Power Consumption-Aware 5G Edge UPF Selection using Deep Reinforcement Learning

In this paper, we propose a Deep Reinforcement Learning (DRL) algorithm for the orchestration of an edge-cloud 5G core network deployment with multiple distributed User Plane Function (UPF) instances. More specifically, our DRL agent provides a policy for the UPF selection procedure when a new connection from a user to the data network is established. The choice of UPF is based on the real-time power consumption metrics gathered from the edge and cloud hosts in addition to the latency and bandwidth requirements of the users. The overall objective is to minimize the system power consumption while satisfying the desired quality of service. We show that the proposed algorithm performs better than other heuristics, resulting in lower power consumption and fewer errors in the latency requirements. Finally, we show the advantages of leveraging the power consumption data instead of CPU utilization metrics as often used in the literature.