Subscribe / Unsubscribe Enewsletters | Login | Register

Pencil Banner

Facebook pushes a new model for buying servers

James Niccolai | Jan. 17, 2013
Facebook has proposed a new model for designing servers that it says will give businesses more choice in selecting components and a smarter way to upgrade systems when needs change, though it remains to be seen how widely its method will be adopted.

Instead, power supplies can be implemented at the rack level, where they can be upgraded as needs dictate, he said.

Likewise, I/O modules, CPUs and other components should not be dependent on each other, Frankovsky said.

"That will allow us to do a smarter technology refresh when it's time to upgrade, where we take out some components but leave others we don't need to change," he said.

The common processor slot will allow companies to "evaluate different CPUs right up to the last hour" when they're selecting a new server design, he said.

It's easy to see how the model could benefit large-scale Internet services companies such as Facebook, which already designs its own servers. It has also attracted cloud service providers such as Rackspace, which also submitted new designs this week.

But some participants here said other large companies can benefit, too.

AMD said the motherboard it developed is designed to bring the Open Compute Project to more than just online giants such as Facebook. It developed the specification, which was codenamed Roadrunner, by working with financial services firms Fidelity Investments and Goldman Sachs.

It fits into a standard server rack, something that already sets it apart from Facebook's own server design, which is for a custom rack. The AMD board includes its Opteron 6200 or 6300 processors, but it allows the customer to choose the I/O module and whether they want a SAS controller, said Bob Ogrey, AMD's "cloud evangelist" and the engineer who designed the board. It also implements a low-cost server management platform that AMD developed with Broadcom, he said.

Companies are testing the board now, and full production will begin at the end of the quarter, he said.

Matt Eastwood, an analyst with IDC, said it makes sense for companies that rely on IT for a competitive advantage to design their own servers. Big banks are willing to invest heavily in powerful systems to run their Monte Carlo simulation programs, for example, which are used to assess investment risk.

Facebook's model may also suit other companies for whom high-performance computing is essential, Eastwood said, such as pharmaceutical firms or companies that do oil and gas exploration.

Still, that's far from most of the world's big businesses. "You're probably talking about a few hundreds of companies" that have an incentive to make use of the Open Compute Project's designs, he said.

He also noted that while Facebook is talking about disaggregation, there's a trend in the industry in the opposite direction, with companies such as Oracle promoting highly integrated systems as a way to achieve maximum compute performance.

Nathan Brookwood, an analyst with Insight64, was also skeptical that the Open Compute project is applicable outside large Internet and cloud service providers. "I'm not sure enterprises can benefit from it," he said.


Previous Page  1  2  3  Next Page 

Sign up for Computerworld eNewsletters.