Open Compute Project To Develop A Network Switch, A First-Of-Its-Kind Open Source Project
Open Compute will develop a specification and a reference box for an open networking switch and will do so from the ground up in the fashion of open-source software efforts, such as those developed by the Apache Foundation.
The OS-agnostic, top-of-rack switch will be the first developed as an open-source project with the spec developed by the Open Compute community.
“Closesd switches are still the primary way things work,” said Frank Frankovsky in an interview this week. Frankovsky is a Facebook vice president in hardware design and supply operations who plays a focal role at Open Compute. “…Networking has always had a black box nature to it. You give it a packet and it gives it back on the other end.”
According to a blog post by Frankovsky, Najam Ahmad, who runs the network engineering team at Facebook, will lead the networking project. The Open Networking Foundation and OpenDaylight group will participate with Broadcomm, Intel, VMware, and Cumulus Networks. Work on the project will begin at the first OCP Engineering Summit, being held at MIT on May 16.
The networking specs will most likely be stripped down, compared to proprietary switches. “Scale is the hardest problem to solve,” Frankovsky said. “Simplicity allows us to trouble shoot faster.”
The network switch will be designed to be independent of the software that runs on top of it. That means customers can configure the technology in the manner that is appropriate for their purposes. The hope is that this “disaggregated,” switch will allow for a faster pace of innovation. Frankovsky said that Facebook has found networking to be a bottleneck. The proprietary switches are not built for scale. More so they are meant for small clusters. Additionally, changes do not happen fast enough in the switches the network vendors offer. Requests go to the vendor who then prioritizes additions in the new version.
There have been open-source hardware projects such as Raspberry Pi but none at the data center level. The need for open-source is precipitated by the surging growth of the Internet, which is forcing companies to process ever larger stores of data.
There are some companies not on the list of participants in the project. Dell, HP and Cisco are noticeably not present. My guess is we can expect to see some of those vendors joining in this open-source effort, especially as demand increases for infrastructure that is not locked down in proprietary fashion but is rather more plug and play.