Facebook-Led Open Compute Project Eyes 4 Network Specs

The group will consider three switch specs from Intel, Broadcom and Mellanox, and networking software from Cumulus.

The Open Compute Project, which in May turned its sights on data center networks, is making significant progress in developing an operating-agnostic networking switch, according to the project's top official.

According to Frank Frankovsky, vice president of hardware design and supply chain for Facebook and president and chairman of the Open Compute Project (OCP), the group is preparing to vote on four contributions to the effort submitted by several technology vendors.

The goal is to create a top-of-rack switch that can run any operating system, a move that OCP participants hope will make networks as dynamic and flexible as virtualization has made servers and storage. It also would create a challenge to Cisco Systems, Juniper Networks and other vendors that make a lot of money by selling expensive and complex switches that run their own operating system.

"Such [an open] switch … would enable a faster pace of innovation in the development of networking hardware; help software-defined networking continue to evolve and flourish; and ultimately provide consumers of these technologies with the freedom they need to build infrastructures that are flexible, scalable, and efficient across the entire stack," Frankovsky wrote in a Nov. 11 post on the OCP blog.

Hundreds of people are involved in the more than 30 potential contributions that are people-developed for the Facebook-led organization, and four of those contributions—from Intel, Broadcom, Cumulus Networks and Mellanox Technologies—most likely will soon be accepted by members of the OCP's Contribution Committee, he wrote.

Intel has created a specification for an open switch that officials with the giant chip maker said will be for a bare-metal top-of-rack switch that can run 10 Gigabit Ethernet or 40GbE. The goal is to offer organizations more networking choice, greater flexibility and improved costs. Intel officials also said they bring with them a growing partner ecosystem to support the spec.

"The specification describes a 48x4 10/40G switch including all necessary subsystems for switching, control CPU, peripherals, external interfaces, power, cooling, and mechanical enclosure," Intel said in describing its contribution.

Intel's efforts in the OCP's networking initiative illustrate the vendor's desires to become a larger player in the data center beyond just supplying the silicon for systems.

Broadcom also is offering a spec for an open switch based on leaf-and-spine configurations. It leverages Broadcom's latest Trident switch architecture—Trident II—which is being embraced by many networking technology vendors. The Trident architecture supports a range of networking OSes and applications.

Mellanox also is offering a spec for an open switch, based on its SwitchX-2 x86-based top-of-rack switch specification. According to the company's description, its switch "supports 48 SFP+ ports and 12 QSFP ports, enabling non-blocking connectivity within the OCP Open Rack, or alternatively, enabling 60 10GbE server ports when using QSFP+ to SFP+ breakout cables to increase rack efficiency for less bandwidth demanding applications." It should offer better power consumption, latency and density, and lead to larger and more cost- and energy-efficient data center designs, officials said.

Cumulus is offering its Open Network Install Environment (ONIE) software as a boot loader that will make it easier to install software on open switches, according to the company. ONIE supports multiple network operating systems that gives organizations greater control and choice in their networks, officials said.

Facebook kicked off the OCP in 2011 in an effort to create open standards for highly efficient data centers and IT hardware. The group has worked on servers and storage in the past, and now is looking at networks. Facebook and other Web-based companies, such as Google and Amazon, are always looking for more power-efficient systems to help keep costs down in their massive, dense data centers. They haven't always been able to find the resources they need from other vendors, and at times have developed their own servers, storage appliances and power supplies made from off-the-shelf components.

They're also embracing new technologies like software-defined networks, in which network intelligence is moved from expensive hardware and housed in software.