We have discussed 3GPP standards for 5G NR (New Radio) https://www.vamsitalkstech.com/5g/5g-new-radio-nr-platform-architecture/ and 5GC (Core) https://www.vamsitalkstech.com/5g/5g-core-5gc-platform-architecture/ architecture. In this post we introduce a software-defined reference architecture based on a hyperscaler deployment model.
5G mobile broadband network operators will need to deploy hundreds to thousands of infrastructure elements. These range from radio networks to core functions. All of these services will need to operate under strict SLAs and requirements for latency, uptime and performance. The goal in moving 5G functions to software is to not only ensure that these requirements can be met but also to ensure seamless management of these functions across various layers – edge, access, and core data networks. For the rest of this blog we consider all these networks functions as CNFs (Container Native Functions).
The above illustration depicts the target 5G architecture. Let us consider the salient points –
- At the bottom are cell sites and other sites where RAN workloads are running. These CNFs are CUs and DUs. The DU runs right at the edge and maybe deployed on small servers running x86 hardware or ARM-based chipsets.
- User Plane functions can be deployed in telco colos/ data centers or specialized hardware servers such as AWS Outposts
- Central control plane 5G core functions, certain user plane functions, OSS/BSS functions as well as the mobile switching office (MSO) or central data center run in the central data center. However, some of these functions can also be pushed to downstream layers
- Container primitives such as load balancing, healing of CNFs, automation are carried out using Kubernetes semantics
- Security boundaries such as limiting blast radius etc are handled using the distribution capabilities e.g Amazon EKS limits blast radius among worker pods using cluster boundaries
- Several new requirements are introduced into this platform that are not present in standard enterprise application deployments. These include support for bare metal deployments, high-performance kernel features,
To break this out into numbers, a typical CSP will deploy hundreds of thousands of network services as virtual network functions (VNFs) or CNFs. VNFs will run in virtual machines that can optionally be run inside a Kubernetes distribution. These can be provisioned, deployed, and load-balanced across many different geographical locations based on performance and throughput requirements. The next few posts will cover the demands imposed on Kubernetes by 5G deployments.