His talk was entertaining anyway — Slessmen is a trash-talker and knows how to work an audience — but the demo was a last-minute idea, so the real concern was that some goofy text message would pop up on the keynote session’s big screen. Slessmen made a crack about Snapchat and said he’d told friends not to contact him in any way for a couple of hours.
The demo involved calling up a server instance in Phoenix from the phone in San Jose, Calif. — an example of “the data center as API,” as Slessmen described it. He pulled it off in something like three minutes, including time for fat-finger corrections.
The success gave Slessmen a chance at his own Richard Sherman moment. “That’s what the world of compute deployment is going to be, period,” he said. “If people don’t see what’s coming, they’re going be run over by it. Data centers will be provisioned with clicks.”
In a way, that attitude summed up the Open Compute Project (OCP). Most speakers didn’t match Slessmen’s bravado, but for even the nicer guys like OCP’s own Frank Frankovsky, the subtext was the same: Open-source data-center hardware is here. Accept that, or kiss asphalt.
A $1.2 Billion Difference
OCP means enough to Facebook, its founding patron, that Mark Zuckerberg (the CEO; you might have heard of him) came out as a surprise keynote guest, chatting briefly with publisher Tim O’Reilly (pictured above). Zuckerberg spent a lot of his time explaining the motivations for Open Compute and defending the idea — nothing earthshaking, but just being there for the live audience was a strong gesture on his part.
Now boasting 3,300 attendees the OCP Summit was first held five years ago in Facebook’s cafeteria, where the 500 attendees had to get everything done in the hours between lunch and dinner, serenaded by the clattering of dishes throughout, Zuckberberg said. OCP’s impact since then has been deep — at least for Facebook, where officials claim to have saved more than $1.2 billion last year by using open-source hardware.
That’s attracting some buttoned-down names to the OCP ranks. IBM and Microsoft joined this week, with Microsoft even contributing the open-source Microsoft Cloud Server Specification, “the most advanced hardware servers we’re deploying in our data centers today, used to power Office 365, Windows Azure, and Bing,” said Bill Laing, a Microsoft corporate vice president, in his keynote.
This ARM’s Got Legs
And then there was ARM. There’s nothing in the Open Compute Project bylaws that says the whole thing has to be one big ARM party, but ARM’s chip design sure hogged the stage at the OCP Summit this week.
Some of that comes from ARM’s anxiousness to get its 64-bit architecture into data centers after getting off to a late start. But in a way, it also reflected the whole philosophy behind OCP: Try to break some rules.
Intel, by contrast, didn’t happen to have a product to announce and got about five minutes on stage. (Several sponsors took little five-minute speaking slots as opposed to longer keynotes.) Of course, Intel will have a lot of say as OCP continues to gather a following. ARM just shouted louder this week.
ARM itself announced the Server Base System Architecture (SBSA), an open-source hardware specification for servers. Ideally, this sets up an ecosystem where application and OS developers can write code one time and apply it to all ARM v8-based servers, Chief Marketing Officer Ian Drew said.
And AppliedMicro AppliedMicro CEO Paramesh Gopi said his company would introduce two more generations of the company’s 64-bit, ARM-based X-Gene family by the end of the year. The first X-Gene, shipping now, was introduced in 2011, and X-Gene 2 is due to sample in the spring, he said.
ARM also got a shout-out from Marc Andreessen, the Mosaic pioneer and now a partner with Andreessen Horowitz. During a fireside chat session on stage, he envisioned a future where the data center starts taking advantage of the smartphone supply chain, down to using ARM processors and flash memory. “I’m kind of a radical about this,” he said — and indeed, Arista chairman Andy Bechtolsheim, who was sharing the stage with him, expressed doubts, especially given how much Intel can accomplish with advanced manufacturing technologies.
Andreessen had a good point, though: “We deal with very few applications at Internet scale that are CPU-bound.” It’s more a question of I/O limitations and plain old costs, he said — and if superior clock speed isn’t a factor, then one of Intel’s strongest suits is nullified.