OpenFlow is just the start. What networking really needs is to become more like computer science, researchers argue at a Stanford get-together

Craig Matsumoto, Editor-in-Chief, Light Reading

October 20, 2011

5 Min Read
The Software Revolution Is Coming

STANFORD, Calif. -- One day, Nikhil Handigol got the idea for trying a different kind of load balancing. Normally, load balancers direct traffic to the lightest-loaded server, but they have no idea how congested the path to that server is. Handigol wondered if he could improve performance by taking the path into account.

Ideas like that can be a dime a dozen, but Handigol, a graduate student at Stanford University , got to see his come to life, first on Stanford's network, then nationwide on the GENI test network.

"The point here is: For the first time I've seen, a graduate student was able to take an idea and run it on a national network," said Nick McKeown, the Stanford professor behind the creation of the OpenFlow protocol.

To McKeown, who was speaking at the Open Networking Summit Tuesday, that's part of the value of software-defined networking and of OpenFlow in particular. They open up DIY possibilities that, in turn, could spark changes in the networking business. But like any movement that promises big change, software-defined networking first has to convince a lot of people to change the way they're thinking.

That was a major theme at the Summit, a three-day event where 400 attendees -- including about 50 from Cisco Systems Inc. (Nasdaq: CSCO), according to organizers -- packed into one of Stanford's most modern buildings to rub shoulders with OpenFlow's elite and to discuss the long-term future of networks.

OpenFlow, a protocol that lets operators reprogram a network's control plane, has been getting lots of attention. But it's just one corner of software-defined networking (SDN) that, in general terms, is about turning the network into a programmable entity.

The concept of software-defined networking has been around for a very long time, but it's suddenly become hip. That's partly because it could be an avenue to reawaken the startup-building machine in networking, continuing the pattern set by OpenFlow startups Big Switch Networks and Nicira Networks Inc. Some also see the potential for new business models in companies old and new. Consulting, for instance -- companies could use SDN expertise to execute some of the customized networking cases that operators dream up.

SDN could also lead to a network that's easier and safer to tweak, opening myriad pathways of innovation; think of Handigol's experiment as the SDN equivalent of a smartphone app. The innovative potential is already playing out with OpenFlow, which has made tremendous progress in the past two years not through standards-pushing, but in that goofy open-source way of letting anyone play around with the technology.

"It all happened through the broader community and not some standards effort," said Martin Casado, a founder of Nicira, in his Summit kickoff speech.

But beyond the business angles, software-defined networking is part of a bigger issue: that unlike computer science, networking has never become a discipline that's been boiled down to some fundamental principles.

The abstract truth
It might seem like a trivial distinction, but in a Tuesday morning talk, researcher Scott Shenker of the University of California at Berkeley made a convincing case that networking's future lies in developing a software-like attitude. And that should scare a lot of networking experts, not to mention companies.

In software, he said, layers upon layers of abstraction have been built up, so that tough problems can be defined and solved in simple terms. That's led to software being boiled down to foundational principles -- provable cases that form the basis for the field.

Networking doesn't have that. It's got abstraction of data forwarding -- that's the seven-layer OSI stack we're all familiar with. But there's no such simplification in the control plane.

So, in a sense, networking "is an utter academic failure," Shenker said. "We built an artifact. We did not build a discipline." And instead of fundamental principles to pass on, networking professors teach "a big bag of protocols."

The problem isn't only that networking is complex; it's that the people mastering the complexity are still the ones who get rewarded, Shenker said. To argue why that's bad, he used the analogy of the personal computer: The person you want to have designing the interface is the one who always hated typing DOS commands.

Networking needs to divide and conquer its functions the way software has, Shenker said. That means developing grand abstractions -- one object that simply represents the state of the network, for example, rather than thinking of different nodes having their own states. By dividing the problem, it becomes possible to tackle bigger problems.

New worlds
But these abstractions mean adding elements of centralized control to the network, and that would be a major change for every vendor in the business. "This spells the end for distributed protocols," Shenker said.

The transition could also cause trouble for the people in networking, because software has developed a different mode of thinking, where standards aren't so important and where complexity is meant not to be mastered, but to be obscured.

The process is going to take years, but the advocates of software-defined networking think it's necessary and even inevitable. "It's a software-oriented culture that's going to rule our field," Shenker said.

— Craig Matsumoto, West Coast Editor, Light Reading

About the Author(s)

Craig Matsumoto

Editor-in-Chief, Light Reading

Yes, THAT Craig Matsumoto – who used to be at Light Reading from 2002 until 2013 and then went away and did other stuff and now HE'S BACK! As Editor-in-Chief. Go Craig!!

Subscribe and receive the latest news from the industry.
Join 62,000+ members. Yes it's completely free.

You May Also Like