AI-Native Disaggregated RAN: Project Sylva Reference Architecture with SUSE Telco Cloud and OCUDU RAN stack

AI-Native Disaggregated RAN: Project Sylva Reference Architecture with SUSE Telco Cloud and OCUDU RAN stack Blog post authored by: Nils Fuerste, DevOps Engineer at Software Radio Systems Alberto Morgante, Principal Telco Engineer at SUSE Introduction Communication Service Providers (CSPs) are transitioning to open, disaggregated, and AI-native network architectures to reduce costs, eliminate vendor lock-in, and accelerate innovation in 5G and 6G services. This shift requires tight integration between carrier-grade infrastructure platforms and high-performance RAN software that can run efficiently on general-purpose hardware. This joint reference architecture documents the technical integration of the OCUDU RAN stack with SUSE Telco Cloud, aligned with Project Sylva, enabling deployment of RAN with AI-native workloads. Validated in collaboration with Software Radio Systems (SRS), the technical contributor of OCUDU, it provides a single-node deployment model using Cluster API (CAPI) for hardware readiness validation and resource sizing. The setup focuses on an O-RAN Split 7.2 gNB configuration with an RU emulator, enabling fully software-based testing and repeatable performance measurements. It demonstrates production-grade performance characteristics including low latency, 1,000+ users per cell, and hardware efficiency gains of 30-50% CAPEX reduction compared to proprietary systems. Market Challenges The telco industry continues to operate under significant structural fragmentation. Most 5G deployments still rely on proprietary, vertically integrated hardware and software stacks from a small number of vendors. This approach creates deep vendor lock-in, inflates both capital and operational expenses, and limits operators’ ability to select best-of-breed components. Operators face three interconnected problems. First, the lack of standardized reference architectures forces custom, site-specific integration work for every new deployment, binding CSPs to individual vendor roadmaps and increasing complexity. Second, existing open source telco infrastructure has historically fallen short of the extreme performance, deterministic latency, and resiliency requirements demanded by modern RAN workloads, especially those incorporating AI. Third, managing distributed edge environments with consistent operations remains difficult when each site uses different hardware platforms and deployment processes. These challenges directly constrain operators. They result in prolonged deployment timelines, higher lifecycle costs, reduced innovation velocity, and slower delivery of new AI-native 5G and future 6G services that require sub-millisecond latency and high user density. Without a unified, horizontal foundation, operators struggle to scale efficiently, control their infrastructure choices, or fully capitalize on the business opportunities of edge computing and programmable networks. Also, a key gap has been the lack of validated, carrier-grade Open RAN implementations that operators can confidently deploy at scale. The Solution SUSE and SRS have jointly developed a fully disaggregated, carrier-grade Open RAN stack anchored in the Project Sylva reference architecture. This solution was demonstrated at MWC 2026, validating its readiness for real-world telco deployments. The stack is structured as a horizontal, integrated platform that begins with general-purpose hardware and extends through the infrastructure layers to the RAN workloads. At the foundation lies high-performance commodity hardware featuring Intel® Xeon® 6 SoC processors and Supermicro edge servers. This choice delivers the necessary compute density and efficiency while avoiding proprietary silos, enabling significant improvements in power consumption and server utilization. Above the hardware, SUSE Telco Cloud provides the operating system and Kubernetes infrastructure layer. SUSE Linux Micro (SL Micro), an immutable and real-time Linux distribution, combined with RKE2, creates a hardened, Sylva-compliant Container-as-a-Service platform. This layer ensures deterministic low-latency performance, CPU isolation, SR-IOV networking with DPDK, and NUMA-aware resource management, all critical for meeting the stringent timing and throughput requirements of RAN workloads. SUSE Telco Cloud architecture SUSE Rancher Prime supplies unified orchestration and management across distributed edge sites. It enables centralized control, consistent policy enforcement, and zero-touch operations through declarative GitOps workflows and Cluster API (CAPI), allowing operators to manage hundreds of edge locations with a single operational model. At the workload layer, OCUDU delivers AI-native, software-defined Central Unit (CU) and Distributed Unit (DU) network functions. Designed as a cloud-native implementation of the 5G RAN stack (supporting O-RAN Split 7.2), OCUDU brings intelligence directly into the RAN, incorporating native design elements for AI/ML-driven optimization, scheduling, and performance enhancement. This enables operators to run intelligent RAN workloads directly on the SUSE Telco Cloud platform with consistent, carrier-grade behavior. As a foundational component for AI-RAN, OCUDU allows seamless integration of real-time AI capabilities into the RAN stack, accelerating the transition toward intelligent, adaptive 5G and 6G networks. Together, these components create a complete horizontal solution anchored in the Project Sylva reference architecture and delivered through SUSE Telco Cloud. By using Kubernetes as the interoperability layer where telco software converges, the stack fully decouples RAN software from proprietary hardware. This architecture directly addresses the fragmentation and lock-in challenges outlined earlier by providing operators with a standardized, portable, and consistent operating model. It eliminates the need for site-specific custom integrations, dramatically reduces deployment times from weeks to minutes, lowers CAPEX through hardware efficiency, and enables true “Develop Once, Deploy Anywhere” portability, all while delivering carrier-grade performance at scale. Architecture Overview and Deployment This reference architecture defines a single-node deployment designed for hardware validation, performance benchmarking, and resource sizing of AI-native RAN workloads. It integrates OCUDU’s gNB implementation with SUSE Telco Cloud, following Project Sylva principles for a consistent, cloud-native telco stack. The architecture follows O-RAN Split 7.2, with all CU and DU functions consolidated in a single gNB pod running alongside an RU emulator. This configuration enables complete software-based testing of the DU-to-RU path over the Open Fronthaul interface without requiring physical radio hardware. The diagram above shows how OCUDU maps the 5G RAN protocol stack onto O-RAN logical components. The blue area represents the OCUDU implementation, while the O-RU is shown as the external radio unit connected through the fronthaul interface. In this reference architecture, Split 7.2 defines where the DU connects to the O-RAN Radio Unit over the O-RAN Open Fronthaul interface. OCUDU provides the CU and DU-side functionality shown in the blue area of the diagram, including O-CU-CP, O-CU-UP, O-DU-high, and O-DU-low. The O-RU sits outside OCUDU and exchanges fronthaul traffic with the DU using the Split 7.2 interface. - O-CU-CP: Provides the control-plane part of the CU, including RRC and PDCP-C. It connects to the 5G Core control plane over NG-C/N2, to the O-CU-UP over E1, to the O-DU over F1-C, and optionally to the near-RT RIC over E2. - O-CU-UP: Provides the user-plane part of the CU, including SDAP and PDCP-U. It connects to the 5G Core user plane over NG-U/N3 and to the O-DU over F1-U. - O-DU-high: Contains the higher DU layers shown in the diagram, including RLC and MAC. - O-DU-low: Contains the lower DU processing shown in the diagram. For Split 7.2, this is the OCUDU component that connects to the O-RU over the O-RAN Open Fronthaul interface. - O-RU: Represents the O-RAN Radio Unit. It sits outside OCUDU and connects to the DU over the Split 7.2 fronthaul interface. The diagram also shows Split 6 and 8 as an alternative lower-layer split option. This blog post focuses on Split 7.2, where the DU-to-RU connection uses the O-RAN Open Fronthaul model. OCUDU RAN Workloads | Component | Purpose | Sizing Guidance | |---|---|---| | ocudu-gnb | Runs the OCUDU gNB deployment variant. In this blog post, this includes the CU-CP, CU-UP, and DU functionality in a single monolithic application. | Size according to the selected RAN configuration. Bandwidth, number of cells, antenna configuration, MIMO layers, numerology, enabled features, and traffic profile all influence CPU, memory, hugepage, and network requirements. | | ru-emulator | Provides an emulated O-RAN Radio Unit for lab and validation scenarios where physical RU hardware is not used. | Size according to the fronthaul configuration and the amount of generated or processed radio traffic. It can be co-located with the gNB for simple lab setups or placed on separate hardware for more realistic deployments. | | linuxptp | Deploys linuxptp components used for clock synchronization in deployments where DU and RU timing must be aligned, such as multi-node or physical O-RU scenarios. | Lightweight compared to the RAN workload. Resource usage is usually minimal, but the configuration must match the selected NIC, PHC, and synchronization design. | Test Mode Capabilities OCUDU test mode exercises the complete software stack without requiring real UEs or RUs. Depending on the test use case, OCUDU can be tested with a physical radio unit or with the radio side fully emulated in software. This makes it useful for both lab validation and performance testing before integrating physical RAN hardware. In this reference architecture, we use the OCUDU RU emulator to fully exercise the OCUDU RAN stack and produce meaningful results in a controlled software-based environment. For the full technical reference documentation, please visit the SUSE Documentation portal here. Conclusion and Next Steps This reference architecture demonstrates how the OCUDU RAN stack can be deployed on SUSE Telco Cloud for hardware validation and RAN workload sizing. The measured resource requirements serve as key input for planning and sizing production deployments. The setup functions as a validation environment that confirms whether the underlying platform delivers sufficient performance and correct configuration for a chosen RAN profile. The deployment validates several key infrastructure capabilities required for carrier-grade Open RAN: - Declarative provisioning using CAPI and Metal3 to build the cluster through an infrastructure-as-code workflow. - High-performance networking using SR-IOV Virtual Functions and DPDK for direct, low-overhead packet processing. - Real-time readiness using the PREEMPT_RT kernel, CPU isolation, hugepages, and tuned profiles to reduce jitter and improve determinism. - O-RAN Split 7.2 validation by exercising the DU-to-RU path through the O-RAN Open Fronthaul interface using the RU emulator. - Comprehensive observability through logs, metrics, and dashboards to identify configuration issues, resource bottlenecks, timing problems, and performance limits. In the single-node scenario, the OCUDU gNB and RU emulator run on the same host. This makes the setup particularly effective for validating CPU allocation, memory configuration, hugepages, SR-IOV device assignment, DPDK configuration, and overall platform readiness without requiring a separate physical RU. The same validation approach can be adapted to different scenarios, including smaller functional tests, larger benchmark systems, dual-node DU/RU deployments, or multi-cell configurations. The critical consideration is that the RAN configuration and hardware profile must be evaluated together. Factors such as bandwidth, MIMO layers, antenna configuration, traffic model, and enabled features directly determine the required compute, memory, and I/O capacity. By combining SUSE Telco Cloud’s robust, Sylva-compliant infrastructure with the OCUDU AI-native RAN stack, operators gain a practical path to overcome ecosystem fragmentation. This joint solution delivers standardized operations, reduced deployment complexity, lower infrastructure costs, and the flexibility to scale AI-native 5G and 6G services across diverse edge environments. Ready to get started? Contact the SUSE Telco team today to schedule a dedicated technical workshop, review the OCUDU Getting Started Guide, or initiate your own validation deployment with OCUDU RAN on SUSE Telco Cloud. Authors: Nils is a DevOps Engineer at Software Radio Systems (SRS), specializing in the development and management of reliable infrastructure for mobile network systems. Since joining SRS in 2021, he has focused on mastering Kubernetes, streamlining deployments, and providing high-level customer-facing engineering support. With a deep-rooted background in IT Security, Nils brings a security-first mindset to infrastructure management. His previous work in mobile security research includes investigating fake base station detection, SS7/Diameter vulnerabilities, and SIM-based attacks, as well as conducting exploit testing on mobile phones and core networks. Alberto serves as a Principal Telco Engineer at SUSE, where he plays a pivotal role in the advancement of telecommunications infrastructure. He is currently spearheading key telco projects focused on the architecture and development of a next-generation telco cloud platform, built upon the highly scalable SUSE Telco Cloud stack. Related Articles Sep 03rd, 2024 Enhancing Security with Confidential Computing: Use Cases Jun 16th, 2025 Are Open Source Kubernetes Monitoring Tools Better? May 28th, 2025