Why 2017 is the year of the ARM server
"First they ignore you, then they laugh at you, then they fight you, then you win."
We are at the laughing/fighting stage now. The paid puppets of the tech press are writing articles citing the demise of ARM in server, and telling us that it should have happened by now if it was going to happen, and other total nonsense. Let me educate you.
Building a new architecture is hard. It's harder when you've a zoo of vendors used to the embedded world to tame in the process, getting everyone on board with how the server industry really works, actively fighting against many good technical people with terrible understandings of the market, of the importance of firmware and platform experiences done right, of one size fits all software. That's what we've been doing for the past 5 years. That's why you don't have an ARM server powering every aspect of your life.
But it's not because ARM isn't ready for prime time.
Now is precisely the moment that ARM is ready for prime time. We're done telling people how they should have done it 3 years ago. The next upstream Linux kernel will finally boot "out of the box" on every shipping ARM server platform adhering to the platform specifications (pretty much all of them at this point), and this is making its way into Linux distributions that people can "just install" like they do on x86. Sure, we should have been here years ago, but we are finally here now.
It's not just magic that things are going to suddenly ramp up over the course of 2017. It's deliberate and determined execution by those of us who have been fighting this battle for nearly 6 years of our lives. We always knew it would be a decade long activity. Not because it should take a decade (it "should" take at most a couple of years), but because the overall computing industry is stodgy, antiquated, slow moving, and full of inertia. You have to channel every ounce of your being, every last joule of energy to compel the world to move in a different direction. We are doing this. We have done this.
Those of you who think we aren't upon the cusp of greatness are living under rocks. I feel very sorry for you. You're going to think ARM servers "suddenly" became a thing next year, rather than realizing it needed all of the puzzle pieces to land together finally in a few top shelf designs, quality open source upstream software, and fantastic out of the box experiences. I'm very, very bullish on the next year ahead.
Jon.
P.S. Next time someone has a new server architecture to land, I can promise to do it faster. I will need total and utter control, a shark tank, and a fleet of black helicopters. That's how you do it in a year. Otherwise you get the above. Thanks.
I don't think that large organizations will easily "jump ship" to a new architecture without some essential shock to the system. One possible discontinuity might be Amazon turning to SOC's developed by Annapurna for large scale deployment in AWS. This should validate ARM use in the Cloud. It might be difficult to build on this outside AWS since Amazon is so secretive and protective of their technology. An other possibility is a new technology arising that will favor ARM technology and allow ARM servers to flourish. This would be the case if Ascender's Remote Android technology receives wide deployment in the Cloud. I frequently reflect on the interminable wait for the "Year of the Linux Desktop" that, like Godot, has not yet arrived. Oftentimes, what we hope for does happen, but in a way that we could not anticipate or immediately recognize as a vindication of our views. The current dominance of Linux in almost all computing endeavors (Cloud, Mobile, Embedded etc.), unanticipated a decade ago, makes its lackluster success on the desktop moot. Something similar might occur in the ARM server market.
Here's how it plays out: AMD sets the building on fire and causes a pricing disruption very soon, ARM offers compelling performance at low power on contemporary process nodes, and the hyperscalers don't have architecture stickiness to care about remaining on x86 for all of their workloads. Intel is about to learn a lesson in competition, but it's not even really about Intel. It's about new infra being built finally having a good choice of alternatives. In the new world, we don't have stodgy 1980s databases and other nonsense that journalists think people care about running. Also, decisions are made by a small number of very technically informed people, not the mass market. Mass market might be sheeple, but the big cloud vendors are sophisticated consumers who can handle an alternative.
I am not optimistic about 2017 being the year of the ARM server. I do think that the task to prepare ARM infrastructure for production work is a Sisyphean job that is necessary but unfortunately not sufficient to upset the x86 monopoly in Cloud and Data Center servers. The reason for my pessimism is the implied assumption that the current monopoly holder, Intel, will watch passively as its dominance is challenged. The ability of Intel to track ARM’s roadmap and to disrupt any compelling cost/performance advantage that ARM licensees currently might have is enormous. As Intel did not succeed in penetrating the mobile-embedded market largely because of market inertia, ARM will find it very difficult to promote their servers without a clear compelling reason for their use. We, at Ascender Technologies, provide just such a compelling reason http://bit.ly/2bZeTO9. Running Android Apps efficiently in a Cloud using ARM servers will leverage the advantage that ARM has in the Android ecosystem and create a new massive Cloud based on Android servers. A demonstration of this technology is shown here https://vimeo.com/180099519