A laboratory comparison of two 5G radio access network stacks found little separation in the time needed to establish a connection, but a clearer advantage for srsRAN in throughput and real application workloads. The result comes from one Docker-based 5G standalone testbed, so it describes how the platforms behaved in that setup rather than setting a universal ranking.
Connection setup was close
For RRC, the connection-setup procedure, the best reported average was 87.7 milliseconds for both OAI and srsRAN in configuration C1. The averages were 88.5 ms in C3 and 91.9 ms in C2. The best isolated result was OAI using a USRP B210 software-defined radio in C3, at 85.3 ms; the second-best result was just 1.65% higher.
The weakest reported case was OAI with a USRP N310 in C2, whose average setup time was 15.2% above the best result. The configurations otherwise showed similar behavior on this measure. Researchers used four real user devices and ran 10 connection-and-disconnection cycles per device, with a script coordinating airplane-mode changes.
The larger gap appeared in throughput
The throughput tests measured end-to-end capacity between the user devices and a server with iPerf3, in both downlink and uplink. Each test lasted 100 seconds, with results reported at one-second intervals. Across the tested cases, downlink reached about 70% of the expected theoretical capacity, while uplink reached only 5% to 49%.
The comparison shifted with the hardware and configuration. In C1, srsRAN's downlink throughput was 18% to 24% higher than OAI's on both the B210 and N310. In C2, OAI was 12% higher with the N310 but 7% lower with the B210. In C3, OAI reached only 45.7% of srsRAN's throughput with the B210 and 43.8% with the N310. srsRAN was generally more consistent, apart from an unstable B210 C3 uplink test that could not be completed.
Application tests favored srsRAN
The differences were also visible in application workloads. In the video-on-demand test, four devices downloaded about 60 ten-second video chunks each. In C1, OAI required more than 2.8 seconds for half of its chunks to load, while srsRAN's median loading time was around 1.5 seconds and its performance was more stable under concurrent load. C2 was the most balanced configuration, and srsRAN had more consistent loading times in C3.
Live-streaming latency generally favored srsRAN, although the size of the difference depended on the configuration. OAI's latency was 33.91% higher than srsRAN's in C1, and OAI lost its connection with the B210. With the N310 in C3, OAI averaged 485.09 ms, compared with 113 ms for srsRAN with the B210 and 119 ms with the N310. In C2, the N310 means were close: 108 ms for OAI and 109 ms for srsRAN.
The cloud-gaming test used three devices simultaneously playing Brawlhalla in offline mode. Across its configurations, srsRAN averaged 24.06 ms of latency; its best case was 23.54 ms with the N310 in C2, alongside 5.31 ms of mean jitter, a measure of variation in delay. OAI showed unstable behavior with high jitter and outliers, although frame drops were close to zero.
A result tied to the testbed
The testbed deployed OAI and srsRAN in Docker on an x86 server, used 3GPP functional split Option 8, and connected real devices through USRP Ettus B210 and N310 radios. The radio evaluation held modulation at 256-QAM and subcarrier spacing at 30 kHz, and compared 20 MHz and 40 MHz one-layer bandwidths, with 2x2 MIMO at 20 MHz.
The authors' overall reading was that OAI had the best RRC result, while srsRAN came closest to the theoretical downlink figures and performed best across the real application workloads. They also highlighted OAI's support for 120 kHz subcarrier spacing in the FR2 range and srsRAN's more structured documentation.
These results are tied to the reported server, software environment, SDRs, fixed radio settings, selected workloads and small number of real devices used here. The authors present the comparison as a within-testbed assessment, so the observed differences should not be treated as a universal performance ranking.
Funding and status
The document is marked arXiv version 1 and dated 26 August 2026, making it a preprint. The work was supported by Motorola Mobility, a CNPq Research Productivity Fellowship under grant 313083/2023-1, and FACEPE under grant IBPG-0130-1.03/23.
Paper data and sources
Original title: Open-Source 5G RAN Platforms: A Dual Perspective on Performance and Capabilities
Authors: Maria Katarine Santana Barbosa, Iasmin Gomes, Vinícius Melo, Kelvin Lopes Dias
Journal/Repository: 10.1109/WCNPS69127.2025.11295920
Status: Preprint, not yet peer-reviewed
First online: 2026-08-26
DOI: Not available
Original paper · Full text