After reaping benefits from modernizing and disaggregating their core networks, communications service providers (CSPs) are shifting their focus to the radio access network (RAN). They want to bring the same agility, lower costs, and speed of deployment by virtualizing the RAN as they did with the core.
Due to the volume of cell sites in the RAN and the difference in size and service density, the platform requirements are different, especially for smaller sites at the far-edge.
Innovating in the RAN
VMware and Intel are working together to provide innovative solutions in the virtualized RAN for CSP all over the world. Leveraging VMware Telco Cloud Platform for RAN powered by 3rd Gen Intel Xeon Scalable processors, CSPs will have flexibility to run far edge RAN sites either with VM based CaaS or bare metal.
VMware and Intel have created a comprehensive lab to test vDU and vCU workloads on both VM based CaaSs and bare metal. This will give CSPs greater choice, flexibility, and feature agility when it comes to deploying their far edge RAN by running vDUs and vCUs directly on a bare metal CaaS while still offering the operational, security and life cycle management benefits customers trust VMware to bring.
Choosing the Right Combination
The choice between the two deployment options depends on the specific RAN site requirements of a CSP, such as performance, scalability, and management complexity. Here’s a general overview of when to use each option:
- VM based CaaS Option Co-hosting vDUs and vCUs from multiple vendors on the same physical server leveraging Intel Cache Allocation Technology (CAT) and Intel Resource Director Technology.
- A consistent cloud management layer to abstract hardware from the application when there are a variety of server types deployed.
- Support for advanced use cases such as high-availability, security, multi-tenancy, data recovery, and shared storage.
- CPU and memory resources readily available and accessible at both near and far edge sites to support the VMware VM based CaaS and Tanzu Kubernetes Grid (TKG) clusters.
Managed Baremetal CaaS Option
- In far edge or near edge sites, there is a need for deploying a single-node Kubernetes cluster to host vDUs or vCUs where the CPU and memory resources at these sites are limited.
- Both vDUs and vCUs are provided by the same network equipment provider (NEP), and therefore, can share the same operating system kernel providing visibility and access to the hardware.
VMware offers a comprehensive portfolio with VM based CaaS based, and a bare metal solution tailored to the requirements of diverse RAN sites, including far-edge locations. This empowers CSPs with unmatched flexibility and scalability, enabling them to adapt swiftly to changing demands.
VMware-Intel Collaboration Proof of Concept
VMware, in collaboration with Intel, have created a bare metal architecture solution, driving innovation with operational excellence at the edge. Leveraging VMware’s software and Intel’s hardware, CSPs now have flexibility to run far edge RAN sites either on a VM based CaaS or bare metal CaaS, based on the cell site requirements.
Our testing processes with Intel evaluates various metrics such as latency, efficiency, operational cost, optimized resource utilization, streamlined management and agility for these deployments.
Photon OS is the Linux distribution of VMware on the server and uses Intel Processors and NIC chipsets to complement the Managed Baremetal CaaS solution option. Our solution demonstrated a platform with a very small footprint to meet the needs of smaller cell sites.
What was tested?
VMware successfully validated running a single-node Kubernetes cluster on an Intel-based server to host vDUs and vCUs, along with radio units, in the VMware Telco Labs. VMware is also working with other NEP partners to host their vDUs and vCUs on both VM based CaaS and Intel-based servers, providing choice and flexibility in designing the RAN deployments.
The testing criteria focused on result metrics for the successful deployment and operation with PhotonOS. Key metrics included the ability to host a collocated instance of vDU and vCU, with TinyTKG deployed using fundamental K8s objects and resources, optimized for minimal resource consumption (pCPU, Memory, and Storage).
The installation time was completed within 2-3 minutes, ensuring efficient setup. Integration with hardware plugins through the K8s Device plugin framework, successful networking for vDU and vCU workloads, and dynamic interface attachment/detachment during LCM actions were accomplished. Access to persistent volumes on local storage and dynamic provisioning of local volumes proved crucial for data management. The availability and usability of required drivers and kernel modules post-installation ensured seamless functionality. Scheduling latency was kept below 10 microseconds for optimal performance. Meeting these result-oriented criteria ensured the effective deployment and functioning of PhotonOS with specific networking and storage capabilities for vDU and vCU workloads.
This PoC showcased the flexibility of VMware Telco Cloud Platform RAN with zero touch provisioning to orchestrate, monitor and verify the automated provisioning steps, including server discovery, firmware updates, OS (Operating System) installation, configuration management, and application deployment on either VM based CaaS or bare metal CaaS options. The infrastructure is powered by Intel’s 3rd Generation Intel ® Xeon ® Scalable Processors. In addition, CSPs are safe guarded with robust security features, coupled with comprehensive management tools to ensure the integrity and reliability of TCP RAN deployments.
Conclusion
CSPs want to bring the same agility, lower costs, and speed of deployment by virtualizing the RAN as they did with the core.
VMware and Intel are working to give customers more choice and flexibility when it comes to deploying their RAN by running vDUs and vCUs, either on a VM based CaaS or directly on a bare metal server.
By combining the best of both worlds, CSPs can securely and effectively manage their RAN while optimizing costs.
