Resource Profiling for Virtualized Radio Access Networks

The mobile networks are rapidly evolving, transitioning from specialized hardware to fully virtualized platforms that run on commercial off-the-shelf (COTS) hardware powered by general-purpose processors. Mobile operators are increasingly adopting Network Functions Virtualization (NFV) to take advantage of virtualization benefits, such as ease of deployment, flexibility, and cost savings. A crucial component of this transformation is the implementation of virtualized Radio Access Network (vRAN) solutions, which provide a cost-effective way to deploy the 5G and beyond RAN as a containerized network function (CNF) on COTS hardware. vRAN is more efficient than traditional RAN, as it allows multiple base station workloads to be multiplexed on the same hardware, potentially co-located with other workloads. However, ensuring consistent performance of vRAN workloads in this consolidated environment is challenging due to the high variability and unpredictability arising from contentions for shared system resources such as CPU cores, Last Level Cache (LLC), and Memory Bandwidth. Thus, there is a strong need to understand the vRAN components if they are affected by any system resource. This paper presents an understanding of how the next-generation scalable processors can be used to run the vRAN components and presents a detailed analysis of the controllable knobs such as CPU, LLC, Memory Bandwidth, and energy consumption. The proposed vRAN profiling shows that the observed throughput varies significantly based on the number of pinned CPU cores (e.g., up to $250 \text{Mbits} / \text{sec}$), and also similar performance observed with other system resources - LLC allocation and Memory Bandwidth.