AWS has launched a new type of its Outposts RAN server, with a form factor and performance designed especially for on-premise deployments at RAN sites.
The company’s Fabio Cerone, EMEA Director, Telco, AWS, said the servers were a continuation of the company’s cloud continuum approach, which sees it use the same EC2 bare metal technology across its central, regional and edge cloud products. It also bundles in the AWS CaaS equivalent – Elastic Kubernetes Service – along with synchronisation technology.
AWS’ Outposts RAN server uses its Graviton 3 chip to handle Layer 2 and above RAN software, with L1 being handled on what Cerone terms a “bring your own” basis (* see Update). “The AWS value proposition is to have a fully managed solution, taking care of the pre-integration of that complexity.” Cerone said.
Hyper fragmentation for the sake of avoiding lock-in is a legend
Cerone said that the cloud vendor has taken this approach to try and unlock the Cloud RAN market for brownfield operators – it announced that Telefonica, Orange and Du would all be taking a look. Although Vodafone is about to announced the results of its latest RAN tender, operators such as Orange and Telefonica are still evaluating their options. The level of integration and the ability to managed a distributed cloud or virtual RAN rollout is one of the key items that operators must assess. That is why AWS is among those who have decided that what the industry needs is a pre-integrated approach that also takes care of much of the lifecycle management (LCM) for operators.
For AWS, the BYO L1 approach has led AWS to Nokia in the first instance. That’s not surprising as it is a continuation of the work the two companies have carried out since Nokia launched its AnyRAN DU accelerator NIC two years ago. Nokia’s AnyRAN takes the shape of a pluggable card that can slot into a server chassis to carry out inline L1 DU acceleration using an Arm-based Marvell SoC.
The Nokia SmartNIC pictures at launch in 2023.
Other vendors do not handle the L1 acceleration in a similar way, with Ericsson and Samsung both using Intel’s Xeon platform to handle acceleration in a lookaside manner. The latest two versions of Xeon SoC – releases four and six – handle the hardest part of the L1 acceleration, Forward Error Correction, within Intel’s integrated vRAN Boost feature.
Some see that approach as creating a hardware dependency with Intel, as it ties the vRAN to the underlying Intel platform via that integrated FEC acceleration. Others see Nokia’s approach as creating a dependency on its inline SoC-based SmartNIC.
For AWS to line up other BYO vendors they would need to have ported RAN software to work in the Arm-based Graviton CPU environment. Cerone said that AWS is having “architectural conversations” with other vendors. “Yes we have found a fit with Nokia, but I fully believe this is a significant step in radio network evolution and we are having conversations with all the others.”
Avoiding fragmentation/ avoiding lockin
Any operator deploying the AWS server will also be handing over the management and control of those Kubernetes deployments to AWS. To date, most have used CaaS from the likes of Wind River and Red Hat to manage their distributed vRAN deployments.
For Cerone, that’s not a problem. “Hardware, OS, Kubernetes layer, this is something we do ourselves. It comes into the operator network in a pre-integrated format. You don’t need to take care of that system integration, lifecycle management, all the different layers. You don’t buy the four different layers from different players, you’re not just buying a piece of hardware, but a fully managed solution.”
For Cerone, “Hyper fragmentation for the sake of avoiding lock-in is a legend. If you take it to extremes it just brings additional complexity and cost and not much operational efficiency.
“This is part of the debate we have had forever. In the end you need to be pragmatic – and here that means to make sure we make available the solution that addresses TCO and LCM, and gives freedom to the operator to select the network function software vendor they want.”
Cerone’s words mirror those of Rakuten CTO Raghunath Hariharan. Hariharan said that Rakuten Mobile had “swung too far out on the disaggregation pendulum”, and needed to come back a little bit with a more pre-integrated solution.
“What we learnt is that there is some amount of re-integration that has to happen. Not like a closed RAN but something intermediate. We were able to bring our cloud which integrates all the Kubernetes plugins, all the timing and synch, all the aspects needed for networking and bring that to the market. Other adopters of Open RAN are seeing that as a benefit and beginning to offer similar solutions.”
One of the operators that has been working with AWS is Orange. Orange’s VP Green & Radio Networks, Elisabeth Py, said, “We are not going into Open RAN to have vendor lock-in on the hardware, chipset or whatever. So we are very careful for that.
“We are considering other kind of chipsets – for example the Arm-based Graviton solution with Nokia and AWS. So we’ll leave our options open.”
Is the Nokia SmartNIC approach itself a hardware dependency?
“It could be. But it has its advantages also. In principle what we would like is to be able to change the CaaS without swapping the hardware, swap the CNF without swapping everything else underlying. But we’ll see. This is what we are thinking.”
That approach would not, it should be noted, accord with the AWS bundled approach, which does indeed include the CaaS (EKS) as part of the service.
For Cerone, such debates are likely to be frustrating. “We can talk about incremental steps in this industry forever, or we might start thinking about things in a different way. We welcome constructive friction and are always ready to disconfirm when the data demonstrates we are wrong. Our thinking is that a fully managed solution brings a level of flexibility, cost of ownership and automation that everyone has been seeking.”
* UPDATE
AWS has asked TMN to clarify that the use of the term ‘bring your own’ is not intended to signify that customers or partners bring their own L1. AWS offers a fully pre-integrated server, working with ISVs to integrate their preferred accelerator for L1.
