Energy-Efficient Baseband Function Deployments for Service-Oriented Open RAN

Recently open radio access network (Open RAN), which splits baseband functions into multiple process units at different locations has received considerable attentions from both industries and academia with the potential to enable a fully disaggregated RAN with more flexibility in delivering energy-saving and latency-sensitive applications. However, the significant increases of resource usage dynamics in both geographical and time and network complexity may lead to unnecessary high energy consumption in RANs without an efficient RAN function management policy. Many studies have proposed baseband function management solutions, however, the activation cost and data network resources of edge computing capacities have not been evaluated in much detail, as far as the authors know. In this paper, with the objective of minimizing energy consumption, meanwhile, satisfying the requests over the network under the constraints of latency and resource capacity, we propose a completed mixed integer linear programming (MILP) formulation, a multi-agent deep reinforcement learning-based algorithm and a heuristic (DCUH), to take user plane functions (UPFs) on the multi-access edge computing servers (MECs) and the activation consumption of MECs into consideration. Moreover, we prototype an OpenDaylight, OpenStack and Open Source Management and Orchestration-based Open RAN testbed to verify the feasibility of the proposed solutions. Results show the importance of hibernating the MEC after a certain time of network vacancy. DRL-based algorithm and DCUH can approach a similar performance as the benchmark of MILP and save more than 40% energy consumption compared to the first-fit algorithm. This study offers an important insight into the design of baseband deployment policies that greatly enhance user experience with better service and save Open RAN operational energy costs.