People driven by causes are necessarily bad at moderation. No member of Extinction Rebellion or Just Stop Oil, UK groups upset about climate change, ever made headlines by calling for a sensible approach to the use of fossil fuels in politely worded letters. Hence the net-zero-by-2025 demands that could only be satisfied if most people agreed to starve or freeze to death over the next couple of years. Among the recent antics, one dedicated protestor jumped on top of a snooker table at a televized game and set off a powder bomb. He was photographed screaming in a sea of yellow mist like the victim of a gas attack.

Telecom has its share of causes, and they have generated some fanaticism, if not outrageous, Extinction Rebellion-like behavior, which doesn't really suit the executive lifestyle. The best example of this today is probably open RAN, a movement pushing for interoperability in mobile networks. Sounds boring? Well, it's surprisingly political, aimed by some of its proponents at curbing the power of big European equipment vendors (Ericsson and Nokia) and nurturing alternatives to dangerous Chinese ones (chiefly Huawei).

Figure 1: Protest movements in the telecom sector are less disruptive to everyday life.
(Source: David Holt via Creative Commons)

Without interoperability between suppliers, operators must buy all the parts for any mobile site from one of the aforementioned big vendors. Telcos don't like this, especially if their governments have banned Huawei and they are left dealing with a Nordic duopoly. If the interfaces between different parts of the radio access network (the RAN) were opened, telcos could buy from specialists, previously excluded from deals because they lacked the full range of products.

It sounds like an admirable ambition, but it's one as hard to realize as that net-zero-by-2025 dream. The campaign for openness originally targeted the "closed" interface between the radios and the computing products, the two big RAN elements. But the replacement cooked up by the O-RAN Alliance, the telco-led group in charge of specifications work, turns out to be rubbish, according to various stakeholders.

Splitting up is never easy

The reason for that requires some elaboration about the technology. In an open RAN, a choice must be made about the apportionment of functions between a radio unit (RU) provided by Supplier A and a distributed unit (DU, responsible for baseband, the computing part) from Supplier B. Open fronthaul 7.2x, the interface developed by the O-RAN Alliance, shunted just about everything into the DU, leaving the RU as a relatively dumb object.

This was arguably in the interests of speedily cultivating a market for low-cost radios. Unfortunately, it necessitates a lot of important shuttling between the RU and the DU. That's not a problem in less advanced radios, but the extra delay could be a performance nightmare for antenna-rich massive MIMO. And the inexorable growth in data traffic means massive MIMO is going mainstream.

Figure 2: Ericsson's Mike Murphy takes aim at today's open RAN specification.
(Source: Iain Morris/Light Reading)

Hence a new proposal for a successor interface to 7.2x that would address this problem. Backed by AMD (a chipmaker), Ericsson, Nokia and ZTE, it would move some key functions back into the RU. Trouble is, it doesn't look quite as open. Some fear Ericsson, which seems to be the ringleader, is trying to slip proprietary tech into the system and thwart open RAN. Another concern is that DU companies might have to spend months redesigning their technology to ensure it is compatible with updated radios. In a LinkedIn comment about Light Reading's original coverage of the issue, Joe Madden, an analyst with Mobile Experts, said 7.2x is "trading off capacity to get openness."

In the meantime, open RAN's most zealous campaigners are pushing for change in other problematic areas. In the most open of open RANs, the software, handled partly by the DU, would run off general purpose processors (GPPs), given a helpful shove by accelerators (other chips) because those GPPs are not as good as customized silicon in the RAN. With a GPP and "virtualization," porting software from one hardware platform to another should be as easy as moving bags between hotel rooms. This is, after all, what separation of hardware from software is all about. But it simply doesn't exist.

One glaring deficiency until now has been the absence of GPP alternatives to Intel's x86 platform. This might slowly be changing, but it currently makes porting of some functions an academic exercise. And while the market for accelerators that handle other functions looks a tad more diverse, software cannot easily be moved from one accelerator to another.

Figure 3: Nvidia boss Jensen Huang is promoting GPUs for inclusion in telco networks.
(Source: Nvidia)

There are various degrees of what telcos might term "vendor lock-in." Trillion-dollar Nvidia is prepared to offer its graphics processing units (GPUs) as accelerators for use with other companies' software, but its own Aerial code is offered only with those GPUs. Ericsson can port its software from Intel to AMD, which uses the same x86 architecture, but seemingly not to anyone using Arm, x86's main (but much smaller) rival. Software built on FlexRAN, Intel's reference design, works only with Intel's chips. Nokia's software looks designed to work specifically with Marvell's. And so on.

Just about everybody except Intel and Nvidia provides accelerators exclusively through cards, too, rather than integrated with the GPP. These cards can be slotted into servers so long as there is support for PCIe, a common standard. But they have been condemned by several industry giants, including Ericsson, Intel and Verizon, as an abandonment of virtualization. Having software tailored to that specific hardware component makes "disaggregation more challenging," said Ericsson and Verizon in a white paper, and "eliminates the possibility to create a common cloud compute infrastructure across the network."

There's always a dependency

Yet these clouds are the absolute worst form of lock-in. Porting software between them is fiendish, operators have repeatedly complained. Accounts make it sound as though extricating systems from a public cloud provider such as AWS, Google Cloud or Microsoft Azure is like disentangling someone from a very persistent boa. An alternative is to use a virtualization specialist such as Red Hat, VMware or Wind River as a kind of abstraction layer that sits above and between those clouds.

"What VMware and Red Hat can do is make it so the platform is consistent," explained Roz Roseboro, a principal analyst with Omdia (a Light Reading sister company). "You don't want to have one version for AWS and one version for Azure and one version for Google. If you use VMware or Red Hat, you can have the same environment."

Figure 4: A different form of lock-in?
(Source: Red Hat)

Unfortunately, this means being heavily reliant on VMware or Red Hat instead. "At some point in that stack, there is going to be a dependency," said Rozeboro. "If you want to have something consistent, something must be a common factor. It's going to be VMware or Red Hat, or it's going to be AWS, or it's going to be Microsoft services. Something must be the same."

This probably won't stop operators from trying to avoid any lock-in. But even if they have success in some of these areas, other dependencies seem to have been largely ignored. Nokia's latest 5G equipment includes five-nanometer chips currently available only from Taiwan's TSMC and South Korea's Samsung. The extreme ultraviolet lithography equipment needed to make those chips comes entirely from ASML of the Netherlands. And as a US government report noted two years ago, the raw materials, the silicon and gallium used to produce chips, are "concentrated in China." No protest movement is going to change that.

Related posts:

— Iain Morris, International Editor, Light Reading