How does Facebook serve up its social network to more than half a billion people around the world–and do it with surprisingly few outages and glitches? It started out seven years ago by leasing servers in data centers as it needed them. Eighteen months ago, it built its own massive data center in Prineville, Oregon–and decided to design its own servers that were as simple and power-efficient as possible. Today, CEO Mark Zuckerberg hosted an event at which the company announced that it’s sharing its server design with the world through an organization it’s created called the Open Compute Project.
Facebook designed its own motherboards from scratch, in both Intel and AMD flavors. They’re not fancy: they have no paint or plastic, no slots, and as few screws as possible. And they’re a bit on the large side, in part to accomodate larger heat sinks and fans that help keep them cool. But in a video shown at the event, Facebook employees who worked on the project said that the servers demonstrate Facebook’s philosophy of being useful rather than glamorous and said that “they’re pretty to us.”
The servers use 22 percent less material than typical servers and have power supplies that are 94.5 percent efficient, according to Facebook. They’re connected directly to transformers without a backup battery, and are 99.9999% reliable. Compared to previous servers, they help the Prineville data center run at a 38 percent increase in energy efficiency, at 24 percent lower cost to Facebook. And unlike typical data centers that must be blasted with air conditioning to prevent servers from overheating, the Prineville facility isn’t air conditioned at all, and has no ducts. (Prineville was chosen for the data center in part for its climate.)
Facebook says that businesses of all sorts who run data centers–large corporations of all sorts, and Web-hosting companies–can benefit from what it’s learned, and it wants feedback so it can do even better. Through the Open Compute Project, the company is sharing its designs, CAD files, and related knowledge. The event included a panel with representatives of Intel, Dell, Zynga, hosting giant Rackspace, and the Department of Energy, all of who lavished praise on Facebook’s work and said it’ll help their organizations do more at lower cost while consuming less power. I also saw folks from such large companies as Delta and Dreamworks in the audience.
Usually, you can judge the success of a new Facebook initiative based on how much time Facebook users spend thinking it. In this case, as with server technologies in general, we’ll know that the Open Compute Project is accomplishing its goals if consumers don’t think about it at all–because all those servers are doing their job quietly, reliably, and efficiently, so Facebook and other sites that use the new designs just keep working. (Judging from how often Twitter makes news for flakiness and data center problems, maybe it should look into this stuff.)