"We can no longer rely on last-generation technology, which has essentially remained unchanged for 40 years, to power Internet performance," says Roberts, who is 69 years old. In September, his start-up, Anagran Inc., introduced a flow router, equipment that analyzes Web traffic to discern whether it is an e-mail, a movie or a phone call and then carves out the bandwidth needed for transmission.

There's also Len Bosack, the 55-year-old co-founder and former chief technology officer of networking giant Cisco Systems Inc., who helped commercialize routers, the core piece of networking equipment that allows computers to communicate with one another. Yet he now terms such gear "less and less adequate" for today's Internet needs. Also in September, his company, XKL LLC, unveiled a system that allows businesses to connect to underground cables that have nearly 100 times the capacity of current telecommunications pipes.

The actions of Bosack and Roberts fuel the growing debate over whether the Internet's current infrastructure is sufficient to handle the explosion of bandwidth-hungry services such as Internet telephony and video. In a recent report, Cisco calculated that monthly Internet traffic in North America will increase 264 percent by 2011 to more than 7.8 million terabytes, or the equivalent of 40 trillion e-mail messages. With such increases, many believe networks could crash or at least slow to a crawl.

"The increasing bandwidth demands on cable operators will soon reach crisis stage," wrote Stan Schatt, research director at ABI Research, in a recent report. Others disagree. Cisco, even with its forecast of great traffic growth, concludes Internet-service providers will be able to cope.

Today, information travels the Web by being broken into tiny bits called packets, which are routed through the least congested pipes to their destination. Once the packets arrive, they are reassembled into their original form. The problem is that the increasing size of files, such as video, has begun overwhelming some equipment handling the traffic, resulting in errant or lost packets.

To tackle the problem, a slew of start-ups are producing gear and software to accelerate Internet traffic or to increase the network's capacity. These include companies run by Roberts and Bosack, as well as Riverbed Technology Inc. and Big Band Networks Inc. Other companies, such as BitGravity Inc. and Limelight Networks Inc., are creating "parallel networks" -- essentially scaled-down versions of the Internet -- to escape the glut of traffic on current networks.

The target buyers for these start-up services are Internet-service providers and big companies with large networks of satellite offices. A business, for instance, can affect how quickly e-mails and other files flow to their branch offices depending on the equipment they use in their network.

But many of the new start-ups aiming to cater to these needs aren't likely to survive. In the late 1990s, start-ups such as Celox Networks Inc., Chiaro Networks Inc. and Centerpoint Broadband Technologies Inc. also emerged to grapple with the growth in Internet traffic. All have since filed for bankruptcy or shut down, casualties of the telecom bust.

Roberts' concern over the Internet's infrastructure stretches back years. While at ARPAnet, he says he was unsure how long the technology could work, especially since the system didn't ensure that information packets would arrive at their destination. His fears crystallized in the late 1990s when companies began to use the Internet to make phone calls and consumers begin to dabble in online video.

Today's Web-traffic-control start-ups "really have their work cut out for them," says Michael Kennedy, a networking consultant with Network Strategy Partners LLC. "Technology doesn't always win the race."

"The Internet wasn't designed for people to watch television," Roberts adds. "I know because I designed it."

In 1998, Roberts, who worked at several networking start-ups in the 1970s and 1980s after leaving the Defense Department in 1971, founded a start-up called Caspian Networks Inc. He raised $317 million from venture capitalists for Caspian to manufacture the flow-based routers that could analyze Internet traffic and improve how that traffic moved. But the equipment -- priced at about $500,000 per router -- was too expensive for many customers.

Roberts resigned in 2004 from Caspian; the company closed its doors last year. But the scientist's desire to improve the Web's infrastructure didn't abate. Two months after leaving Caspian, Roberts founded Anagran in Redwood City, Calif., and raised $22 million in venture funding to continue his work. This time around, he says, Anagran's product is cheaper -- $70,000 -- and there is a more urgent need for such equipment.

Anagran has already picked up customers such as Merit Networks Inc., of Ann Arbor, Mich., a nonprofit organization that builds networks for universities, and the International Center for Advanced Internet Research at Northwestern University. "Larry's equipment is built for the more complex traffic like Internet television," says Jim Chen, assistant director of networking at Northwestern. "It's a perfect fit for us."

Roberts and Bosack haven't met but know of each other through industry circles and say they have been following one another's progress.

"We're pushing for the same thing," says Bosack. "The public needs something better than what's currently available."