Circumventing traffic jams on the Internet.

During the 6 months that Carolina Cruz-Neira of Iowa State University in Ames and her collaborators at various institutions in Illinois worked to develop an elaborate, interactive computer simulation of molecules, they actually met in person only twice. Except for a few telephone calls, everything was done over the Internet-from sending messages to transferring programs and data.

"We were able to exchange information quickly, reliably, and easily many times per day," says Cruz-Neira.

In creating the Virtual Biomolecular Environment, Cruz-Neira and her coworkers envisioned a system that any scientist with a desktop computer connected via the Internet could use to study biochemical interactions. The Internet, however, doesn't yet have the capacity or reliability to handle such a challenge.

"Our application requires real-time manipulation of a remote simulation as the simulation is running," Cruz-Neira says. Even a 0.1-second delay in transmitting commands could be so frustrating that a researcher might not want to use the system.

Similar problems have hindered other research projects, ranging from direct control of a remote electron microscope to on-line visualization of black holes. "If you look at the Internet as a whole, you can't reliably deliver the kind of performance you need at busy times, even for existing applications," says Mark Luker, who manages the network connections program at the National Science Foundation in Arlington, Va.

To help remedy this situation, NSF last week introduced a grant program to fund specific scientific and engineering projects that require innovative ways of regulating traffic flow on the Internet.

At present, the Internet encompasses more than 100,000 networks connecting millions of computers throughout the world. It carries all the information in small packets that are treated equally, whether they consist of simple messages or portions of complex video images, whether they are part of vital research or a student's home page photo.

As traffic increases, the Internet responds by slowing down.

"We need to take a deeper look at this, not just try to expand capacity and stay ahead of demand," Luker says. "Some traffic needs better or different service than other traffic."

This means developing schemes for setting priorities and implementing them. It may be possible, for example, to tag certain data streams so they have right-of-way on the Internet just as emergency vehicles do on highways. Other solutions may involve setting aside specific channels for priority traffic during peak periods. As a temporary measure, research groups can apply for links to NSF's experimental, high-performance vBNS system, which connects its five supercomputer centers.

"There's a vision here of a more effective Internet-one that has different qualities of service for different needs and can guarantee the kind of service you need to do what you want to do," Luker says.