Open-Source Cloud Hardware Grows Up Fast

6 Ways SDN Shakes Up The Enterprise

The Open Compute Project founded by Facebook has made rapid strides in what has been a narrow sphere of influence. The cloud is an x86 world, and, in its first three years, OCP has put out several motherboard and server designs for cloud projects.

The early adopters have been primarily big financial services companies, such as Bank of America, Fidelity Investments, and Goldman Sachs. At the Open Compute Project Summit this week in San Jose, Calif., OCP showed it may be ready to grow beyond one industrial segment into others, such as online gaming and pharmaceuticals, and expand the reach of open-source hardware in other ways.

OCP broadened its approach to licensing, as well. The hardware designs are available under an Apache Software Foundation-style of license, where the licensee may make modifications, sell them, and keep the additions proprietary. It's a "permissive" license, in the parlance of open-source. The adopter may do just about anything with the code without obligation. To encourage contributors, the Open Compute Project Foundation may move to a second model, a GPL-like license, where anyone may modify an Open Compute Project design, but if they sell a product based on it they must contribute their changes back to the community. The GPL governs Linux use. Its giveback clause makes it more restrictive, or "prescriptive," in how for-profit modifications must be handled.

"Soon we will roll out a second, more 'prescriptive' license," wrote Open Compute Project founder Frank Frankovsky in a Jan. 28 blog.

It will require anyone who modifies an OCP design and then sells a product based on it "to contribute the modified version back to the foundation. It's our hope that having multiple licensing options will lead to even more OCP technology contributions," he wrote.

[Cloudscaling's Randy Bias has a specific point of view about what makes a cloud service "open." See When The Open Cloud... Isn't.]

In another sign of maturity, OCP Summit speakers and panelists seemed to leave behind their references to converting thousands of servers and embraced both larger expectations and bigger numbers.

Facebook founder and CEO Mark Zuckerberg got the process underway Tuesday by estimating that Facebook's efficient cloud server and datacenter designs had saved the company "more than a billion" dollars over the last three years, compared to traditional datacenters. The savings came from less-expensive manufacturing of its first OCP motherboard server and from its datacenters' reduced electricity use. Facebook saves energy in its design of both equipment and datacenter facilities. The latter have massive air-flow handling and air-cooling techniques that allow ambient air to cool the equipment without air conditioning.

Figure 1: Facebook's Prineville, Ore., datacenter. (Source: Wikimedia Commons.)

"We've saved the equivalent of the electricity to supply 40,000 homes," said Zuckerberg in an interview with O'Reilly Media CEO Tim O'Reilly. The reduced energy use reflects a Facebook contribution to the environment equivalent to taking 50,000 cars off the road.

Bill Laing, corporate VP of Microsoft's Servers and Cloud unit, came up with some big numbers of his own. Microsoft has invested over $1 billion in cloud datacenters, which contain 1 million servers, he said. At the OCP Summit, Microsoft donated its design of a 12u cloud server that mixes layers of CPU and storage blades.

Cade Metz, a senior editor at Wired, carried the big numbers a step forward as he moderated a panel on the future of open-source hardware in networking. Martin Casado, chief network architect for VMware, said he's watched networking move from highly proprietary, inflexible sets of devices that tie a customer to a vendor, to a field where network switches may be programmed and reprogrammed dynamically throughout the day. "Five years ago, I was a lot more pessimistic about the future of networking," said Casado.

Metz deadpanned, "$1.26 billion later, he's more optimistic" to the laughter of the audience. Casado sold his networking startup Nicira to VMware in 2012 for that amount. He and fellow panelists agreed that Open Compute Project specs will allow an OCP-specified set of switch designs to be produced by a variety of manufacturers and used in software-defined networks.

Several new members of the OCP Foundation were announced at the summit. Microsoft's presence adds another big x86 hardware user, possibly making OCP seem more of an organization with "Intel inside," but open-source hardware is in actuality a chip-agnostic movement.

Gathering in the wings of the fifth annual OCP Summit was a powerful alliance of ARM-device manufacturers. The ARM chip is found in many mobile devices and is noted for its low power consumption. While normally competitors, the ARM manufacturers agreed to unite behind a shared specification for datacenter servers. The Server Base System Architecture will force ARM server makers to concentrate on value-added features to a standard server. Software developers will then be able to produce applications for a broad range of servers, built on a common hardware base. The aim of the architecture is to prevent developers from having to port their applications to each manufacturer. If it succeeds, it's likely to greatly increase the number of applications available to run on ARM as it emerges as a server CPU.

Ian Drew, ARM's CMO, said in a keynote session Wednesday that he had to "knock heads together" to get the likes of HP, Applied Micro, Dell, Citrix, HP, and Cavium to agree to follow the architecture. But they, along with Linaro and Canonical, which port open-source software to ARM, have agreed to do so. Intel's dominance of the x86 space and, consequently, cloud servers, has prompted the ARM manufacturers to compete less with each other and confront the threat that Intel dominance represents to them.

Drew predicted that ARM manufacturers will produce the servers that populate the cloud datacenters of the future due to their smaller footprints, energy efficiency, and compatibility with mobile devices.

Andrew Feldman, corporate VP of AMD, said in a keynote Tuesday: "By 2019, ARM will command 25% of the server market," and custom ARM CPUs or purpose-designed ARM chips "will be the norm for mega-datacenters," such as those used by Facebook, Microsoft, Google, and Amazon.

AMD announced Tuesday that it has produced its first 64-bit ARM motherboard, Seattle, for use in such servers. ARM has had an extremely limited presence in the datacenter because ARM chips have been largely 32-bit designs. Feldman also acknowledged that the ecosystem of software developers for ARM "is in its gangly youth" compared to the mature x86 software market.

Andy Bechtolsheim, a founder of Sun Microsystems, now CEO of high-performance switch developer Arista, scoffed at earlier predictions that 32-bit ARM chips would gain market share. Rather, 64-bit chips and better development tools and software development kits from manufacturers could start to turn that around. Meanwhile, Intel will drive down x86 chips' power consumption in its relentless drive to put x86 into all kinds of servers and devices.

Feldman was unfazed. "AMD will be a leader in ARM servers," he predicted, and will find itself, like Intel, part of an expanding universe.

Charles Babcock is an editor-at-large for InformationWeek, having joined the publication in 2003. He is the former editor-in-chief of Digital News, former software editor of Computerworld, and former technology editor of Interactive Week.

Find out how a government program is putting cloud computing on the fast track to better security. Also in the Cloud Security issue of InformationWeek.