At Facebook’s Open Compute Summit in New York City today, technology industry titans railed against proprietary, non-interoperable server and data center designs.
Andy Bechtolscheim, chief development officer at Arista Networks but more notably one of the co-founders of Sun Microsystems, encouraged attendees to support emerging Open Compute Project (OCP) standards and the newly formed Open Compute Foundation to speed up the pace of hardware innovation.
“What’s been missing is standards at the systems level,” Bechtolscheim said, resulting in hardware designs filled with “gratuitous differentiation” – functionally similar products that do not interoperate with one another. An example of non-interoperable products is blade servers from vendors like Hewlett-Packard, Cisco and IBM, whose servers are largely similar, but do not fit into one another’s chassis.
“There’s nothing more frustrating to the customer,” Bechtolscheim said. “The status quo, quite frankly, benefits the vendor, not the customer.”
Open Compute Project grows up
Facebook launched the Open Compute Project last April when it open-sourced the proprietary server and data center designs it used to build its energy-efficient Prineville, Ore., data center.
At the time, data center experts took note of Facebook’s interesting, if highly proprietary, server, chassis, power supply and data center electrical and cooling designs. By sharing those designs with the data center community at large, Facebook reasoned it could speed the pace of innovation, particularly in its rarified world of “scale computing,” where football-field data centers are the norm.
But there’s a chance that work being doing in the Open Compute Project will trickle down to hardware designs used by mere mortals. Today, ASUS submitted a new motherboard design, Mellanox announced a 10 Gigabit Ethernet mezzanine adapter for OCP servers, and Red Hat committed to supporting Red Hat Enterprise Linux on certified systems.
Other vendors, meanwhile, joined the foundation, including Dell, Huawei, Arista, Hyve Solutions, and Silicon Mechanics. Board members of Open Compute Foundation vote to approve a project, and members include Arista’s Bechtolscheim; Frank Frankovsky, Facebook director of technical operations; Don Duet, a managing director at Goldman Sachs; Mark Roenigk, Rackspace chief operating officer; and Jason Waxman, general manager for Intel’s data center group.
Open Rack now, open storage, networking, systems management next?
To go along with the new foundation, Open Compute has a new standard under its bailiwick: Open Rack 1.0, describing the server rack, its power distribution and cooling properties.
Other possibilities for the Open Compute Foundation to tackle include storage formats, virtual I/O, data center design for different geographies, and open systems management, all of which were discussed in afternoon breakout sessions.
A larger question for the Open Compute Foundation is whether the open source model pioneered by the Linux community can have as much impact on hardware designs. Facebook, for one, thinks it can.
“When you look at the pace of innovation in software compared with hardware, it’s night and day,” Facebook’s Frankovsky told the audience. “Let’s make it day and day.”