Open Compute Project Tests Three Open Switch Designs

In just six months, Broadcom, Intel and Mellanox have all submitted independent open switch proposals

The Open Compute Project (OCP), a Facebook-backed initiative that aims to develop cheaper, more energy efficient data centre designs, is currently testing not one, but three different specifications for open network switches.

The reference designs have been proposed by Broadcom, Intel and Mellanox, while Cumulus Networks has contributed its Open Network Install Environment (ONIE) software.

“Taken together, these contributions are tremendous steps forward toward our vision of a truly disaggregated network stack. They are also vivid proof of the OCP community’s ability to work together, in the open, to develop innovative new technologies – and to do so at an almost unheard-of pace,” says Frank Frankovsky, chairman and president of the Open Compute Project Foundation.

Open switches adopted by OCP are expected to compete with proprietary hardware made by companies like Cisco and Juniper Networks.

Facebook and friendz

The OCP was launched by Facebook in 2011 to share custom-made data centre technology and today, it unites more than 50 member companies, including Seagate, EMC, ARM, Dell, Rackspace and Red Hat. In the last two years, the project has produced designs for everything from racks and storage to power supplies, motherboards and interconnects.

Intel switch reference designIn May 2013, OCP announced it would develop an open, OS-agnostic top-of-rack network switch. It was believed that such equipment would help decrease the cost of building networks, and “help software-defined networking continue to evolve and flourish”.

Broadcom was the first to develop a full specification and implementation of an open switch, based on the widely deployed Trident II architecture and addressing the popular leaf and spine configurations. “We believe this specification will enable faster innovation in the market and more choice for data centre operators and telecom service providers,” the company says.

Intel has proposed a bare-metal 48×4 10/40G switch design that includes all necessary subsystems for network switching and controlling the CPU, peripherals, external interfaces, power, cooling, and the mechanical enclosure. It can be built with parts from other vendors, although Intel would obviously want everyone to use its hardware, namely the FM6764 Ethernet switch silicon, developed with SDN in mind.

“To complement the proposed contribution to the OCP working group, Intel brings an ecosystem of partners ready to supply production level systems with a variety of solution capabilities,” it adds.

The specification developed by Mellanox is based on its SwitchX-2 product, and supports 48 SFP+ ports and 12 QSFP ports, enabling non-blocking connectivity within the OCP Open Rack. Alternatively, users can have all 60 10GbE server ports when using QSFP+ to SFP+ breakout cables, increasing rack efficiency for less bandwidth-hungry applications.

Meanwhile, Cumulus Networks has agreed to share its ONIE boot loader, used to install software on network switches. ONIE is already familiar to various hardware vendors across the globe and supports multiple network operating systems, making it perfect for Open Compute. It is worth noting that the switch design from Mellanox is the first to enable ONIE on x86 architecture.

At the moment, it is not clear if the OCP will choose one design, or continue developing several. But considering the pace at which the organisation moves, we should see at least one finished specification early next year.

Meanwhile, Cisco is moving forward with its Application-Centric Infrastructure (ACI) strategy, which combines some SDN elements with optimised hardware and includes a new family of Nexus 9000 switches.

Are you an expert on Facebook? Take our quiz!

Originally published on eWeek.