NSF Guru of Net Architecture Makeover Takes Over Stanford Program

Parulkar becomes executive director of Stanford’s Clean Slate Internet Design Program, said McKeown, the program’s founder and director. The program already involves several research projects, including on security and wireless spectrum policy, and could end up trying to offer a full- blown new Internet architecture, he said. Parulkar was a founder and chief technology officer of Growth Networks, which Cisco bought in 2000 for $350 million, according to a resume on his Web site. He also was a board member of NetSift, acquired by Cisco for $30 million in 2005. McKeown told us that Parulkar’s Cisco connections were “pretty much unrelated” to the Clean Slate Program.

The Stanford program “could easily find” federal money “with no strings attached” to cover its $1 million annual budget, but “we very deliberately did not do this,” McKeown said in a campus presentation Friday. Instead, it decided to “work with a focused group of industrial partners” that also includes NEC and Xilinx, he said. Three unspecified companies are in discussions about joining, McKeown said. “Our goal is to have members from across the networking industry,” he said in an e-mail. “We are in no hurry to add additional companies, but plan to do so in the coming year.”

The participating companies and program researchers get to pick each other’s brains, McKeown said in his presentation. The companies have a “vested interest,” like much of the rest of the world, in legacy Internet technology -- but they also have an interest “in how it will be,” he said. A predecessor to the Clean Slate Program got part of a 5 year, $7.5 million NSF grant, according to a 2003 Stanford news release. Its projects aim for “revolutionary” effects on the Internet in 10 to 15 years, McKeown said. The program “exploit[s] Stanford’s breadth and depth” of expertise, which no company or government can rival -- only MIT and the University of California- Berkeley, he said. “Stanford, more than anywhere else, is in a position to have an enormous impact on what the Internet will look like in 10 or 20 years.”

The Internet’s architecture dates from the Ed Sullivan era, when computing was based on IBM mainframes instead of Apple iPhones, McKeown said. Sweeping trends to run “the Internet over everything” and “everything over the Internet” have greatly strained its transport protocols, such as TCP and SIP, he said. The Internet wasn’t set up to be secure or to run video, McKeown said.

The architecture is “stuck -- it’s very to hard to change,” a victim of its own success in creating a huge technology legacy and installed base, McKeown said. The failure of IP multicast and snail’s pace of IPv6 show this, and so do the success of businesses supplying Band-Aids like firewalls and middleware, McKeown said. The cell phone industry has supplied “the most innovation in the field of course in the past ten years,” but mobile companies have been so busy growing, “they can’t really give… the attention they would like to” to replacing the Internet’s architecture, he said.

Led by Parulkar, the NSF has promoted efforts toward blank-slate rethinking of the architecture the past three years, McKeown said. It has prompted several research projects under the umbrella of GENI, the Global Environment for Network Innovation, which NSF has contracted with ARPANet creator BBN. The result is a “sea change in the thinking about network research,” he said.

“There’s chaos” in computer science research, McKeown said. “There’s panic. There’s pandemonium.” With the rethinking, graduate students don’t know what wrong step might keep them from doctorates and instructors worry about getting tenure, he said. There’s “very, very healthy debate and lots of disagreement.”

GENI’s aim is financing an experimental facility to advance radical new ideas, McKeown said. It aims to fill the void created by the inability of any particular technology vendor or venture capitalist to back a promising departure without widespread cooperation throughout the Internet world and without the millions of users needed to test the innovation, he said. GENI has “a very successful first stake in the ground” in PlanetLab at MIT, but that’s limited to a programmable platform, overlaid on the Internet, to run services over, McKeown said.

There’s “pretty widespread agreement” that the work should be organized around existing switching centers and 20 to 50 will be needed for a nationwide test network, McKeown said. Anyone with an account on the network could try out would-be architecture innovations on a virtual “slice” of the system and promote its adoption by others, he said.

The NSF has several hundred million dollars to spend yearly on an important infrastructure project under the Major Research Equipment and Facilities Construction program, McKeown said. He said the award never has gone to information technology, but now the foundation is “trying to decide if and how this would be funded.” The testbed won’t be funded before 2010 or deployed before 2012, McKeown said. Research efforts meanwhile will concentrate on small extensions of PlanetLab’s work, he said. He told us his program doesn’t seek to build the GENI testbed.

It would be “unfortunate” if a new architecture required a “flag day” -- the equivalent of a hot cut in telecom terms -- when everyone on the Internet was expected to switch from the current one, McKeown said in his presentation. But even that shouldn’t deter the necessary effort, he said. If the new architecture can be spelled out, “over time maybe companies will see and plan accordingly.” It doesn’t matter whether organizations such as ICANN, IEEE and the ITU “lag or lead” in the effort, McKeown said. “It will be the ideas that win in the end.”

The Clean Slate Program seeks a network that is “economically sustainable,” unlike the Internet today, McKeown said. “Nobody is actually making any money on deploying the Internet infrastructure today,” he said. Economics dictates that the prices competing providers charge tend toward the cost of adding a user, which is nothing, McKeown said. The only alternatives to putting the Internet infrastructure on a solid business footing are “not palatable at all in the political climate,” McKeown said: “Nationalize the infrastructure… upon which people will compete” or set up a regulated monopoly or duopoly to run it. But how to do it isn’t obvious, he said.

The other goals are transforming the infrastructure by making it “trustworthy” -- secure, robust and manageable -- and having it provide “mobility by default,” “performance to blow your socks off” and support for access technologies and “applications you haven’t thought of,” McKeown said. Higher performance is needed because video accounts for most Internet traffic, “probably 80 to 85 percent,” he said. If one percent of what Netflix ships daily were downloaded instead, “the Internet would shut down,” McKeown said. And within a year, video probably will account for 90 to 95 percent of traffic, he said.

Clean Slate’s strategy is working from the bottom up -- from “innovation in the small” to “themes” and “large collaboration projects” and maybe then a complete Internet architecture, McKeown said. Current Stanford projects include developing a theory for making data “flows not packets” the basis for network traffic. Another aims to propose an alternative to packet switching that would allow the use of optical technology, which promises “enormous capacity” and the reduction of power requirements that impose the greatest limitation on scalable routers.

The wireless-policy project anticipates the opening of new spectrum and seeks to be prepared with policy recommendations and related technology plans, McKeown said. The premise is that spectrum scarcity results from inefficiencies and tight government control, he said. The current availability of only a small amount of spectrum for wireless broadband “is not sustainable” and “is going to change,” McKeown said.

The Clean Slate Project is “already talking to some government agencies about how we can deploy” a security architecture that has been running since winter with about 300 users in a computer sciences building named for Bill Gates, McKeown said. The greatest worry is networks connecting to the Internet, and the first requirement “obviously is being able to identify reliably who is who” on the Internet, he said.