How to Deploy Open Source 5G SA Solutions

Having a private 5G SA network enables the creation of a highly scalable and resilient solution that supports various dedicated services such as IoT and automation. 5G core network services are widely available for installation in the open-source community. However, one of the most crucial aspects of implementation is ensuring that the solution meets enterprise requirements.

Performance testing is essential for the evaluation of throughput, scalability, latency, and reliability. It also ensures that customization meets industry-specific needs and competes with commercial solutions. These tests help confirm whether an open-source platform is a viable and efficient alternative to paid solutions and if it can be integrated with commercial radio access network (RAN) vendors.

In this article we’ll summarize the results of tests conducted by Software Mind and Airspan to confirm that an open source 5G core solution meets performance requirements for production systems.

Introduction to the PoC

To meet industry-specific requirements for data transfers between user equipment (UE) and core mobility elements, Software Mind decided to provide a proof of concept (PoC) solution to verify whether a successful implementation could be achieved based on an Open5GS project.

Software Mind partnered with Airspan, a recognized leader in Open RAN and end-to-end 5G solutions, to validate the integration of Open5GS with a commercial-grade RAN solution. This collaboration ensured that open-source core networks can effectively interoperate with carrier-grade RAN infrastructure

The NG-RAN and UE were isolated within a dedicated chassis to prevent interference with commercial services, while the Open5GS core elements operated on a single bare-metal server, with specified services exposed for integration. The PoC setup also included a network switch with 1 Gb/s interfaces, meaning all results were expected to remain below this throughput threshold. A simplified diagram is presented below.

Our first test scenario was to establish a connection between two UEs. The bitrate was halved because network traffic was shared between radio and network resources. Additionally, the packet round-trip time (RTT) also impacted the achieved data transfer rate relative to the expected bitrate level.

NG-RAN

During our tests, radio network coverage was confined to specialized enclosures, ensuring no interference with commercial cellular network providers. The antenna and gNodeB network element were supplied by Airspan.

To ensure a real-world deployment scenario, Airspan provided a fully integrated NG-RAN solution. The selected gNodeB model, AV1901, was configured with a 40/40/20 DL/UL/FL frame profile to test performance under commercial-grade conditions. (in default, DL- downlink, UL-uplink,FL- flexible) and 64 QAM DL/UL modulation.

User equipment (UE) in the radio wave isolation cabinet

5G core elements

The following core elements were provided to fulfil requirements: AMF, AUSF, BSF, NRF, NSSF, PCF, SCP, SMF, UDM, UDR and UPF. These elements form a complete 5G core network and enable full support for 5G services. The latest Open5GS 2.7.2 version was used.

All provisioning operations were set via Open5GS web UI.

One of our PoC requirements was to set all services, including user plane function (UPF) into a single bare metal server, thus we placed all 5G services into a single server with exposed services to integrate with NG-RAN.

Challenges

RAN Integration with 5G core services

At first glance, one of the potential challenges anticipated by our team was the integration of the RAN with 5G core services like AMF, SMF, and UPF. However, these services were seamlessly integrated with Airspan’s infrastructure, so we could focus on aspects like network throughput and latency.

TCP throughput limitations

During testing, we observed a TCP throughput limitation, where a single session was capped at 300 Mb/s. This issue, documented in Open5GS (GitHub issue #3306), was resolved in July 2024 through an update to packet buffer handling, which improved performance by 20%.

The specific code fix involved in modifying the packet buffer handling:

/*

sendbuf = ogs_pkbuf_copy(recvbuf);

if (!sendbuf) {

ogs_error(“ogs_pkbuf_copy() failed”);

return false;

}*/

sendbuf = recvbuf;</div>

This change resulted in a 20% performance gain, enabling throughput of up to 400 Mb/s on a single TCP session.

– RTT (Round-Trip Time) challenges
RTT proved to be another significant challenge, especially for applications requiring low latency. During our tests, we observed high latency between two UE devices while testing direct connection services between two smartphones over 5G. To mitigate the effects of high RTT, we realized it might be necessary to adjust the TCP buffers on the UE devices and identify the source of the high RTT within the network, which we successfully carried out.

– Unexpected network mask assignment
Unexpected behavior was the random network mask assignment to UEs. Although the IP addresses were correctly allocated from the defined address range, the network mask lengths assigned by Open5GS varied. This inconsistency could block communication between devices when not required. In this case, the client specifically requested open communication within a common APN, which highlighted the importance of addressing this issue.

– Radio profile
The radio profile is a crucial aspect that should be adjusted based on industry-specific needs. The spectrum is divided into uplink (UL) and downlink (DL) bands to facilitate efficient two-way data transmission. In the RAN configuration, you can define a profile that specifies the percentage of bandwidth allocated to DL, UL, and FL (flexible) parameters, ensuring that the spectrum is used for designated purposes. Generally, the DL parameter is the most critical for UEs.

– UPF test insights
Our tests revealed that the UPF implementation in Open5GS appears to operate in a single-threaded mode, making the choice of CPU (processor generation, clock speed, etc.) crucial. For broader commercial applications, deploying multiple UPF instances is essential to meeting network performance demands.

Results

Thanks to well-defined APIs, integrating open-source and commercial products in 5G networks is a straightforward process and a significant advantage. Whether using commercial or open-source solutions, organizations can achieve new levels of cost efficiency while simultaneously addressing their business requirements.

Throughput based on a radio frame profile:
Local speed test results using the radio frame profile (40/40/20 DL/UP/Flexible).

Local iperf3 using frame 40/40/20

Local speed test result using frame 70/20/10 

Local iperf3 result using frame 70/20/10

Video streaming was also possible, even in 4K resolution. Below you can see results presented with ping to UPF’s gateway while watching a video. 

Based on the resources available in the PoC setup, the 1 Gb/s limitation at the network switch level, and the applied radio modulation, our team fully saturated the network resources in terms of throughput.

To further enhance 5G network performance, it might be worth considering a switch to alternative open-source UPF solutions, such as UPG-VPP UPF (https://github.com/travelping/upg-vpp), eUPF (https://github.com/s5uishida/install_eupf), or free5GC UPF (https://github.com/free5gc/upf).

The number of parallel sessions also influences the maximum throughput between UEs (within a shared radio spectrum). Implementing multisession setups and defining the number of threads or UPF instances impacts how quickly the core network element can switch between sessions to ensure optimal data transfer conditions.

Below you can find an example result from multisession tests, where a single UPF instance was used, and UEs established multiple TCP connections to measure available throughput.

In this scenario, the total available throughput was approximately 200 Mb/s. One of the factors affecting this result was the context switching of the UPF thread between TCP sessions, and the shared allocation of resources among UEs.

In another scenario, the UE was configured to communicate with the Internet during a speed test, with only a single TCP session established. The throughput for this single TCP session exceeded 400 Mb/s. Below is a chart presenting the test results, prepared based on captured packets (pcap) and analyzed using Wireshark. 

Looking at the data, it is evident that there are specific areas where network performance can be optimized. A more advanced switch, enhancing TCPbuffers for more stable packet transmission or trying to apply different UPF implementations may have an important impact on network scalability.  

Open-source project advantages

One of the greatest advantages of 5G technology is the standardization of interfaces, which enables the integration of components from various vendors. This flexibility makes it possible to select solutions that best meet project requirements – in terms of performance and cost-efficiency – and replace inefficient network elements with modern, more functional alternatives.

Open-source projects can serve as cost-effective alternatives that integrate well with commercial products, such as NG-RAN, including Airspan’s RAN. While this approach is attractive, it is important to recognize that open-source projects may come with significant challenges due to functional limitations or missing features necessary to meet business requirements.

During our PoC project, we developed a comprehensive solution that addressed industry-specific needs related to communication flows, network throughput, and overall performance. Integrating the 5G core network with NG-RAN, optimizing TCP throughput and latency, and reducing high round-trip time (RTT) presented substantial challenges throughout the process. Software Mind successfully tackled these challenges by implementing tailored technical solutions and optimizing network architecture.

Our team focused particularly on ensuring comprehensive coverage for user equipment communication scenarios, minimizing delays, and maximizing data throughput to meet expected service quality standards. As a result, our team completed the project successfully, delivering a stable and efficient network infrastructure that fulfilled the requirements of our business case.

Therefore, for a successful commercial launch, it is essential to collaborate with an organization that is open and willing to effectively address challenges in a cooperative manner. This successful PoC demonstrates how integrating Open RAN and open-source core networks can provide cost-effective, scalable 5G solutions. If your organization is looking to explore open-source 5G deployments with carrier-grade reliability, contact us to discuss how we can support your project.