Ilmu Komputer & AI editorial
Open AccessOA2026
PRO-RAN: Processor-Level Characterization of Open RAN Centralized and Distributed Units
A controlled IMRAD digest of function-level CPU and microarchitectural profiling for disaggregated O-RAN CU and DU software
Moojan Kamalzadeh; Larry Horner; Linqi Xiao; Abhishek Bhattacharyya; Ehsan Bahaloo Horeh; Padmapriya Patil; Venkateswarlu Gudepu; Andrea Fumagalliยท 2026ยท DOI 10.48550/arXiv.2608.26498
The core problem
Open Radio Access Network (O-RAN) disaggregates RAN protocol functions and enables Centralized Unit (CU) and Distributed Unit (DU) software to execute on general-purpose computing platforms. Because CU and DU carry different protocol responsibilities, they produce different processor workloads and execution paths. Conventional performance metrics, including CPU utilization and throughput, quantify aggregate resource usage without identifying function-level execution costs or processor microarchitectural bottlenecks. Processor-level characterization, by contrast, provides insights into resource provisioning, function placement, software optimization, and hardware acceleration. The paper addresses this gap by describing a controlled characterization framework that evaluates independently deployed CU and DU functions under matched hardware and traffic conditions. The central premise is that aggregate utilization is insufficient for O-RAN deployment decisions: two functions may consume similar CPU percentages while stressing entirely different microarchitectural resources, such as front-end fetch, back-end execution, memory hierarchy, or branch prediction. The authors motivate function
Innovation
Under traffic load, accumulated process CPU time increases from 17.3 s to 37.4 s for the CU and from 462.0 s to 628.4 s for the DU during equal 300-s profiling intervals. These figures indicate that the DU consumes substantially more accumulated process CPU time than the CU across the measured intervals, and that both functions increase their CPU time under load. The CU grows from 17.3 s to 37.4 s, while the DU grows from 462.0 s to 628.4 s. The absolute difference between CU and DU CPU time is large: the DU baseline of 462.0 s is more than an order of magnitude above the CU baseline of 17.3 s. The measurements identify distinct CU and DU execution characteristics and motivate function-specific processor analysis and optimization. Because the intervals are equal at 300 s, the accumulated CPU time values are directly comparable within each function and across the two functions under the matched conditions.
Open Radio Access Network (O-RAN) disaggregates RAN protocol functions and enables Centralized Unit (CU) and Distributed Unit (DU) software to execute on general-purpose computing platforms. Because CU and DU carry different protocol responsibilities, they produce different processor workloads and execution paths. Conventional performance metrics, including CPU utilization and throughput, quantify aggregate resource usage without identifying function-level execution costs or processor microarchitectural bottlenecks. Processor-level characterization, by contrast, provides insights into resource provisioning, function placement, software optimization, and hardware acceleration. The paper addresses this gap by describing a controlled characterization framework that evaluates independently deployed CU and DU functions under matched hardware and traffic conditions. The central premise is that aggregate utilization is insufficient for O-RAN deployment decisions: two functions may consume similar CPU percentages while stressing entirely different microarchitectural resources, such as front-end fetch, back-end execution, memory hierarchy, or branch prediction. The authors motivate function-specific processor analysis as a prerequisite for efficient O-RAN infrastructure design.
The experimental platform integrates the Linux Foundation OCUDU implementation with an emulated User Equipment, a ZeroMQ-based radio interface, and an Open5GS core. Automated validation confirms registration and bidirectional packet delivery before process-scoped Intel VTune Hotspots and Top-Down Microarchitecture Analysis are applied. The framework enforces matched hardware and traffic conditions so that CU and DU measurements are directly comparable. Profiling is process-scoped, meaning that only the target CU or DU process is instrumented, avoiding contamination from unrelated system activity. Equal 300-second profiling intervals are used for both functions. The workflow can be summarized as follows:
Why it matters
The reported CPU time values expose a structural asymmetry between CU and DU workloads. A simple normalized comparison can be expressed as:
The DU's load-induced CPU time increase is roughly 8.28 times that of the CU, reinforcing that the two functions should not be treated as interchangeable workloads. Aggregate metrics such as CPU utilization and throughput would report resource usage without revealing which microarchitectural bottlenecks drive these differences. Top-Down Microarchitecture Analysis and Hotspots profiling are therefore positioned as necessary complements: they attribute execution cost to specific processor behaviors, informing resource provisioning, function placement, software optimization, and hardware acceleration. The framework's matched-hardware, matched-traffic, process-scoped design supports function-specific optimization rather than one-size-fits-all RAN deployment. The authors conclude that distinct CU and DU execution characteristics justify dedicated processor-level analysis for each function.
Who should read this
CS practitioners and researchers
Opening member contentโฆ