Breakfast Bytes Blogs

So your first thought on reading the title to this post might be that having to stay at home has finally got to me and made my brain turn to mush. Or perhaps that I'm really, really short of material to write about. After all, isn't 5G the next big thing. Why would turning on a 4G network be newsworthy of a blog post? And when was the last time anything significant happened in mobile in Japan? i-mode was launched in 1999, over 20 years ago.

Here's why this is significant. I'll give you the opening paragraph from GSMA. They are the operator organization, and also the people who organize the MWC events, the big one being in Barcelona (not this year, of course).

Rakuten Mobile turned on commercial 4G service after a six-month delay, to become the fourth mobile operator in Japan, a move it claimed made it the first operator with a live service based on fully virtualised infrastructure.

The key thing in there is the term "fully virtualized". That is why this is significant for 5G, too. What the phrase means is that the network is largely built on commodity hardware (cheap) and special software (expensive to develop but cheap to scale). Obviously, you cannot completely implement a mobile network purely in software—at some point, you need radios and antennas to talk to the handsets. Historically, networks have been implemented using proprietary hardware and software from one of the network equipment suppliers, of which there used to be many and of which there are really only three that operate at scale today.

A similar thing happened with data center networking, where we have gone from proprietary routers running proprietary software from one of the big router companies, to software-defined networking (SDN) where the forwarding (requiring special chips) is separated from all the control (implemented purely in software).

A fully virtualized network has the possibility to be a lot cheaper to build and operate. Rakuten themselves said:

Its launch was the world’s first large-scale deployment using a fully virtualised mobile network, resulting in substantial reductions of capex and opex.

In the context of a mobile network, capex (capital expenditure) is the cost of constructing the network. Opex (operating expenses) is the cost of running the network.

O-RAN

One of the most specialized areas of mobile is the radio access network, usually abbreviated to RAN. This is a fancy name for all the basestation hardware and software that is specific to the standard being implemented. Further up the stack are functions like interfacing to the internet, or billing, which are naturally implemented as software running on standard hardware, or even running in the cloud. Just as with data center networking, there is still a need for specialized hardware in the mix. In the data center, since 2012, one driver has been the Open Network Foundation (ONF), who have been driving open standards for SDN. There is a similar organization in mobile, called O-RAN. One important aspect of O-RAN is that it is driven by the network operators more than the network equipment suppliers. Its mission is:

Leading the industry towards open, interoperable interfaces and RAN virtualization

As with SDN, the expectation is that by driving open specifications, the O-RAN alliance will allow operators to purchase kits from a wider range of suppliers, including smaller ones. As more and more of the system is implemented in software, the remaining hardware component becomes more of a commodity. I think that there is an analogy with data centers, server manufacturers, and operators of the biggest data center networks. Those companies were all happy for Linux to become the operating system (OS) of choice. If, instead, it was an IBM operating system, or Windows (or, earlier, Solaris or VMS) then it would be both more expensive (have to purchase OS licenses) and would restrict the hardware choices (Solaris only ran on Sun SPARC chips in practice). I'm not making a prediction as to what will happen, I'm too far removed. Plus, I should point out that nobody predicted Linux would take over the datacenter...and that includes Linus Torvalds.

This may not be the endpoint. In HOT CHIPS: The AWS Nitro Project, I wrote about how Amazon/AWS is increasingly running on chips that they designed themselves, running software that they wrote themselves. The first data center I came across was the one we built at Ambit (although we called it a server farm, it was not yet on the scale of true data centers in their own massive buildings). I wrote about that recently in Getting to Hyperscale Data Centers: PCs to Clouds. it was about 40 Sun workstations and 20 HP workstations. For this post, the important factor is that all the hardware was proprietary, designed and controlled by Sun and HP, running proprietary OSs. The next phase was commodity servers running Linux, with chips supplied by the semiconductor industry (for the actual processors, Intel and AMD), for other chips mostly a range of suppliers. Amazon has gone to the next stage, building all its own chips except for the processor itself. And they've even dipped their toe in that water, as you can read in my post Xcelium Is 50% Faster on AWS's New Arm Server Chip. For deep learning in data centers, there are chips in use at both Microsoft and Google that they designed themselves, and there is a phalanx of semiconductor startups eager to supply chips waiting hopefully in the wings. I remain skeptical that many of them will succeed. For example, Wave Computing recently shut down, and I considered it one likely to succeed when I first heard about it and wrote about it in Wave Computing: a Dataflow Processor for Deep Learning. I'm not predicting most network operators are going to suddenly start building their own chips, most networks in the world are not at the sort of scale where that would make sense, but it wouldn't surprise me if a few do.

This post is not intended to be some sort of commercial for Rakuten. But virtualized open mobile networks will be important for 5G, too. Indeed, Rakuten says they will roll out 5G in June in Japan, although with the country currently locked-down, that timing might well change. In fact, I wouldn't bet on any 5G rollout timetables at present.

More Information

The GSMA piece I referred to at the start is Virtualised Rakuten Mobile 4G network goes live.

After I had largely written this piece, The Economist covered some of the same ground, although their focus was less technical. They have even come up with a catchy name for the business segment 5Geopolitics. I don't think I'd go anywhere near as far as they do when they say:

If netheads have their way, it may prove as momentous as Amazon’s launch in 2006 of its trailblazing cloud-computing arm.

Amazon/AWS today has something like 50% market share in cloud computing. Rakuten has zeroish in mobile. Of course, when AWS was launched, it had zero market share, too, but it was launching into open water. The mobile industry has players locked in place and subscriber growth has plateaued. But there is a parallel too: Rakuten is the largest e-commerce company in Japan. And there is always room for a shakeup—just look at Reliance Jio in India.

For general coverage on 5G, you can turn almost anywhere. But, similar to what I wrote in my post 5G in 2020, The Economist article warns:

Many telecoms experts dismiss 5g as hype by makers of devices and networking kit drumming up business. They are partly right. 5g will not profoundly alter consumers’ lives.

For my own coverage of 5G, start at the piece referenced above. Or look at:

• What Is 5G?
• Why Is 5G Such a Big Deal?
• 5G in US vs Rest-of-World

For details of Cadence's 5G portfolio, see 5G Systems and Subsystems on our website. The rollout of 5G will take place over a period of years, but now is the time that the chips and systems have to be designed.

Sign up for Sunday Brunch, the weekly Breakfast Bytes email.