pith. sign in

arxiv: 2605.22638 · v1 · pith:RN2PXW6Znew · submitted 2026-05-21 · 💻 cs.NI

An intensive vRAN deployment with OpenAirInterface

Romain Beurdouche , Raymond Knopp This is my paper

Pith reviewed 2026-05-22 03:48 UTC · model grok-4.3

classification 💻 cs.NI
keywords vRANOpenAirInterface5Gvirtualized radio access networksoftware stack adaptationmulti-instance scalingperformance measurementcomputer architecture
0
0 comments X

The pith

Adapting OpenAirInterface enables scaling multiple vRAN instances on a shared server with measurable performance effects.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

This paper examines how to run an intensive deployment of virtualized 5G radio access networks using the OpenAirInterface software stack on standard computer hardware. The authors adapt the stack to better use processor and accelerator features while finding ways to pack several vRAN instances onto one server. They report the specific changes made, the resulting performance numbers, and observations about how different computer architectures behave under this load. A sympathetic reader would care because 5G networks need both high real-time performance and the flexibility to run on affordable general-purpose machines instead of fixed custom hardware. If these adaptations succeed, operators could increase network density without adding more physical servers.

Core claim

We adapted the OpenAirInterface software stack to leverage the capabilities of hardware and developed methods to scale the vRAN deployment with several instances sharing a server. We describe the improvements to the stack and their effect on performance, along with observations on how computer architectures influence the deployment and the remaining limitations that call for further work.

What carries the argument

The adapted OpenAirInterface stack configured for multi-instance vRAN scaling on shared general-purpose processors and hardware accelerators.

Load-bearing premise

The tested computer architectures and chosen scaling configurations are representative enough that the reported performance effects and architectural observations will generalize to other intensive vRAN deployments without substantial additional tuning.

What would settle it

Deploying the same multi-instance configuration on a different processor family or with a higher number of concurrent vRAN instances would reveal whether performance scales as reported or requires further software changes.

Figures

Figures reproduced from arXiv: 2605.22638 by Raymond Knopp, Romain Beurdouche.

Figure 1
Figure 1. Figure 1: channel coding time on the EP-RFSoC system with [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Architecture of the EP-RFSoC system with envi [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Distribution of time for channel coding of a slot, [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
read the original abstract

The advent of 5G virtualized Radio Access Networks (vRANs) brings a new challenge with regards to computer architectures. It requires to select or design computing technologies that provide a sufficient level of performance while maximizing the flexibility and efficiency of the implemented networks. Several solutions addressing this challenge were proposed, relying on general purpose processors as well as hardware accelerators. This work describes our effort to enable an intensive vRAN deployment using the 5G software stack OpenAirInterface on top of these computer architectures. We had to adapt the software stack to leverage the capabilities of hardware and to find how to scale up the vRAN deployment with several vRAN instances sharing a server. We describe in this work our improvements to the stack and their effect on performance. We also share our observations on the behavior of the computer architectures and how they affect our deployment. We finally discuss the limitations of our deployment and further efforts to implement better vRAN deployments.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The manuscript describes adaptations to the OpenAirInterface 5G software stack to support intensive vRAN deployments, specifically enabling multiple vRAN instances to share a single server. The authors detail modifications to leverage hardware capabilities of general-purpose processors and accelerators, report the performance effects of these changes, and share observations on how computer architectures influence such deployments. The work concludes by discussing limitations and suggesting further improvements for better vRAN implementations.

Significance. If the reported adaptations produce reproducible performance gains and the architectural observations hold beyond the tested setups, the work offers practical value for scaling open-source vRAN systems on shared commodity hardware. It addresses real deployment challenges in 5G virtualization and could inform efficient resource sharing strategies, though its significance depends on the strength of the empirical evidence and generalizability.

major comments (2)
  1. [§4 (Performance Evaluation)] §4 (Performance Evaluation): The central claim that the stack adaptations enabled scaling with measurable performance effects lacks quantitative metrics, baseline comparisons to unmodified OAI, error bars, or statistical details. This is load-bearing for validating the improvements and their impact.
  2. [§5 (Architectural Observations)] §5 (Architectural Observations): The observations on computer architecture behavior are drawn from a small number of tested configurations without systematic variation of CPU models, memory hierarchies, or accelerators. This undermines broader conclusions about intensive vRAN deployments.
minor comments (2)
  1. [Abstract] The abstract summarizes claims qualitatively but would be strengthened by including at least one key quantitative result or comparison.
  2. [§3 (Adaptations)] Notation for vRAN instance scaling parameters could be clarified with a table or diagram in the methods section.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the detailed and constructive review. We address the two major comments point by point below, indicating where revisions will be made to strengthen the empirical presentation while preserving the scope of the work.

read point-by-point responses

  1. Referee: [§4 (Performance Evaluation)] The central claim that the stack adaptations enabled scaling with measurable performance effects lacks quantitative metrics, baseline comparisons to unmodified OAI, error bars, or statistical details. This is load-bearing for validating the improvements and their impact.

    Authors: We agree that Section 4 would be strengthened by additional quantitative detail. The current manuscript reports measured effects on throughput, latency, and scaling when multiple vRAN instances share a server after our modifications. To address the concern, we will add explicit baseline comparisons against unmodified OpenAirInterface, include error bars on all performance plots, and report basic statistical measures (means and standard deviations across repeated runs) in the revised version. revision: yes

  2. Referee: [§5 (Architectural Observations)] The observations on computer architecture behavior are drawn from a small number of tested configurations without systematic variation of CPU models, memory hierarchies, or accelerators. This undermines broader conclusions about intensive vRAN deployments.

    Authors: The observations reflect the specific commodity platforms and accelerators available for our experiments, which are representative of current vRAN testbeds. We did not perform an exhaustive parameter sweep, as that would exceed the scope of demonstrating practical scaling on shared hardware. In revision we will explicitly list the tested CPU models, memory configurations, and accelerators, add a limitations paragraph clarifying the generalizability of the observations, and avoid language that implies universality. revision: partial

Circularity Check

0 steps flagged

No circularity: empirical deployment report with direct measurements

full rationale

The paper is an empirical description of software adaptations to OpenAirInterface for scaling multiple vRAN instances on shared servers, including performance measurements and architectural observations on tested hardware. No mathematical derivations, equations, fitted parameters, or predictions appear. Claims rest on direct testing and reported effects rather than any self-referential reduction to inputs. No self-citations form load-bearing premises, and the work is self-contained against external benchmarks of vRAN performance. This is the expected non-finding for a deployment-focused systems paper.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Review based on abstract only; no explicit free parameters, axioms, or invented entities are stated. The central claims rest on the implicit assumption that the chosen hardware platforms and OpenAirInterface modifications are sufficient to demonstrate general scaling behavior.

pith-pipeline@v0.9.0 · 5683 in / 1167 out tokens · 53201 ms · 2026-05-22T03:48:55.955647+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

56 extracted references · 56 canonical work pages

  1. [1]

    White paper, China Mobile Research Institute, April 2010

    C-RAN The Road Towards Green RAN. White paper, China Mobile Research Institute, April 2010

    work page 2010

  2. [2]

    White paper, O-RAN Alliance, October 2018

    O-RAN: Towards an Open and Smart RAN. White paper, O-RAN Alliance, October 2018

    work page 2018

  3. [3]

    White paper, NTT DOCOMO, June 2021

    5G Open RAN Ecosystem Whitepaper. White paper, NTT DOCOMO, June 2021

    work page 2021

  4. [4]

    Solution brief, AMD, January 2021

    MODERNIZE YOUR NETWORK TO MAXIMIZE 5G OPPORTUNITIES. Solution brief, AMD, January 2021

    work page 2021

  5. [5]

    Solution brief, AMD Xilinx, 2021

    Xilinx T2 Telco Accelerator Card. Solution brief, AMD Xilinx, 2021

    work page 2021

  6. [6]

    https://doc.slices-sc

    Slices-ri blueprint, 2023. https://doc.slices-sc. eu/blueprint/

    work page 2023

  7. [7]

    Solution brief, Lenovo Press, September 2023

    Virtual Radio Access Network Distributed Unit (vRAN DU) with Lenovo ThinkEdge SE455 V3. Solution brief, Lenovo Press, September 2023

    work page 2023

  8. [8]

    What is the difference between inline and lookaside ac- celerators in virtualized distributed units? White paper, Fujitsu, 2023

    work page 2023

  9. [9]

    White paper, AMD, May 2024

    4TH GEN AMD EPYC™ PROCESSOR ARCHITEC- TURE. White paper, AMD, May 2024

    work page 2024

  10. [10]

    White paper, AMD, March 2025

    5TH GEN AMD EPYC™ PROCESSOR ARCHITEC- TURE. White paper, AMD, March 2025

    work page 2025

  11. [11]

    https://gitlab.eurecom.fr/ oai/openairinterface5g/-/blob/2026.w06/ ci-scripts/yaml_files/sa_fhi_7.2_vvdn_gnb/ docker-compose.yml?ref_type=tags

    Docker compose file for the deployments of a 100MHz 4T4R numerology 1 FHI 7.2 gNB, February 2026. https://gitlab.eurecom.fr/ oai/openairinterface5g/-/blob/2026.w06/ ci-scripts/yaml_files/sa_fhi_7.2_vvdn_gnb/ docker-compose.yml?ref_type=tags

    work page 2026

  12. [12]

    OpenAir- Interface Documentation, February 2026

    OAI 7.2 Fronthaul Interface 5G SA Tutorial. OpenAir- Interface Documentation, February 2026. https:// gitlab.eurecom.fr/oai/openairinterface5g/-/ blob/2026.w06/doc/ORAN_FHI7.2_Tutorial.md

    work page 2026

  13. [13]

    OpenAirInterface Documentation, Febru- ary 2026

    OAI LDPC offload (O-RAN AAL/DPDK BB- DEV). OpenAirInterface Documentation, Febru- ary 2026. https://gitlab.eurecom.fr/oai/ openairinterface5g/-/blob/2026.w06/doc/ LDPC_OFFLOAD_SETUP.md

    work page 2026

  14. [14]

    https://advancedwireless

    Platforms for advanced wireless research, [Online; ac- cessed 17-April-2026]. https://advancedwireless. org/

    work page 2026

  15. [15]

    https://www.slices-ri

    Scientific large scale infrastructure for comput- ing/communication experimental studies, [Online; ac- cessed 17-April-2026]. https://www.slices-ri. eu/

    work page 2026

  16. [16]

    srsRAN Project website, [Online; ac- cessed 17-April-2026]

    srsRAN Project. srsRAN Project website, [Online; ac- cessed 17-April-2026]. https://www.srslte.com/ 5g

    work page 2026

  17. [17]

    OCUDU Ecosystem Foundation, [Online; accessed 22- April-2026].https://ocudu.org/

    work page 2026

  18. [18]

    https://www.amd.com/en/products/ adaptive-socs-and-fpgas/versal/rf-series

    AMD Versal™ RF Series, [Online; accessed 23- April-2026]. https://www.amd.com/en/products/ adaptive-socs-and-fpgas/versal/rf-series. html

    work page 2026

  19. [19]

    Product brief, AMD, [Online; accessed 23-April-2026]

    AMD VERSAL™ RF SERIES. Product brief, AMD, [Online; accessed 23-April-2026]. https://www.amd. com/content/dam/amd/en/documents/products/ adaptive-socs-and-fpgas/versal/rf-series/ versal-rf-series-product-brief.pdf

    work page 2026

  20. [20]

    3GPP.5G; NR; Multiplexing and channel coding, Jan- uary 2022

    work page 2022

  21. [21]

    openairinter- face5g

    OpenAirInterface Software Alliance. openairinter- face5g. OpenAirInterface Git Repository, Febru- ary 2026. https://gitlab.eurecom.fr/oai/ openairinterface5g/-/blob/2026.w06

    work page 2026

  22. [22]

    https://docs.amd.com/v/u/en-US/58268_ amd-epyc-8004-tg-architecture-overview

    AMD.AMD EPYC™ 8004 PROCESSOR ARCHITECTURE OVERVIEW, May 2024. https://docs.amd.com/v/u/en-US/58268_ amd-epyc-8004-tg-architecture-overview

    work page 2024

  23. [23]

    https://docs.amd.com/v/u/en-US/58462_ amd-epyc-9005-tg-architecture-overview

    AMD.AMD EPYC™ 9005 PROCESSOR ARCHITECTURE OVERVIEW, April 2025. https://docs.amd.com/v/u/en-US/58462_ amd-epyc-9005-tg-architecture-overview

    work page 2025

  24. [24]

    Open, Pro- grammable, and Virtualized 5G Networks: State-of- the-Art and the Road Ahead.Computer Networks, 182:107516, 2020

    Leonardo Bonati, Michele Polese, Salvatore D’Oro, Stefano Basagni, and Tommaso Melodia. Open, Pro- grammable, and Virtualized 5G Networks: State-of- the-Art and the Road Ahead.Computer Networks, 182:107516, 2020

    work page 2020

  25. [25]

    Agora: Real-time massive MIMO base- band processing in software

    Jian Ding, Rahman Doost-Mohammady, Anuj Kalia, and Lin Zhong. Agora: Real-time massive MIMO base- band processing in software. InProceedings of the 16th International Conference on Emerging Network- ing EXperiments and Technologies, CoNEXT ’20, page 232–244, New York, NY , USA, 2020. Association for Computing Machinery

    work page 2020

  26. [26]

    Redfish®

    DMTF. Redfish®. Redfish website, [Online; ac- cessed 17-April-2026]. https://www.dmtf.org/ standards/redfish. 13

    work page 2026

  27. [27]

    DPDK. 7. Intel® vRAN Boost Poll Mode Driver (PMD). DPDK Documentation, [Online; accessed 17-April-2026]. https://doc.dpdk.org/guides/ bbdevs/vrb1.html

    work page 2026

  28. [28]

    DPDK. 7. Linux Drivers. DPDK Documentation, [On- line; accessed 17-April-2026]. https://doc.dpdk. org/guides/linux_gsg/linux_drivers.html

    work page 2026

  29. [29]

    DPDK. 8. Wireless Baseband Device Library. DPDK Documentation, [Online; accessed 17-April- 2026]. https://doc.dpdk.org/guides/prog_ guide/bbdev.html

    work page 2026

  30. [30]

    Scientific large-scale infrastructure for com- puting/communication experimental studies, [Online; accessed 17-April-2026]

    ESFRI. Scientific large-scale infrastructure for com- puting/communication experimental studies, [Online; accessed 17-April-2026]. https://roadmap2021. esfri.eu/projects-and-landmarks/ browse-the-catalogue/slices/

    work page 2026

  31. [31]

    SLICES, a scientific instrument for the networking community.Computer Communica- tions, 193:189–203, 2022

    Serge Fdida, Nikos Makris, Thanasis Korakis, Raffaele Bruno, Andrea Passarella, Panayiotis Andreou, Bartosz Belter, Cédric Crettaz, Walid Dabbous, Yuri Demchenko, and Raymond Knopp. SLICES, a scientific instrument for the networking community.Computer Communica- tions, 193:189–203, 2022

    work page 2022

  32. [32]

    Concordia: teaching the 5G vRAN to share compute

    Xenofon Foukas and Bozidar Radunovic. Concordia: teaching the 5G vRAN to share compute. InProceed- ings of the 2021 ACM SIGCOMM 2021 Conference, SIGCOMM ’21, page 580–596, New York, NY , USA,

    work page 2021

  33. [34]

    Sutton, Pablo Serrano, Cristina Cano, and Doug J

    Ismael Gomez-Miguelez, Andres Garcia-Saavedra, Paul D. Sutton, Pablo Serrano, Cristina Cano, and Doug J. Leith. srslte: an open-source platform for lte evolution and experimentation. InProceedings of the Tenth ACM International Workshop on Wireless Network Testbeds, Experimental Evaluation, and Characteriza- tion, WiNTECH ’16, page 25–32, New York, NY , USA,

    work page

  34. [36]

    Scalable Distributed Massive MIMO Baseband Processing

    Junzhi Gong, Anuj Kalia, and Minlan Yu. Scalable Distributed Massive MIMO Baseband Processing. In 20th USENIX Symposium on Networked Systems Design and Implementation (NSDI 23), pages 405–417, Boston, MA, April 2023. USENIX Association

    work page 2023

  35. [37]

    FlexRAN™ Reference Architecture for Wireless Access, March 2023

    Intel. FlexRAN™ Reference Architecture for Wireless Access, March 2023. https://www. intel.com/content/www/us/en/developer/ topic-technology/edge-5g/tools/flexran. html

    work page 2023

  36. [38]

    FlexRAN, September 2025

    Intel. FlexRAN, September 2025. https: //github.com/intel/FlexRAN/tree/ c441ed3b3d692457428ee855e9e13d76b0a36167

    work page 2025

  37. [39]

    Intel® intrinsics guide, [Online; accessed 17-April-2026]

    Intel. Intel® intrinsics guide, [Online; accessed 17-April-2026]. https://www.intel.com/content/ www/us/en/docs/intrinsics-guide/index. html#avx512techs=AVX512_FP16

    work page 2026

  38. [40]

    Products formerly Sapphire Rapids Edge Enhanced

    Intel. Products formerly Sapphire Rapids Edge Enhanced. Product Specifications, [Online; accessed 17-April-2026]. https://www.intel.com/content/ www/us/en/ark/products/codename/235054/ products-formerly-sapphire-rapids-edge-enhanced. html

    work page 2026

  39. [41]

    Anuj Kalia, Nikita Lazarev, Leyang Xue, Xenofon Foukas, Bozidar Radunovic, and Francis Y . Yan. To- wards energy efficient 5G vRAN servers. InProceed- ings of the 22nd USENIX Symposium on Networked Systems Design and Implementation, NSDI ’25, USA,

    work page

  40. [42]

    Driving innovation in 6G wireless technolo- gies: The OpenAirInterface approach.Computer Net- works, 269:111410, 2025

    Florian Kaltenberger, Tommaso Melodia, Irfan Ghauri, Michele Polese, Raymond Knopp, Tien Thinh Nguyen, Sakthivel Velumani, Davide Villa, Leonardo Bonati, Robert Schmidt, Sagar Arora, Mikel Irazabal, and Navid Nikaein. Driving innovation in 6G wireless technolo- gies: The OpenAirInterface approach.Computer Net- works, 269:111410, 2025

    work page 2025

  41. [43]

    NVIDIA Aerial GPU Hosted AI-on-5G

    Anupa Kelkar and Chris Dick. NVIDIA Aerial GPU Hosted AI-on-5G. In2021 IEEE 4th 5G World Forum (5GWF), pages 64–69, 2021

    work page 2021

  42. [44]

    Chip choices kickstart open RAN war between lookaside and inline.Light Reading, August 2023

    Iain Morris. Chip choices kickstart open RAN war between lookaside and inline.Light Reading, August 2023

    work page 2023

  43. [45]

    Nvidia ai aerial

    NVIDIA. Nvidia ai aerial. NVIDIA devel- oper website, [Online; accessed 17-April-2026]. https://developer.nvidia.com/industries/ telecommunications/ai-aerial

    work page 2026

  44. [46]

    O-RAN Alliance.Control, User and Synchronization Plane Specification, October 2025

    work page 2025

  45. [47]

    O-RAN Alliance.Management Plane Specification, October 2025

    work page 2025

  46. [48]

    O-RAN Alliance.O-RAN Acceleration Abstraction Layer General Aspects and Principles, October 2025

    work page 2025

  47. [49]

    O-RAN Alliance.O-RAN Architecture Description, October 2025

    work page 2025

  48. [50]

    Benefits of Virtual- izing the Layer 1 in a RAN Stack

    Niall Power and Sindhu Xirasagar. Benefits of Virtual- izing the Layer 1 in a RAN Stack. White paper, Intel, 2022

    work page 2022

  49. [51]

    Programmable Millimeter-Wave MIMO 14 Radios with Real-Time Baseband Processing

    Zhenzhou Qi, Zhihui Gao, Chung-Hsuan Tung, and Tingjun Chen. Programmable Millimeter-Wave MIMO 14 Radios with Real-Time Baseband Processing. InPro- ceedings of the 17th ACM Workshop on Wireless Net- work Testbeds, Experimental Evaluation & Character- ization, WiNTECH ’23, page 17–24, New York, NY , USA, 2023. Association for Computing Machinery

    work page 2023

  50. [52]

    Savannah: A Real-time Programmable mmWave Baseband Processing Framework

    Zhenzhou Qi, Chung-Hsuan Tung, Anuj Kalia, and Tingjun Chen. Savannah: A Real-time Programmable mmWave Baseband Processing Framework. InProceed- ings of the 30th Annual International Conference on Mobile Computing and Networking, ACM MobiCom ’24, page 1736–1738, New York, NY , USA, 2024. Asso- ciation for Computing Machinery

    work page 2024

  51. [53]

    Savannah: Efficient mmWave Baseband Processing with Minimal and Heterogeneous Resources

    Zhenzhou Qi, Chung-Hsuan Tung, Anuj Kalia, and Tingjun Chen. Savannah: Efficient mmWave Baseband Processing with Minimal and Heterogeneous Resources. InProceedings of the 30th Annual International Con- ference on Mobile Computing and Networking, ACM MobiCom ’24, page 1500–1514, New York, NY , USA,

    work page

  52. [54]

    Association for Computing Machinery

    work page

  53. [55]

    Blueprint-based reproducible research with the slices research infrastruc- ture

    Damien Saucez, Sebastian Gallenmuller, Raymond Knopp, Nikos Makris, and Serge Fdida. Blueprint-based reproducible research with the slices research infrastruc- ture. InIEEE INFOCOM 2024 - IEEE Conference on Computer Communications Workshops (INFOCOM WKSHPS), pages 01–02, 2024

    work page 2024

  54. [56]

    Ayala-Romero, An- dres Garcia-Saavedra, Marco Fiore, and Xavier Costa- Perez

    Leonardo Lo Schiavo, Jose A. Ayala-Romero, An- dres Garcia-Saavedra, Marco Fiore, and Xavier Costa- Perez. YinYangRAN: Resource Multiplexing in GPU- Accelerated Virtualized RANs. InIEEE INFOCOM 2024 - IEEE Conference on Computer Communications, pages 721–730, 2024

    work page 2024

  55. [57]

    CloudRIC: Open Radio Access Network (O-RAN) Virtualization with Shared Heterogeneous Computing

    Leonardo Lo Schiavo, Gines Garcia-Aviles, Andres Garcia-Saavedra, Marco Gramaglia, Marco Fiore, Al- bert Banchs, and Xavier Costa-Perez. CloudRIC: Open Radio Access Network (O-RAN) Virtualization with Shared Heterogeneous Computing. InProceedings of the 30th Annual International Conference on Mobile Computing and Networking, ACM MobiCom ’24, page 558–572,...

    work page 2024

  56. [58]

    Demystifying 5g polar and ldpc codes: A comprehensive review and foundations.TechRxiv, 2025(0925), 2025

    Mody Sy. Demystifying 5g polar and ldpc codes: A comprehensive review and foundations.TechRxiv, 2025(0925), 2025. 15

    work page 2025