Network Slicing – Hyperscaler Reference Architecture

Introduction 

Network Slicing is a key feature of 3GPP proposed 5G specifications. As previous posts have discussed, network slicing enables the 5G network to offer business applications that offer various levels of services to customers. And this is a key part of network monetization. Let us consider a real-world architecture on AWS which builds on the generic architecture we covered a few posts ago – https://www.vamsitalkstech.com/5g/the-systems-architecture-of-network-slicing/. 

Network Slicing Architecture 

A hyperscaler oriented real-world representation of network slicing is shown below. 

Key points about the above network slicing architecture –

1. A network slice is an end-to-end logical network that can be dynamically created. The typical workflow is that User Equipment (UE) access network slice(s) over the same base station gNB. Each slice can offer an agreed type of service with its own SLA. 
2. There are two aspects of network slicing from a network standpoint – Core Slicing and RAN slicing. The RAN aspect deals with which segments to transmit and the Core deals with scaling various microservices and placing them on available compute instances.
3. The slice is first instantiated on a CU/DU via an activation command for a certain kind of business service which has various parameters pre-provisioned. The service itself is defined in the CU. To make the end-to-end slicing capability work, the RAN interworks with the Core from a slicing standpoint. In the case of AWS, this leverages core AWS services such as EC2, EKS, Lambda, CloudFormation etc running on Outposts servers while being orchestrated by either an AWS or 3rd party network orchestrator. 
4. As shown, the UE signals the Network and identifies the slice via the Single Network Slice Selection Assistance Information (S-NSSAI). The 3GPP currently allows up to eight network slices between the UE and the Network.
5. The goal of RAN slicing is to enable multiple containers CU/DUs placed on the available to compute instances based on scale needs and slicing policies for different applications. 
6. Given that multiple and concurrent NSIs (Network Slice Instances) can be deployed on common infrastructure, isolation is a key requirement in network slicing. This is from both a security and a performance guarantee standpoint. The lifecycle of an NSI should not impact or congest or compromise the security or performance of other NSIs. 

Conclusion

The next blog post will introduce the O-RAN alliance and its approach to network slicing. The alliance is a global community of network operators, vendors, and research institutions who aim to reshape the RAN industry into an open and fully interoperable ecosystem. 

Image source: Nery Montenegro