Remember when we said that Facebook was going ARM a few years ago and all the pundits laughed? Guess what Facebook and Calxeda are announcing today at the Open Compute summit?
The big news today is on multiple fronts, a new spec, companies filling it, and a new take on an older spec, all involving Calxeda. Calxeda, formerly known as Smooth-Stone, makes ARM chips for servers. Open Compute is a Facebook project that Calxeda is releasing designs for today. The idea is to nail down standards for their own internal consumption, to create an ecosystem around that, and generally ensure devices are there for them. See where this is going?
The new spec is the big bang, and that one is called Common Slot. In a similar fashion to how the Open Compute server platform defines an xU rackmount server ecosystem, Common Slot defines the microserver variant of it. Think about Calxeda and the boxes they make possible, then Seamicro, and a host of others, see a pattern? That pattern is proprietary cards for the whole microserver paradigm, exactly what Open Compute is trying to avoid. Common Slot is just a way to specify the mechanical and electrical interfaces of microservers to make them plug and play compatible among vendors for data center buyers.
Common Slot will do that nicely, and Calxeda is one of a very small handful of companies making Common Slot servers. In fact, Common Slot is not fully baked, today is only the official unveiling of the spec, currently on v.3. With four or so SoC vendors in agreement, and Facebook putting their purchasing might behind it, the remaining .7 of the spec is likely to get worked out in a hurry. Then microservers can be interoperable, ARM or x86, mix and match, pick your chip, vendor, and all the rest from a long list. There are only a few vendors allowed in this early, did we mention Calxeda is one? Any guesses what that means?
Moving on, there is one other part to mention, something that was called Project Knockout by Calxeda. One of the first Open Compute projects was a spec called Open Vault, basically a standardized NAS box spec that ODMs could build to. It was dumb storage that required a Xeon box as the head node for every handful of Open Vault boxes.
Calxeda steps in to this ring with Knockout, and the idea is to put one of those little EnergyCore nodes in each storage array. This distributes the compute, adds redundancy, and potentially lowers cost and energy use too. Knockout is not an Open Vault standard yet, but it is a good idea. How well it is accepted by the rest of the Open Compute players is not known yet, but the concept is technically solid. You have to wonder if Open Vault spec current+1 will have a Common Slot slot, it does make a lot of sense from every angle.
That is the news, lets get back to the background. Nearly two and a half years ago, August 23, 2010 to be exact, SemiAccurate reported that Facebook was going to jump in to ARM servers. People laughed. People mocked us. People claimed we were making things up for hits, and those were the nice ones. Everyone who was anyone in the industry claimed they knew better, and we were wrong. Guess what? Not only did SemiAccurate nail the vendor, Calxeda was known as Smooth-Stone back then, we got the rest dead on too.
Needless to say, the reaction by everyone was extremely negative, but none more so than Intel. The article went up in the afternoon of that summer day, and later on the following morning I received a phone call from an irate Intel contact demanding the article be removed. The phrasing was unusually harsh and insistent, tones that this mild mannered person rarely uses. I listened to his concerns, very understandable given the nature of the story and it’s potential effect on Intel, and heard nothing I didn’t expect.
As is SemiAccurate’s policy, we offered him as much space as he wanted to publish an unedited rebuttal to anything we wrote. Normally this gets a dry paragraph or two in the finest legal approved PR speak, and that is that. When a company knows they are in the wrong, they usually decline this offer, but rarely politely. This time, my contact got more upset and still insisted that the article be pulled down, retracted, and danced upon by a team of overweight men in golf shoes. This was not the normal peeved PR operative on the other end of the phone.
So we offered him SemiAccurate’s second option after the right of reply, and nicely suggested that he do something that rhymes with buck toff. He still insisted that the article was wrong, and it needed to come down now, earlier if possible. Something was up, so I decided to refer to the source and told him to ask Facebook directly though his contacts.
“X and I were at the CTO’s office at 9am this morning” was the reply, paraphrased, that I was not expecting to hear. An article on SemiAccurate does not usually prompt in-person meetings with the CTO of Facebook and two Intel employees, one of which has a title that begins with “Head of” first thing the following morning. Something was up. The Intel employee told me that no less than the CTO of Facebook himself said there was no truth to the article. He probably wasn’t expecting my response either, half of it was. “No”. The other half can’t be reprinted without compromising sources.
The article stood, the Intel contact was less than pleased, and the news spread around the net. Journalists were lining up to prove SemiAccurate wrong, from analysis to questionable quotes from Facebook, ARM, and Intel, there was a lot of negativity. Our forum thread on the subject was a bit less than supportive, but that does happen. Then no less than the VP of Technical Operations at Facebook, Jonathan Heiliger, posted the following comment on our site. (Note: Due to a change in SemiAccurate’s back end, older comments were lost over a year ago. The comment is no longer on the page, but it is authentic.)
“Facebook continuously evaluates and helps develop new technologies we believe will improve the performance, efficiency or reliability of our infrastructure. However, we have no plans to deploy ARM servers in our Prineville, Oregon data center.” Pretty telling right? If you parse it out logically, you will realize three things, Jonathan is a very cunning wordsmith, he doesn’t actually deny a single bit of the article, and it is Prineville, not Pineville like we thought for many years.
From there, that leads to a very obvious conclusion, and it is one that we have been pointing out for a while now, the modern press is full of dumb, tame chimps that have as much trouble with simple logic as they do with journalistic ethics. Brian Caulfield of Forbes did half of the right thing by asking Facebook, but didn’t bother to ask SemiAccurate anything before attacking us in print. Same with Rich Miller of Datacenter Knowledge, but he doesn’t do much more than rewrite Forbes without adding so much as an original thought, fairly typical it seems. There were many many others, but no one that we know of had the sources to actually verify the story independently, just reprint the official word while trying to look superior.
Step forward two and a half years, and guess what? In case you didn’t put the dots together, Facebook is going to move to ARM, and is going to move to ARM with Smooth-Stone/Calxeda hardware. No, this wasn’t a lucky guess back then, SemiAccurate knew what was coming, where and why, and we stuck by our article in spite of all the criticism and clueless lackeys trying to make a name for themselves. We were right, they were dead wrong, and Facebook’s CTO has a pretty thorny credibility problem about now. One has to wonder when the retractions will come out, or if the originals will just be silently pulled from their respective sites. Any guesses?
That said, Facebook is going ARM, no question. As of this moment, it is official, and the hardware is specified as well. Calxeda is one of the vendors making ARM servers to Facebook’s design, and while deployment is not announced yet, we still trust our original sources. There were no changes to the story, no backpedalling, and no waffling. Years later, others tentatively wrote up small bits of the story, but most still have no clue. To date, no one got it fully. I’ll bet these same people still doubt this, and this, and doubt the most recent industry changer, although not this, or this anymore.
SemiAccurate gets news first, gets it right more often than almost any other source, and gives you analysis that no one else can match. And the best parts are times like this when we get to say, “Told you“.S|A
Latest posts by Charlie Demerjian (see all)
- Intel kills off the 10nm process - Oct 22, 2018
- HyperX ships it’s 60 millionth enthusiast memory module - Oct 15, 2018
- Bittware/Nallatech water cools 300W of Xilinx FPGA - Oct 12, 2018
- More on Intel’s 10nm process problems - Sep 17, 2018
- Intel puts out another 14nm 2020 server platform - Sep 11, 2018