At the Open Compute Summit this week in New York, Frank Frankovsky, Facebook’s director of hardware design and supply chain, opened the proceedings by saying, "Open source is not just something that you can use to describe software, but also to describe the hardware space."
That is the goal for the Open Compute Project, which aims to spur the development of cheaper servers and more efficient data centers. The project was kicked off by Facebook in April 2011 and has shared details of the social networking giant's customized server specifications and data center design principles. At the summit, a board of directors was announced that includes members of Facebook, Intel, Arista, Amazon, Goldman-Sachs and Dell. But does this project really herald an open source era for hardware? Yes and no.
The Open Compute Project certainly is based on open source principles and guidelines. And it has definitely embraced many of the aspects of successful open source software groups, such as the Apache Project, including an open model for contributions and project organization.
But people shouldn’t expect to see full-fledged products emerge from this group in the same way that you see code coming from Apache or Mozilla. What the Open Compute Project is releasing are specifications for data center hardware, including everything from servers to racks to batteries (but not networking equipment). With these open specifications, companies and vendors will be able to build products designed to these open hardware specifications.
Currently, the efforts of the Open Compute Project will be of most interest to companies at the highest end of the data center picture--companies such as Facebook and Amazon that design massive data centers that often take up entire buildings. For example, one of the specifications launched at the summit was for a giant triplet rack designed to handle Open Compute servers. This is a gigantic rack that wouldn’t fit in the data centers of many companies.
When I asked the Open Compute Project board when and if some of these designs would be useful for the "smaller" data centers found in some businesses (like, you know, those with just 1,000 servers as opposed to 40,000 or 50,000), they said that down the road, the specifications would adapt to designs more common in smaller data centers, such as single racks.
Sitting through the sessions at the Open Compute Summit, I definitely saw quite a bit of exciting technology. There is a great deal of potential for savings in power, better and cheaper cooling and much greater interoperability for hardware in the data center. For example, the power supply models are designed to handle power from the utility with less conversion needed, cutting down on waste, and the rack and data center designs focus on avoiding the use of air conditioning and other expensive methods for cooling systems, relying more on regular room venting.
However, in many ways it felt the same as if I was looking at a Formula 1 car: The technology is really cool and some of it will eventually trickle down, but the advances in that race car (massive data center) won’t apply much right now to my regular sedan (typical data center).