O one owns it. And no one in particular actually runs it. Yet more than half a billion people rely on it as they do a light switch.

The Internet is a network whose many incarnations — as obscure academic playpen, information superhighway, vast marketplace, sci- fi-inspired matrix — have seen it through more than three decades of ceaseless evolution.

In the mid-1990's, a handful of doomsayers predicted that the Internet would melt down under the strain of increased volume. They proved to be false prophets, yet now, as it enters its 33rd year, the Net faces other challenges.

The demands and dangers — sudden, news- driven traffic, security holes, and a clamor for high-speed access to homes — are concerns that bear no resemblance to those that preoccupied the Internet's creators. For all their genius, they failed to see what the Net would become once it left the confines of the university and entered the free market.

Those perils are inextricably linked to what experts consider the Internet's big promise: evolving into an information utility as ubiquitous and accessible as electricity. That, too, was not foreseen by most of the engineers and computer scientists who built the Net in the 1960's and 70's.

Ten years ago, at the end of 1991, the same year that the World Wide Web was put in place but a good two or three years before the term Web browser became part of everyday speech, the Net was home to some 727,000 hosts, or computers with unique Internet Protocol, or I.P., addresses. By the end of 2001, that number had soared to 175 million, according to estimates by Matrix Net Systems, a network measurement business in Austin, Tex.

For all that growth, the Net operates with surprisingly few hiccups, 24 hours a day — and with few visible signs of who is responsible for keeping it that way. There are no vans with Internet Inc. logos at the roadside, no workers in Cyberspace hard hats hovering over manholes.

Such is yet another of the Internet's glorious mysteries. No one really owns the Net, which, as most people know by now, is actually a sprawling collection of networks owned by various telecommunications carriers. The largest, known as backbone providers, include WorldCom (news/quote), Verizon, Sprint and Cable & Wireless (news/quote) USA.

What, then, is the future of this vital public utility? Who determines it? And who is charged with carrying it out?

For the Internet's first 25 years, the United States government ran parts of it, financed network research and in some cases paid companies to build custom equipment to run the network. But in the mid-1990's the Net became a commercial enterprise, and its operation was transferred to private carriers. In the process, most of the government's control evaporated.

Now the network depends on the cooperation and mutual interests of the telecommunications companies. Those so-called backbone providers adhere to what are known as peering arrangements, which are essentially agreements to exchange traffic at no charge.

"Peering fits right in with the overly loose way the Internet is provided," said Scott Bradner, a senior technical consultant at Harvard University, "which is unrelated commercial interests doing their own thing." Mr. Bradner, co-director of the Internet Engineering Task Force, an international self-organized group of network designers, operators and researchers who have set technical standards for the Internet since the late 1980's, said that peering remains a remarkably robust mechanism.

And for now, capacity is not a particularly pressing problem because the backbone providers have been laying high-speed lines at prodigious rates over the last few years.

"We've got a lot of long-distance fiber in the ground, a lot of which isn't being used, but it's available," said Craig Partridge, a chief scientist at BBN Technologies, an engineering company that oversaw the building of the first network switches in the late 1960's and is now owned by Verizon.

Still, the fear that the Net is not up to its unforeseen role still gnaws at prognosticators. Consider the gigalapse prediction.

In December 1995, Robert Metcalfe, who invented the office network technology known as Ethernet, wrote in his column in the industry weekly Infoworld that the Internet was in danger of a vast meltdown.

More specifically, Dr. Metcalfe predicted what he called a gigalapse, or one billion lost user hours resulting from a severed link — for instance, a ruptured connection between a service provider and the rest of the Internet, a backhoe's cutting a cable by mistake or the failure of a router.

The disaster would come by the end of 1996, he said, or he would eat his words.

The gigalapse did not occur, and while delivering the keynote address at an industry conference in 1997, Dr. Metcalfe literally ate his column. "I reached under the podium and pulled out a blender, poured a glass of water, and blended it with the column, poured it into a bowl and ate it with a spoon," he recalled recently.

The failure of Dr. Metcalfe's prediction apparently stemmed from the success of the Net's basic architecture. It was designed as a distributed network rather than a centralized one, with data taking any number of different paths to its destination.

That deceptively simple principle has, time and again, saved the network from failure. When a communications line important to the network's operation goes down, as one did last summer when a freight-train fire in Baltimore damaged a fiber-optic loop, data works its way around the trouble.

It took a far greater crisis to make the Internet's vulnerabilities clearer.

On Sept. 11, within minutes of the terrorist attacks on the World Trade Center, the question was not whether the Internet could handle the sudden wave of traffic, but whether the servers — the computers that deliver content to anyone who requests it by clicking on a Web link — were up to the task.

Executives at CNN.com were among the first to notice the Internet's true Achilles' heel: the communications link to individual sites that become deluged with traffic. CNN.com fixed the problem within a few hours by adding server capacity and moving some of its content to servers operated by Akamai, a company providing distributed network service.

Mr. Bradner said that most large companies have active mirror sites to allow quick downloading of the information on their servers. And as with so many things about the Net, responsibility lies with the service provider. "Whether it's CNN.com or nytimes.com or anyone offering services, they have to design their service to be reliable," he said. "This can never be centralized."

Guidelines can help. Mr. Bradner belongs to a Federal Communications Commission advisory group called the Network Reliability and Operability Council, which just published a set of recommended practices for service providers, including advice on redundant servers, backup generators and reliable power. "Still, there are no requirements," Mr. Bradner said.

If the government is not running things, exactly, at least it is taking a close look.

Dr. Partridge of BBN Technologies recently served on a National Research Council committee that published a report on the Internet. One of the group's main concerns was supplying households with high-speed Internet service, known as broadband.

Some 10.7 million of the nation's households now have such access, or about 16 percent of all households online, according to the Yankee Group, a research firm.

Only when full high-speed access is established nationwide, Mr. Partridge and others say, will the Internet and its multimedia component, the Web, enter the next phase of their evolution.

"We need to make it a normal thing that everyone has high-speed bandwidth," said Brian Carpenter, an engineer at I.B.M. (news/quote) and chairman of the Internet Society, a nonprofit group that coordinates Internet-related projects around the world.

Yet there is no central coordination of broadband deployment. Where, when and how much access is available is up to the individual provider — typically, the phone or cable company. As a result, availability varies widely.

Control falls to the marketplace. And in light of recent bankruptcies and mergers among providers, like Excite@Home's failure and AT&T (news/quote) Broadband's sale to Comcast (news/quote) late last year, universal broadband deployment may be moving further into the future.

The one prominent element of centralized management in Internet operations — the assignment of addresses and top domain names, like .com or .edu — reflects the tricky politics of what is essentially a libertarian arena. That is the task of the Internet Corporation for Assigned Names and Numbers, or Icann, which operates under the auspices of the Commerce Department. Its efforts to establish an open decision-making process became mired in disputes over who the Internet's stakeholders actually were.

And even as Icann and its authorized registrars take over administration of the Internet's naming system, a different problem nags at computer scientists: the finite number of underlying I.P. addresses.

In the current version of Internet Protocol, the software for the routers that direct Internet traffic, there is a theoretical limit of four billion addresses. Some 25 percent are already spoken for.

The solution, Mr. Carpenter said, is bigger addresses. "This means rolling out a whole new version of I.P.," he said.

Although the assignment of I.P. addresses falls to Icann, inventing a new protocol is essentially a research problem that falls to the Internet Engineering Task Force.

As the Internet continues to grow and sprawl, security is also a nagging concern. The Internet was not built to be secure in the first place: its openness is its core strength and its most conspicuous weakness.

"Security is hard — not only for the designers, to make sure a system is secure, but for users, because it gets in the way of making things easy," Mr. Bradner said.

There is no centralized or even far-flung security management for the Internet. The Computer Emergency Response Team at Carnegie Mellon University is mainly a voluntary clearinghouse for information about security problems in Internet software.

The lack of a central security mechanism "is a mixed bag," Mr. Bradner said. A centralized system that could authenticate the origin of all traffic would be useful in tracing the source of an attack, he said.

That is where a delicate balance must be struck: between the ability to trace traffic and the desire to protect an individual's privacy or a corporation's data. "It's not at all clear that there's a centralizable role, or that there's a role government could play without posing a severe threat to individuals," Mr. Bradner said.

Past plans for identity verification have failed because of the complexity of making them work on a global scale, he said.

Such are the challenges that face the Internet as it continues its march.

"The really interesting question to ask is whether we can build a next generation of applications," Mr. Carpenter said. "Can we move from what we have now, which is an information source, to a network that's really an information utility, used for entertainment, education and commercial activities? There's tremendous potential here, but we've got a lot of work to do."

As that work progresses, another question centers on what role the government should play. Many carriers who bear the cost of expanding the infrastructure favor federal incentives for carriers to invest in new broadband technology. The Federal Communications Commission is also mulling policy changes, soliciting suggestions from the communications industry for making broadband access more widely available.

Dr. Metcalfe predicts that the next big step is what he calls the video Internet. "We're done with just voice and text," he said. "No one is quite sure what the killer app will be, but we want to see stuff move, and we want it to be better than television."

Despite his joke about eating his words, Dr. Metcalfe said he was unrepentant about his forecast of a gigalapse.

"There's a gigalapse in our future," he said. "The Net's getting bigger all the time and there are basic fragilities." Since there is no formal tracking mechanism for connection failures, he argues, his gigalapse may very well have happened already without anyone noticing.

"I'm sure there are outages every day, but because of the Internet's robust nature they are generally not noticed," he said. "We do control-alt-delete and chant, and eventually the connection comes back."

Indeed it does.