Vodafone slams Intel and its chip rivals on standardization

FlexRAN by name, inflexible by nature. It's a harsh assessment of Intel and its reference architecture for a virtualized radio access network (RAN). Nor is the critic easy for the US chipmaker to dismiss.

Yago Tenorio is not only the network architecture director for Vodafone, an operator that aims to use new open RAN technologies across 30% of its European footprint by 2030. He is also the current chair of the Telecom Infra Project, the Facebook-led group zealous about new-look network technologies.

For all its claims to be open, Intel is about as closed as a monk's lips after a vow of silence, judging by Tenorio's comments. "Even if you use FlexRAN, you have to design your software in a particular way using the instruction set FlexRAN gives you," he said in conversation with Light Reading this week. "Once coded for FlexRAN, it is not portable for another accelerator – you can only use Intel from that point on."

Accelerator, in this context, refers to the silicon platforms designed to boost RAN performance. Even Intel has acknowledged that its x86, general-purpose processors are less power-efficient than more customized silicon, assembled by the likes of Ericsson, Huawei and Nokia in a traditional RAN.

Intel's original answer was a technique called "lookaside," which hands off some of the baseband (so-called Layer 1) functions to a separate accelerator but continues to rely mainly on the x86 processor. More recently, it has been touting an "integrated" accelerator, heavily x86-dependent but promising better performance than lookaside. Rivals, meanwhile, have settled on an alternative known as "inline" acceleration, which introduces more energy-efficient silicon and leaves the x86 chip with far less to do. Regardless of these choices, software developed for one vendor's platform cannot be used on another, said Tenorio.

Figure 1: Vodafone's Yago Tenorio (right) chats to AvidThink's Roy Chua in front of a pineapple at the food-themed FYUZ event in Madrid this week.
(Source: Iain Morris/Light Reading)

"In simple terms, what the accelerator gives you is a new instruction set," he told Light Reading. "That instruction set, until we can standardize it, is proprietary for that piece of hardware. If you are Samsung, you can probably have one version for Marvell and one for Qualcomm and one for Intel, but it is not ideal and therefore the next wave of standardization needs to be on the Layer 1 interfaces, which are very difficult to abstract with middleware."

The problem is recognized by Joel Brand, the senior director of product marketing for Marvell, who admits there is no silicon chip that will run everything. FlexRAN, he notes, needs an instruction set known as AVX-512, and only Intel supports it. "You can't run FlexRAN on an x86 from AMD," he said.

The lack of software portability could explain why Ericsson's virtual RAN offerings are currently limited to Intel technology. Nothing seems to have come of a long-standing agreement between the Swedish kit vendor and Nvidia, although Brand derides Nvidia's chips as too power-hungry to be competitive. Approached by Light Reading, Ericsson acknowledged that it would need to make software changes to accommodate other hardware platforms, while playing down the effort involved.

"Ericsson cloud RAN software is highly modularized to cater to different accelerator technologies, should the need arise," said the company via email. "The actual adaptation needed will depend on the architecture and interfaces that are selected."

Silicon implants

The other side of this argument is the need for differentiated software products. If US chipmakers take market share from Ericsson and Nokia, this could become even more important. "When merchant silicon is better than custom silicon, does it make sense to build custom silicon or focus resources on the software that is on top of it, where they deliver the most differentiation?" said Sachin Katti, the chief technology officer of Intel's network and edge group. "If anything, I think it will allow them to focus resources on the thing that differentiates them the most."

FlexRAN is not forced on developers using Intel's x86 technology, Katti also points out. It comes free of charge and can be modified, but there is no requirement to use it at all. Any developer, if it preferred, could instead build its software entirely "from the ground up," said the Intel executive.

Marvell seems broadly in agreement about software design. "We are trying to work with our customers on developing silicon that allows them to differentiate," said Brand. "At the same time, we are trying to make that same silicon available to others so that if an operator wants to buy something from Mavenir, Mavenir can run on silicon that was originally optimized for Nokia and Samsung."

As a relatively small developer of baseband software, Mavenir itself plays down the issue of standardization, too. "From base FlexRAN code, suppliers will have to harden and add their value on the basic reference design, allowing vendors to differentiate," said John Baker, Mavenir's senior vice president of business development, via email. Hardware deployed in a production network is unlikely to be changed for years, he added.

Want to know more about 5G? Check out our dedicated 5G content channel here on Light Reading.

Still, it is hard to believe any software developer would welcome having to write different code for each silicon platform. The ideal would surely be software that works on any hardware with minimal tweaking but still offers scope for differentiation from software rivals.

The O-RAN Alliance, the group behind open RAN specifications, has at least worked on developing an interface between Layer 1 and Layer 2, the data link part of the network. This should offer more freedom to combine vendors. But without some harmonization of instruction sets at Layer 1, that software job looks more onerous than it should really be. The feelings on that topic of one of Vodafone's most senior technology executives will be hard for the industry to ignore.

Related posts:

— Iain Morris, International Editor, Light Reading