Frank Frankovsky Facebook
Facebook
Facebook open source hardware guru Frank Frankovsky
FBAug 10 08:00PM
21.81
Change
+0.80
% Change
+3.79%
Something extraordinary is happening at Facebook. The company is working on an idea that that could disrupt some of the largest enterprise tech companies in the world like IBMHPDell.
Facebook is leading a project that pushes hardware vendors into a new, and open-source way of building servers.
It's called the Open Compute Project. Its goal is to do for commercial hardware what Linux did for commercial software -- change the way it is designed, built, sold and supported.
Previously, Facebook would buy off-the-rack hardware from a company like Dell or HP. But now it custom designs the hardware it needs, then it has a manufacturer in Asia build it. After that, through OCP, Facebook releases the designs for anyone else to use or modify as they wish.
A lot of other companies who buy a lot of servers have taken notice and joined the effort -- as have the companies making the hardware. The OCP now has dozens of members like HP, AMD, Fidelity, Quanta, TencentSalesforce.com, VMware, Canonical, DDN, Vantage, ZT Systems, Avnet, Alibaba, Supermicro, and Cloudscaling.
The designs OCP comes up with are posted to GitHub, a site that hosts all sorts of open source projects.
This week, a group of OCP server vendors landed at Facebook's headquarters in Palo Alto to show of their wares and talk about the next steps.
We caught up with Frank Frankovsky, the man leading the OCP charge. Frankovsky is director of hardware design and supply chain at Facebook.
He told us:
  • After the team built its first data center in Prinevill, Ore, which opened in April, 2011, they saw that it was 38% less expensive to operate, gave them a "24% capex savings advantage" compared to buying gear from typical vendors and that their data centers consumed a lot less electricity, too.
  • Traditional server vendors have had mixed reactions to OCP, but are starting to come around. Both HP and Dell have created experimental servers.
Here is a lightly edited transcript.
BI:  How did the Open Compute Project begin?
FF: We were leasing data center space and buying-off-the-shelf server and storage products that were built for a different use case.  They are are great for enterprise IT users who might have a few racks of equipment here and there. But when you are deploying at scale, you have a different set of requirements, a different cost structure and different environmental impact.
We decided we needed we needed to build our own infrastructure. Once we put it into production we thought, wow, this is more efficient than we thought. It's the most efficient data center that we're aware of in the world. How much greater would the impact be if everyone had a blueprint on how to build these?
BI: What's unique about your hardware versus the ones built by commercial vendors?
FF: We have this concept of "vanity free designs." There's a lot of non-value-added materials in server or storage designs. For example, when you look at a HP or a Dell server, they have this beautiful plastic bevel on the front that allows them to put logo on the front. That plastic not only impedes the air flow, which causes the fans to work harder to cool the servers, but when you are deploying servers by the 10's of thousands, that's a lot of wasted material that is someday going to be decommissioned and put into the waste stream. 
BI: Are there any other measurable benefits of an open compute server vs a commercial server?
FF: Once in production with these servers, the time that it takes to repair the servers [is quicker]. We do time and motion studies on how long does it take a technician to repair a failed component on an open compute server relative to a Tier 1 HP or Dell server. In some areas there's an 8x decrease in the amount of time. That allows us to have one technician supporting up to 15,000 servers. In some other environments, you might have one technician for 150 servers.
BI: How did the server vendors react when you made your designs an open source project?
FF:  It's been an interesting year. There's been some really difficult discussions and also some really fruitful discussions with our core suppliers. The difficult discussions typically centered around, 'Wait a minute, if you are going to be building your own stuff, how am I ever going to compete for your business? I'm building a product portfolio designed for the masses, not for a Google or an Amazon orMicrosoft or a Facebook. You guys are small number of very large consumers.'
The trend towards cloud computing leaves the HP's and Dells of the world in a challenging position. Everyone has a finite number of engineering resources.  But along comes open compute and they see a light. One of our functions is to give a cleaner signal to the HP's, Dell's, IBMs, Seagates, FusionIOs of the world of what our market really wants.
BI: Do you see your designs and your open source design method being useful for enterprise data centers?
FF: I believe where we're focused is the leading indicator and not lunatic fringe kind of design. The question is who wouldn't want a more efficient server?
BI: But right now, the enterprise can't get them, can they? Unless they grab your design and go off and build them themselves?
FF: Well they actually can. And that's one of the really cool offshoots of the OCP. We're starting to see open compute solution providers launching new businesses to distribute these designs to end users. Not everybody has to go and make the investment in developing their own supply chain.
This week, we've got all the Open Compute designers at Facebook for a summit of all those companies.  Companies like Synnex has launched a new division called Hyve around developing and supporting Open Compute hardware. Avnet is investing in Open Compute labs around the country where end users can come in and work on the Open Compute product, customize it to their liking, then procure it and have it be supported through Avnet. Quanta has launched a new business called QCT where they are bringing new Open Compute technology directly to end users.
It's really exciting. My prediction is we're going to see similar things with Linux and Red Hat. [Red Hat] allowed companies to adopt Linux and know that there's a technology company behind it to support them. That's the same trend we're seeing with the Open Compute hardware business.