Cutting-edge engineering feats will likely continue to come from established hardware vendors. Consider one recent example, in which IBM redesigned solid state disk so that it can fit into the dual in-line memory module slots on a standard motherboard. IBM moved the fastest storage available inside the server case; that's proprietary engineering that only a few manufacturers are capable of undertaking. OCP is more about making the most cost-effective use of existing standard components, not doing expensive reengineering.
Intel, though a member of Open Compute, thinks companies will continue to pay for the sophisticated integration and testing done by HP, Dell, and Lenovo, whether they're building boxes with their brand names on them or not. "Vanity is often in the eyes of the beholder," warns Eric Hooper, director of cloud service optimization for OEMs and open design manufacturers at Intel.
In an interview with InformationWeek last fall, Cisco CEO John Chambers dismissed the white-box approach in networking -- where the hardware is dumb and controlled entirely by a software overlay -- as "fatally flawed." Most big companies look at the total cost of networking, including integration expense and technology upgrades over time.
"What customers are after is time-to-applications and future-proofing their environments, whereas the majority of the cost is combining vendors that weren't designed to work together," Chambers said. "And the only thing worse than vendors that weren't designed to work together are white boxes that weren't designed to work together."
Brady says Fidelity doesn't expect or get the same level of support from white-box suppliers that it gets from HP or Dell. But then again, the white boxes are part of Fidelity's cloud operation, built on the premise that when a piece of hardware fails, the rest of the servers pick up the slack while the failed machine is replaced.
IT leaders at Fidelity and Goldman say OCP will help them get to a simpler, easier-to-manage environment. A standard management interface for a full rack of 42 servers, for example, would let them update or reconfigure those servers at the same time, rather than one after the other, as is usually done today. Simplified management in turn will exert a continuing influence over how future datacenter environments are built.
Will new vendors emerge?
Fidelity and Goldman are welcoming new x86 server suppliers, as open-source hardware creates new industry opportunities. Take Avnet Embedded, part of $26 billion-a-year electronics distributor Avnet Inc. Avnet Embedded was already an experienced hardware assembler and integrator when OCP published its motherboard spec in April 2011. But with those specs, Avnet could become a supplier of entire OCP-based servers. "You don't have to be as large as HP, IBM, or Dell. It's more like a cooperative," says DaWane Wanek, director of Avnet datacenter infrastructure and OCP solutions.
Avnet's 228,000-square-foot integration facility in Chandler, Ariz., houses an OCP Innovation Lab and produces OCP servers. The center also integrates and supports Cisco-NetApp FlexPods and HP Converged Infrastructure, and handles custom orders from system architects and independent software vendors. It ships about 475,000 systems a year. The company declined to say how many systems are based on OCP specifications, but it's a small fraction of the total. Avnet operates seven other assembly centers worldwide, including two in China, in Tianjin and Beijing.
In the past, custom server manufacturing involved orders for hundreds or a few thousand units. With OCP, there's the opportunity to follow a standard design and compete for sales to cloud computing and web service providers, online gaming companies, and search engine companies -- organizations that buy servers by the tens of thousands.
Facebook's Frankovsky says the major manufacturers may try to minimize the impact OCP already has had, but he warns that proprietary server-makers ignore open-source hardware at their peril. "This could be the best thing that has ever happened to them," he says, "or it could be the worst."
Frankovsky says he sees enough venture capital and startup interest in Open Compute to counter concerns that there won't be enough implementers of its designs. "All our members are real smart people," Frankovsky says. "I don't think they will sit and be caught on the wrong side of history." Venture capitalist Marc Andreessen was among the speakers at the foundation's Open Compute Summit in San Jose, Calif., Jan. 28 and 29. His firm, Andreessen Horowitz, has invested in Cumulus Networks (and Facebook).
While most of the Open Compute energy has been around servers and motherboards, Fidelity and Goldman are keen on storage and networking as well.
But critics point out that early adopters are companies with a few ultrahigh-capacity applications -- such as Facebook's social network -- and not the wide mix of variable-use applications common outside the web and financial industries. One project, Open Rack, was so specific to Facebook's datacenters that it couldn't be wheeled into any typical company's datacenter. The standard enterprise datacenter rack has 19-inch openings, and Facebook wanted wider openings to accommodate motherboards that had rearranged components for smooth airflow over the surface. The redesign prevented taller components from "shadowing" shorter ones, like CPU chips, lest the chips be deprived of their share of the cooling airflow.
Open Compute's potential beyond financial services is tied to the more general movement toward private cloud datacenters. Private cloud architectures like the one at Fidelity let users provision their own computing capacity as needed, so that the company can react more quickly to changing business conditions. Brady knew Fidelity couldn't get to a private cloud without learning new approaches from leaders such as Facebook. "If we had tried to tinker with the old datacenter, it wouldn't have worked," he says.
Duet at Goldman hasn't converted datacenters to OCP-based designs, but his staff has experimented with prototypes and is ready to adopt open-source hardware on a larger scale. "Management at scale is enormously important to us," Duet says. "We think 2014 will be a breakthrough year."
Read the whole Feb. 3, 2014, issue of InformationWeek.