With Open Compute, Facebook is saving billions and moving markets

Even Facebook never really imagined its decision to open source its server and data center designs three years ago would have quite this effect. Today, the Open Compute Project — the open source foundation launched in conjunction with those designs and spearheaded by Facebook — is on its way to saving the company billions of dollars. Much of that comes out of the power bill, but it’s touching other areas, as well.

During on-stage appearances at this year’s Open Compute Summit, which took place earlier this week in San Jose, both Facebook CEO Mark Zuckerberg and VP of Engineering Jay Parikh highlighted the cost savings: Development of energy-efficient technologies has saved Facebook $1.2 billion over the past three years.


Zuckerberg (hoodie) on stage with Tim O’Reilly.

That’s not just in energy-efficient servers, storage and data center designs, but all the way up the stack to efforts like its HipHop VM technology for converting PHP to C++ to run more efficiently. One of the company’s latest innovations is a system for slashing hardware costs and energy consumption by storing many petabytes of data on Blu-ray discs. It’s also experimenting with ways to standardize certain aspects of data center construction to the point where they’re the webscale equivalent of building a desk from Ikea.

But the “really pleasant surprise” about Open Compute, VP of Hardware Design and Supply Chain Optimization Frank Frankovsky told me during an interview at the show, is that it’s actually helping minimize the amount of work Facebook has to do deliver these savings. The foundation was formed to push server vendors to build products that large-scale buyers wanted — and that has undeniably worked — but beyond that, he acknowledged, Facebook didn’t see how it would directly benefit from open sourcing so much technology. Now, with community members contributing the majority of intellectual property and helping steer the ship in new directions, Facebook is finally reaping the rewards of other members’ investments.

Bending markets around your will

The big-picture effects of the Open Compute Project are really big. Market-moving big, in fact. Alone, no one company could push entire sectors to bend to its will, but Open Compute members like Facebook, Goldman Sachs, Baidu, Fidelity, Rackspace and Microsoft combine to buy hundreds of thousands of servers per year. With that kind of purchasing power comes the power to move even deeper into the data center.

What began with servers and storage is moving into networking gear and even microprocessors. The message from Open Compute is that if its members align behind a certain design, vendors can expect they’ll buy it. And if traditional vendors don’t want to build those products and make those sales, there’s always someone that will.

An array of alternate server designs from Intel.

An array of alternate server designs from Intel.

At this point, though, Frankovsky expects vendors will deliver the goods without forcing companies like Facebook to build even more of their own stuff. Take, for example, low-power ARM-based processors, which Facebook and other Open Compute members are publicly excited about for certain workloads. Asked directly if Facebook would ever acquire an ARM license and design its own chips like Google is rumored to be doing, Frankovsky replied, “Only if people aren’t going to build it for us.”

But, he added, “To me, that’s almost an admission of failure because you’ve been unable to influence the supplier community to build exactly what you need.” Chipmakers Intel, AMD and AppliedMicro are already members of, and contribute to, Open Compute, so they’re aware of what the community wants.

Open Compute could even help Facebook realize the “kind of a crazy idea” of renting microprocessors to ensure it always has the latest and greatest ones in its data centers. Theoretically, Frankovsky explained, by the time mainstream IT buyers have tested new models of processors, Facebook might have been running them for 10 months and be ready to resell them at a discount and move onto the latest model. Swapping out CPUs can be messy and tedious with current motherboard designs, but Open Compute’s “group hug” design was built with modularity in mind and could ease the process, Frankovsky said.

Frank Frankovsky of Facebook holding an AppliedMicro board.

Frank Frankovsky of Facebook holding an AppliedMicro board at last year’s summit.

Lessening Facebook’s load

But Facebook isn’t just benefiting from Open Compute members’ IT budgets. It’s also benefiting from their IP. At this point, “[It’s] far less than half now that Facebook is contributing to the project,” Frankovsky said. “… Probably less than 20 percent.”

Facebook is still picking up its pace of innovation, he added, but so is everyone else. When Facebook needed some new designs for power supplies and storage architectures, it used designs from fellow member Rackspace. Open Compute is close to releasing its first reference design for open source top-of-rack network switches, an idea that came from members Cumulus Networks and various open networking consortia. Frankovsky predicts the foundation’s next big push will emerge from the community rather than from Facebook, too.

So, while Zuckerberg used the occasion of his Open Compute Summit appearance to talk mostly about the company’s Internet.org effort, the fact that he was there at all did most of the talking about the value Open Compute is delivering to Facebook. It’s the same signal Amazon CEO Jeff Bezos sent when he keynoted the inaugural re:Invent conference for Amazon Web Services, the company’s cloud computing division, in 2012: Now that this thing is denting the bottom line, it has the blessing from on high.

And Zuckerberg did remind the audience that, in business, most thing do come down to money in the end. Doing the right thing tends to catch up with you over time, he said, and done right, “there’s probably a way to make a profit from it later on.”