Vodafone to keep Intel in check with Arm server for virtual RAN

Intel has long been a flaw in the argument that open and virtualized networks spur competition. The US chipmaker's dominance of the computer world is hard to understate. Last year, for instance, it shipped more than 70% of the central processing units (CPUs) installed in data center servers, according to Counterpoint Research. If it carries this into the virtual radio access networks (RAN) market, telcos desperate to avoid dependency and vendor "lock-in" may have a problem.

The situation would already be troubling were open and virtualized RAN infrastructure a much bigger deal. In the run-up to this year's Mobile World Congress, Intel was boasting a 99% share of the market. There is no hysteria only because these new concepts still account for a thin slice of the total RAN market. But their telco champions dream of the day when open RAN is the de facto design.

Hence the push for alternatives powered by Arm, the UK-based chip designer whose blueprints are more commonly recognized for their ubiquity in mobile phones. Having already made plenty of noise at this week's FYUZ event in Madrid, arranged by the Telecom Infra Project, Vodafone is now promising an Arm-based, open RAN server next year.

It is due for lab tests at the start of 2024 and expected to land in the field, hopefully intact, by the middle of the year. "All this is aimed at having a commercially available Arm server for open RAN this time next year, which is probably sooner than most people are thinking," said Santiago Tenorio, the network architecture director for Vodafone Group, during a conversation with Light Reading.

A hand for Arm

This potentially has major implications for Intel, which is already being challenged by the use of hardware accelerators as a CPU offload in virtual RAN. To address the unsuitability of general-purpose processors for RAN compute, Intel backs a technique called "lookaside" acceleration that shifts one or two especially demanding physical-layer (or Layer 1) functions onto separate silicon. But others prefer what's known as "inline" acceleration, which moves the entire Layer 1 onto a more customized chip. Inline acceleration is a possible threat to Intel because it might allow operators to make do with a less powerful (and cheaper) CPU.

Whatever happens in this fight between lookaside and inline, CPUs are still needed to process higher-layer functions in a virtual RAN. But the server market is one Arm has long struggled to crack. That has left AMD as the only sizable alternative to Intel. The rival chipmaker also uses x86 architecture and is similarly a fan of the lookaside technique. Ericsson recently claimed to have deployed the same Layer 1 code on both Intel and AMD platforms.

An all-x86 market based on Intel and AMD is not ideal, as far as Tenorio and various other stakeholders are concerned. But several are now giving Arm a hand. Ampere Computing, a silicon designer funded heavily by Oracle, is currently working with Hewlett Packard Enterprise (HPE) to develop an Arm-based server for virtual RAN. Tenorio says testing has also involved Japan's Fujitsu as the provider of the RAN software that would run on the Arm platform.

"We are testing with Fujitsu software because they were more open and more available than the others, but more vendors will come," he said. "I know Ericsson and Nokia and Samsung are all paying attention to this because they really want competition to Intel."

One possible advantage of inline over lookaside is that its silicon usually comes on a separate card based on the PCIe standard. Provided all server makers adhere to that, an inline card could just as easily connect to an Arm-based CPU as an x86 one. This is partly why Nokia says it is backing inline. Ericsson, which prefers lookaside, would instead need to ensure its Layer 1 software can work with Arm. It is currently collaborating with several Arm licensees on CPU development.

Portability problems

A drawback noted by Matteo Fiorani, the head of Ericsson's cloud RAN product line, is that Arm remains less powerful than x86. But he is encouraged by recent architectural updates including support for a library called SVE2 that partly mirrors Intel's AVX-512 instruction set. "That is basically vector processing and, when they get that in, we think we can squeeze some good capacity out of an Arm system," he previously told Light Reading.

Another concern is the apparent incompatibility of the Arm and x86 systems. Fiorani, however, is optimistic that software can be made more portable, and Tenorio agrees. "It is much easier because this kind of difference – x86 versus Arm – this is something that the operating system can almost abstract completely," he said, highlighting the role that cloud platforms would play. "If you are just deploying on Red Hat, they will take care of abstracting that to the extent that porting one to the other is not a huge effort."

Still, the Arm ecosystem remains immature compared to x86, as Intel happily points out. Nor do Arm's various licensees necessarily have much in common. The special sauce that each one adds to Arm's basic recipe means these licensees are not easily comparable. In that respect, they certainly appear less "general purpose" than Intel and AMD. For ecosystem partners, that could be a challenge.

Enthusiasm for Arm has not stopped Vodafone from deepening its relationship with Intel. The operator has been exploring silicon designs at a facility in Malaga, Spain, where Intel has been a prominent partner. And at FYUZ on Monday morning, Tenorio announced that Vodafone is now starting to codevelop products with Intel, combining its intellectual property with forthcoming chips. "You can expect a lot of innovation in terms of algorithms that will eventually go into silicon," he said.