Cloud enablement

Open Compute Project Takes on Networking

LAS VEGAS -- Interop -- The Open Compute Project, which has brought open-source attitude (and results) to servers and storage, is going to tackle networking. Frank Frankovsky -- the OCP's chairman and, by day, vice president of hardware design at Facebook -- announced the OCP Networking Project at the end of his Interop keynote Wednesday morning. The group's first meeting happens next week at the Massachusetts Institute of Technology (MIT), and, in fine open-source fashion, everyone is welcome.

Frank Frankovsky delivers the good news at Interop. Remember: It's not really open-source
unless guys with beards are involved.
OpenDaylight has already launched as an open-source effort for software-defined networking (SDN), but this is a potentially bigger effort, as it would tackle networking in general. It's likely the project would seek ways to curb the energy usage of networks, for instance. OpenDaylight is a founding member of the OCP Networking Project, as is the Open Networking Foundation (ONF). (See OpenDaylight SDN Group Breaks Cover.) Vendors involved include Big Switch Networks, Broadcom Corp., Cumulus Networks, Facebook, Intel Corp. and Netronome -- those were the ones listed on Frankovsky's slide, anyway, with many others "coming together," he said. Exactly what this means for today's networking vendors is up to them -- at least according to Frankovsky. The OCP so far has shown the extent to which server and storage vendors underestimated the knowledge in their user base, as well as the discontent lingering within that base, he said. Vendors can consider the swelling tide to be a threat or to be a sweet wave they can ride. And there's no question the OCP is on the rise. You can see it in the numbers. The first OCP summit, in late 2011, drew about 200 people; the most recent one, held in Santa Clara, Calif., in January, drew 1,500. The OCP isn't about beating up vendors, though. The real purpose reflects the original goals of open-source: to share knowledge so that more brainpower can be applied to a problem. "The more people we can get to work on these large, complicated problems, the better," Frankovsky said. "Until you've open-sourced something, you can't get all of the best engineers in the world to work on a problem." — Craig Matsumoto, Managing Editor, Light Reading
Telco 5/9/2013 | 2:41:32 PM
re: Open Compute Project Takes on Networking You are right, ONF referencing OpenFlow and OpenStack with Virtual Network Controllers are too big of a packet routing issue for all but the largest enterprises and carriers. The only benefit to mass equipment sales is further integration of Domain Switching with inter-domain routing that we already have. That is why we have Google already making the swap in January, completely unannounced. It is required, it is their secret sauce and it has not benefit in helping others achieve.

As an analogy, when we sought to sync Ethernet/IP, IETF NTPv4 correctly steered TicToc committee to recognize time stamping packets to 8 decimels was required only by service providers. Thus the verbage is there but not adopted by anyone but OpenCable Labs DTI and MEF 22/8 when requiring SyncE and IEEE-1588.

There is a small community of those who require very robust use; while the mass market only needs a lighter-weight if any solution what-so-ever.
jhoosac2 5/9/2013 | 3:10:17 AM
re: Open Compute Project Takes on Networking Vaporware, because first, only super centers will benefit, since they have the resources to "finish the last foot" by fixing/tuning/extending the product. Others can't. Even the super centers themselves may not agree on an open platform. It is too close to their competitive advantage. It's part of their secret sauce.

Secondly, there won't be any good developer community. This stuff has no "social value". You can't watch movie or listen to mp3 on it. It's a niche industry,
requiring niche skills that has no value in career advancement outside of a handful prospect employers. Plus this stuff needs hardware with substantial cost. The chips are not cheap, to start with. It is not pure software like Linux. New release involves new hardware.

And the chips can never be cheap, in relative terms, because, as big as the DC
market is, it is in no way a mass market, not PC, nor mobile devices, not even close. The market volume is very limited. Vendors know this. They have to make money off that many switch ports.

BTW, there has never been a lack of low-priced switches. Ask FB why they don't use Netgear? or ask Netgear to make one for them? Change some fans here, load some new software there, tweak the case, done, right? Why not? Because it doesn't work!

So why the vaporeware? It's a bargaining chip. FB will spend X amount on this project so they can get Y discount from vendors like Cisco. Y > X.
DanJonesLRMobile 5/8/2013 | 9:03:41 PM
re: Open Compute Project Takes on Networking Certainly not unheard of in the tech world already. I suspect that people that want super high performance will still prefer to build their own.
Craig Matsumoto 5/8/2013 | 6:25:26 PM
re: Open Compute Project Takes on Networking Open-source hardware for networking - this is potentially huge, right? Or am I just being taken in by open-source magic and Frank's beard?
Sign In