There is a joke among some software-defined networking and OpenFlow insiders here at the Open Networking Summit in San Jose: The Open Networking Foundation (ONF) will not be like the IETF -- it will not develop thousands of complicated protocols and standards based on one company's proprietary method; it will leave as much open for development as possible.
Dan Pitt, ONF executive director, doesn't crack this joke but does make it clear that the foundation is about much more than developing the OpenFlow standard. The ONF is as much focused on fostering the kind of open application-development community for networking that can be found in other parts of the open-source software world. This is a massive cultural shift for a proprietary ASICs and hardware-oriented networking market.
It's not that Pitt is anti-standard. With software-defined networking, the control plane is separated from the physical network and can separately control every flow on the network, depending on the need of the applications that reside in the upper layers. In this scenario, the only place where a standard must come into play is in the language the controller uses to translate information from the applications to the underlying physical and virtual switches. That's where the OpenFlow standard comes in, Pitt says. Above that, in the application layer, anything goes.
So this week at the Open Networking Summit, there's lots of talk about these potential applications doing everything from prioritizing video and unified-communications traffic to controlling mobile access and managing intricate security strategies that differ for every tenant on a network.
Actual product releases are still scarce here at the summit, and there's some anxiety about the lack of OpenFlow-friendly switches available (especially Cisco's lack of commitment) to ultimately support the ecosystem of applications. Yet, at this year's summit, conversation has evolved from defining OpenFlow and SDN to laying out actual use cases and how the technology can address customer need, Pitt said. The next step will be product releases.
One thing is clear: There's no lack of excitement around the Open Networking Foundation and OpenFlow. In the year since the ONF was founded, membership has grown to 66, from 17, and members include every major networking vendor, including Cisco, Google, Yahoo, Facebook and VMware. The Open Networking Summit, with upward of 800 attendees, is completely sold out.
Pitt talked to SearchNetworking about ONF and OpenFlow's general direction.
Many companies at the Open Networking Summit will talk software-defined networking controllers and applications, but these won't work without OpenFlow-friendly switches. Where are we in that process?
Dan Pitt: We had a plugfest last month in which we had at least a dozen switch code bases and four controller code bases and the flow visor. What we're seeing is that every company we talk to is doing something in terms of product development or quiet trials to be prepared for this market. What I am hearing from the vendors is that their customers are asking them what their capabilities are for OpenFlow and software-defined networking. Everybody is trying to figure out how to do it, how to get out there first and which customer sectors to attack.
Is there concern that key players, namely Cisco, will not go OpenFlow? If that's the case, will that hinder advancement for the whole ecosystem?
Pitt: I can't speak for what Cisco's plans are. Our experience at ONF is that Cisco is an outstanding technical contributor and technical advisor. At this point, every participant is trying to understand where this is going and how to make it customer-satisfying technology. There are a variety of approaches to software-defined networking, including some proprietary ones. Some of these predate OpenFlow as a popular standard for the communication between the control plane and the forwarding plane.
One thing about all of the incumbents is that they know what their customers' problems are, and they've been constrained by the culture of distributed RFC standards and then having to accommodate their proprietary operating systems and their shipping schedules. Now they can just write the software that controls the network directly to meet those customer needs. That said, when it comes to open software, it's anybody's ballgame. There will be lots of players, and to the benefit of the customers, there will be competition to see who can provide this software and this customization aspect fastest.
Last year we heard about the basics of separating the control plane in SDN. But now I am hearing more about specific applications emerging. Can you talk about what these applications will be?
Pitt: With software-defined networking, you can write software to make the network do exactly what you want it to do. So we have this logically centralized control plane that we call a controller, which is just a software function.
The controller conveys to the switches what they should do with traffic when it comes in. So, for example, if you want to multicast today, you have all these protocols to configure these trees, and it's so complex that nobody uses it. If you want to multicast a flow with OpenFlow and SDN, the controller loads the flow pattern into the switches, so when a packet comes in, it sends it downstream to the switches and ports from there. There's no need for the network to configure itself. It's just direct programming of the routing algorithm.
You can have a module above the controller that dictates access control [using user-based policy], or traffic engineering, or security, or compliance. Those are modules that influence calculation of the paths through the network.
And what's more important is that software in the control plane can be written by anybody. An operator can write their own software, a vendor can supply their own software, independent software vendors who do nothing but software can supply these as products. Enterprises can hire their own staff or contractors to write this software. I think there will be a big market for networking apps once we have common agreed APIs [application programming interfaces].
Application development communities have not been part of the networking culture. Will networking pros easily make that cultural shift?
Pitt: This is part of what's fundamentally different and exciting about networking. It's taking networking into the realm of software, and it will be a big cultural change. We have been hardware- and protocol-oriented, and now we're going to be software- and API-oriented. We will create standards like we're doing for OpenFlow, but that doesn't mean that all we do is create standards. We are going to standardize as little as necessary. [In the past] the market has arrived at conventions before a smoke-filled room of a standards committee. We are encouraging [a move away from that] into software culture.
What's different at Open Networking Summit 2012, compared to a year ago?
Pitt: A year ago everyone was excited by the abstract notion of this stuff. This week we will see a lot of progress about making this real.
What does "making it real" mean at this point? Does that mean product announcements?
Pitt: That means people finding controllers and switches. We're seeing implementations in use of merchant silicon, ASICs, network processors and purely in software. The basic thing about controllers is what do they contain? What are the boundaries of a controller? Nobody knows. BigSwitch has already released an open-source basic controller, which translates logical directives into flow tables and switches. We'll see more of how these pieces come together to be offered commercially. Then I am looking forward to hearing how the carriers are adopting it and deploying it.
I know there is a difference in adoption need (and therefore uptake) in SDN in the enterprise vs. service providers. For service providers, there is a pain point, so they need SDN. But for the enterprise, that may not happen for a while. What will change that?
Pitt: We talk to people in the financial services community, and they are under these compliance mandates where they have to separate investment banking from commercial banking and their money from their customer's money, so they are running completely different infrastructures that are very expensive. They are looking at this as a way of having a common physical infrastructure with all those cost benefits. They can now have separate logical overlays that are auditable. When you are talking to enterprises, you have to consider different sizes of enterprises. The larger enterprises are going to [do this] first. Smaller and medium companies are going to look for cloud providers to build and manage their plumbing for them. It's just not part of their core competence.
The Open Networking Research Center (ONRC) was announced last week. What will the relationship be between the ONRC and the Open Networking Foundation?
Pitt: We don't know the nature of a formal relationship yet, but we are intending to be partners. They are on the research side, and we are on the commercialization side, so we have complementary roles.