March 19, 2008: Features

Rethinking the Internet
Some computer scientists think it’s time for an overhaul

By Kenneth Chang ’87

Reinvent the Internet.

That’s what some computer-networking researchers would like to do — and they’re hoping that the National Science Foundation will invest several hundred million dollars to build a testing ground where they can try out their ideas.

The Internet, which perhaps has transformed society more than any other invention of the late 20th century, is far from crippled. Many millions of people tap in every day. It has created gargantuan and growing swaths of the economy that did not exist a couple of decades ago. So why reinvent something that’s not broken?

One answer: What is good can always be better. Innovations that have so changed modern life — like Skype, Napster, and even Google, for example — reflect applications on the edges of the Internet. Larry L. Peterson, the chairman of Princeton’s computer science department, and other researchers believe that experimenting on the Internet’s core could lead to a faster, more secure, more robust, and more flexible network, giving rise to new uses that the current Internet cannot reliably deliver. Now, data are encoded as light pulses that travel through glass fibers rather than as electrical signals in old-fashioned copper wires — but one can imagine that the network could be fine-tuned to take advantage of the unique properties of light. And perhaps, some computer scientists suggest, a more advanced Internet could better serve less-developed parts of the world — where network connections are intermittent — with improved handling of data.

But how to get there? The most innovative experiments cannot be done on the existing Internet, because industry — which has played a dominant role in its growth and direction — relies on it to operate smoothly, all the time. “Often people kill off interesting lines of inquiry because they aren’t compatible with the Internet as it exists today,” says Jennifer Rexford ’91, a Princeton professor of computer science.

Instead of merely jury-rigging fixes into the existing Internet, Peterson and Rexford believe much can be learned about possible improvements by designing a new network from the ground up — one that parallels the existing Internet, on which researchers can run their most innovative experiments. Both are key members of the planning group working to create such a network, a project known as the Global Environment for Network Innovations — or GENI, for short. The National Science Foundation already has established a GENI project office, run by BBN Technologies of Cambridge, Mass., the same firm — then known as Bolt Beranek and Newman Inc. — that the Defense Department tapped in 1969 to design the forerunner of the Internet, called the ARPANET. (ARPA stood for Advanced Research Projects Agency, the branch of the Department of Defense that funds research aimed at revolutionary breakthroughs useful for the military. Today, it’s known as DARPA, with “Defense” added to the name.)

Over the next few years, BBN will flesh out a proposal for what GENI would consist of and the kinds of research it would be used for. Then the National Science Foundation will decide whether to actually build it, with new computers, fibers, and switching equipment — at a price tag estimated to be between $300 million and $500 million. GENI, if it becomes a reality, probably would take three years to build, and could open as a research facility around 2013 for computer scientists to begin running their experiments.

Peterson calls the project a “moon shot” for a field where innovation often has been made on modest means, by one or two people working in a basement or garage, not by giant, expensive collaborations. (Think Steve Jobs and Steve Wozniak putting together their first computer in the Jobs family garage, or Marc Andreesen writing the first graphical Web browser while an undergraduate at the University of Illinois.) The project would mark computer science’s entry into the arena of Big Science, like the giant particle colliders of physicists and the genome projects of biologists.

“Computer science has never done this,” says Peterson, who already has initiated two smaller efforts at Princeton designed to jump-start networking research. “If you look at the other sciences, they build large scientific instruments all the time. There’s a pipeline of projects to study one scientific question or another.”

Before explaining how GENI looks to reinvent the Internet, it is perhaps necessary to explain what the Internet is. For most people, the Internet is a magical conduit that fetches information from Somewhere Out There and plops it on the computer screen. The Internet is routinely confounded with the World Wide Web, the most familiar application that runs on the Internet. That’s like confusing cars with the road beneath them.

The Internet does not work the way the other familiar communications network — the telephone system — traditionally has worked. When you place a phone call, there is a circuit connecting your phone with the phone of the person to whom you’re talking. The telephone system has an elaborate switching system for opening circuits for each phone call.

But the Internet is, in a rough analogy, more like the U.S. mail. Write a letter — the old-fashioned kind, using ink on paper. Then cut it up into pieces. Place each piece of paper in an envelope, and address the envelopes to the same recipient. Drop the envelopes into a mailbox. A postal carrier picks up the envelopes and takes them to the post office. Based on the postal address, the envelopes will be routed to the appropriate truck headed to another post office. After a few more jumps, another postal carrier will deliver the envelopes to their destination, where the recipient can open them up and piece together your original letter.

That, in oversimplified essence, is how the Internet works. In Internet parlance, a “packet” is the equivalent of the piece of paper in the envelope, and the “header” of the packet provides the address of the computer to which it’s headed. Instead of post offices, devices called routers send the packets merrily along their way.

The genesis of the Internet is in large part due to Robert E. Kahn *64. After receiving his master’s and doctoral degrees in electrical engineering from Princeton, he worked at Bell Laboratories, then became an assistant professor of electrical engineering at MIT. He took a leave of absence from his faculty post for his first foray into designing computer networks, working for Bolt Beranek and Newman Inc. on a proposal to build the ARPANET. He never returned to MIT, staying at BBN when it won the contract.

In 1972, Kahn left BBN to head the Information Processing Techniques Office at ARPA, located in Virginia, just outside Washington, D.C., and there he started the Internet. By then, the first segments of the ARPANET, one of the first networks to connect computers at widely separated locations, were up and running. Plans had been drawn up to extend the network to Europe via satellite. Kahn started another project called Packet Radio — a network of computers communicating with each other via radio signals. For the military, that could enable computer links to Navy ships and Army infantrymen carrying computers in their backpacks.

Kahn had a big idea: He imagined linking everything together, so that the backpack computers and shipboard computers also could talk to the military mainframes on the ARPANET. He enlisted the help of Vinton G. Cerf, a computer scientist at Stanford who also had worked on ARPANET. In 1974, they published a paper that described the underlying protocols for the Internet, which are now known as TCP/IP. (TCP is “Transmission Control Protocol,” which specifies the set of rules for two computers to talk to each other; IP is “Internet Protocol,” which defines an address for each computer.)

Today, the fundamental idea of the Internet — everything connected to everything — seems quaint and obvious. But it was not self-evident then. In 1974, the information superhighway was still just a few dirt roads. The king of computing was the mainframe computer, the minicomputer had just been invented, and the personal computer would not be invented for a few more years. ARPANET was the only computer network, and computers on the ARPANET numbered in the tens.

Just as IBM mainframes could not run software written for competing mainframes from Burroughs, NCR, and Honeywell, IBM was not interested in developing technologies that would make it easier for its computers to talk to other computers. Perhaps in a different world, if computer companies and not DARPA had laid the groundwork, there might be several overlapping and incompatible versions of the Internet, each connecting a different brand of computer.

In any case, when Kahn and Cerf got to work, the idea of a wider Internet was not one that intrigued many. That made the task of inventing the Internet easier. “We were very fortunate to be at a place and time when the technology allowed this to happen, we had the resources, and nobody cared,” Kahn said during a September panel discussion in Princeton. The event, called “Re-imagining the Internet,” also featured Peterson and Rexford, and marked Peterson’s ascension to a new endowed chair named in honor of Kahn.

The brilliant simplicity of Kahn and Cerf’s idea can be seen in its name: Internet. Inter Net. Interconnection between networks — a notion that’s essential because different types of networks can operate quite differently. How data fly across a cell-phone data network is quite different from how Ethernet connects computers on a small computer network. In essence, one is speaking Chinese and the other is speaking Swahili; they cannot directly talk to each other. What Kahn and Cerf did essentially was to create an Esperanto that all computer networks, despite their different underlying protocols, could understand. That allowed the ARPANET, and many other computer networks, to join the Internet later and become an interconnected larger network.

Looking back, Kahn says that if he knew then what the Internet would become, he would have made some changes. He would have provided for a larger, more flexible system for assigning network addresses, because the numbering system for computers on the Internet will run out of numbers — in much the same way that a shortage of telephone numbers required rule changes and additional area codes. He also would have worked on establishing a check that the sender of data was indeed the source whom it claimed to be, which would have made many of the hacker attacks today much more difficult.

The fact that Kahn and Cerf’s Internet protocols made so few assumptions about the underlying hardware has allowed the physical networks to change entirely — from the copper wires of the 1970s to the optical fibers and Wi-Fi of today — while the protocols, with some updating, still work. It’s almost as if the street signs and traffic rules designed before the invention of the automobile still worked effectively with today’s Interstate highways. The flexibility has enabled programmers to invent the World Wide Web, video conferencing, music sharing, and other programs without altering the network itself.

Still, over the years, as the Internet outgrew some of its original design, a committee of experts would gather to fix emerging problems. It was not easy: Efforts to implement even small adjustments often ran into opposition from one or more of the many parties that now have a stake in the Internet — from Cisco, the maker of network-routing equipment, to the commercial companies that now run most of the Internet, to other nations. If Cisco declined to implement a change that could speed the Internet, for example, then the good idea likely would fall by the wayside. Other impediments have been geographical, with some countries complaining that the United States still exerts too much control over the now-international Internet. Some researchers — including Peterson, who served on one of the task forces — wondered if this growing patchwork of small fixes was the best approach or whether it was time for the Internet to undergo a gut renovation.

Peterson recalls the frustration he felt as the task-force discussions progressed. “These are the best and the brightest, who have been thinking about the Internet over the past 20-some years,” he says. “And we’re spending the entire meeting trying to figure out how we can convince Cisco to reinterpret one bit in the IP header. That’s what we’ve been reduced to, and that’s when I realized there’s got to be a better way. My conclusion was, OK, the Internet’s done. Now what can we do?”

So in an effort to start from scratch, about four years ago Peterson co-founded PlanetLab, a Princeton-led project that now allows researchers around the world to test new networking applications. There is a catch. PlanetLab is what is known as an “overlay” network: The new protocols are translated into the old Internet ones, and the data are sent across the same Internet connections. If the Internet clogs up, so does PlanetLab.

Peterson also is participating in a follow-up project called VINI, short for VIrtual Network Infrastructure. VINI allows researchers to tap into routers and start inserting their own programs — a sort of mini-GENI. But that depends on the kindness of the routers’ owners. “Doing it on a shoestring is encumbering what we can accomplish,” he says.

GENI, Peterson says, would allow innovation not just at the edges of the network, but also in the guts. “You could ... do experiments directly on top of the circuits instead of directly on the Internet” — a big advantage, he says.

Kahn is not against change or improvements. A year after leaving DARPA in 1985, he founded the nonprofit Corporation for Network Research Initiatives, and he is still chairman, chief executive, and president of the organization. At first glance, CNRI’s endeavors sound similar to GENI’s, and in the 1990s the company worked on building a prototype high-speed computer network. But its portfolio is wider, and it also has consulted about nanotechnology, sponsored development of a programming language called Python, looked at how best to store information on the Internet, and financed development of software for visualizing scientific data.

Kahn supports the goals of GENI, but is not yet convinced that it is worth its expensive price tag. He says he has not seen the research proposals that would require GENI to be carried out, though he acknowledges that “that doesn’t mean they’re not out there.” Beyond the cost of building the GENI test bed, it might cost $30 million to $75 million a year to run, given the equipment and people needed to maintain it. That is money that otherwise would go to research grants. “It’s a big tax on the research community going forward,” says Kahn, who wonders if much of the research can be accomplished on PlanetLab-like overlays at much lower costs.

Even if some groundbreaking invention comes out of GENI, will governments and networking companies be convinced to spend the money to upgrade their equipment in a way to adopt the new idea? Kahn isn’t sure. “I really liken it to changing the wings and the engines on a flying aircraft without being able to ever land it,” he says.

Peterson does not prognosticate about what breakthrough application might arise from the research that could proceed if GENI gets off the ground. “It’s not so much solution X or challenge Y,” he says. “It’s how do we accelerate and broaden the ability to innovate on the Internet.” He does allow that the project could enable experimentation to improve network security, recently telling a writer for Princeton’s School of Engineering and Applied Science that “if industry continues to chart the course of the Internet, we won’t ever be able to have a national debate about privacy and security.”

Some computer scientists say the experimentation possible with GENI might lead to vast arrays of sensors continuously monitoring the environment. Rexford, a member of one of GENI’s working groups, envisions the possibility of “wreckless driving” — fleets of robotically driven cars all in constant, instantaneous communication with each other so that when one brakes, others behind it immediately will slow down, too.

One developing problem, she says, is how to handle computers in motion. The Internet was designed for stationary computers, sitting in the same place all of the time. But in a few years, mobile devices — think iPhones, BlackBerries, and the slew of “smart phones” — may outnumber the stationary computers. The current solution is not an efficient one. To return to the postal analogy, it’s a simple forwarding address — as if all of your mail is first delivered to your home and is then forwarded to your vacation location, which adds a delay. That roundabout routing is one reason why it is slower for a Web page to appear on a cell phone than on a desktop computer. And as hand-held computers proliferate, that problem will worsen.

Rexford, who was a network researcher at AT&T before joining Princeton’s faculty, instead would like to explore ways to inform other Internet routers of your changing location so that the data can be routed to you more directly. She can test her ideas via simulation and perhaps run small-scale experiments, but the true efficiency of a new protocol would not be clear until actual people with mobile devices started using it on a wide scale. “Real users, and real network conditions, always stress your ideas in surprising ways you can’t imagine in advance,” she says. GENI would allow her to test these ideas — and others — in ways that the current Internet or even PlanetLab and VINI cannot. That, scientists believe, would turn ideas into innovations more quickly.

The crucial thing now, Rexford and Peterson say, is to put the foundation down right. Others will decide what to build on top.

Kenneth Chang ’87 reports on science for The New York Times.