstrates that the dynamic selection and routing significantly reduce energy usage compared to static mapping, without compromising QoS. The findings suggest that DRL-based SFC provisioning can serve as an effective control mechanism for energy-efficient and resource-adaptive O-RAN implementations.