A Measurement-Based Power Consumption Model for Networks With Intelligent Control (RIC)

Open Radio Access Network (Open RAN) triggers a shift towards multi-vendor deployments with increased programmability, flexibility, and intelligence. Still, several critical questions remain unanswered, particularly regarding Open RAN's impact on power consumption. New interfaces and procedures, increased softwarization, real-time analytics, and optimization tasks introduce complexities whose effects on energy usage are currently unknown. Moreover, well-established power consumption models for RAN do not address the additional elements of Open RAN architecture, such as the RAN Intelligent Controller (RIC), the central element for near-real-time decision-making and network optimization. In this context, this paper presents the first evaluation of the power consumption of the RIC platform across diverse scenarios. Specifically, we address key questions about RIC's consumption by performing power measurements under varying workloads, assessing RIC's scalability, and the impact of different parameters on power consumption. Additionally, also for the first time, this work introduces a RIC power consumption model, which can be directly utilized to assess the system-level power consumption of RIC on different RAN site scenarios (macro, micro, and pico sites), with a mean absolute percentage error (MAPE) of only 0.54%. Interestingly we find that RIC's power consumption alone, even when operating only with its essential functionalities, may even exceed 14 kW while overseeing 300 sites, this is equivalent in terms of power to over 160 micro sites. In cases, RIC may even impose a power consumption overhead that exceeds 100% of the RAN power consumption. Additionally, when monitoring 100 heterogeneous network clusters (HetNets), this power may scale up to 60 kW, resulting in an overhead of over 30%. These conditions present significant challenges to the scalability of RIC and raise questions on the feasibility of achieving net energy savings through RIC under specific scenarios. Finally, we discuss and propose potential strategies for optimizing RIC energy efficiency in next-generation Open RAN systems.