"At the beginning of 1997, for example, we completed a 622-megabit backbone system. We just doubled that, so we are running on all the links at twice that bandwidth."

The real issue for me is whether or not the vendors of the switching equipment will be able to keep up with the speeds we know we can drive the optical circuits.

This is a self-healing problem. The Internet service providers who are unable to make the investment to increase capacity will eventually exhibit poor quality service, and customers will go elsewhere. So long as some people are making the investment, they will garner more and more of the traffic. The system will continue to grow, because revenues will be falling to the places that offer the best service.

SCHINDLER: A number of solutions have been proposed. You mentioned a 40-gigabyte backbone.
CERF: Gigabits, actually, but what's a factor of eight among friends. ... And besides, in the lab they have demonstrated terabit fibers, that's trillions of bits per second. That's not in production anywhere, but it is well within the range of feasibility for a single-mode optical fiber to do this. I'm anticipating that we will have plenty of fiber capacity as time goes on.

SCHINDLER: Plenty of bits on the backbone, but a small pipe to the household. Which solution do you favor for that last mile?
CERF: It depends on where you are in the country. The cable modems which have been demonstrated work quite well, but they work best when the cable system has a hybrid fiber/coax network to support it. Not all cable companies are prepared to make an investment in hybrid fiber/coax, and as a result, they may not be able to support that solution.

The telcos, working with our old friend twisted-pair copper have demonstrated up to 52 megabits on relatively short hops, down to, let's say, a megabit or 1.5 megabits as far as 18,000 feet, which covers 95 percent of all homes on twisted pair.

I have left out things like point-to-point radio links, which also might be a solution to the problem.

There is even something quite exotic. Power companies may decide they want to have the ability to control appliances on an appliance-by-appliance basis, because they will save money, not having to build excess power generation, if they can control the appliances that are the most energy consuming during periods of peak load. If you could save several billion dollars by not having to build a power-generation plant, you might spend some of that money on optical fiber to go to someone's house, then turn around and resell the capacity, which you didn't need to control an appliance -- and you only need a few bits per second for that.

We could have some surprising solutions become available that might not have been anticipated.

"Wouldn't it be nice to have a VCR on the Net, pull up a Web page, click on the programs you want, and have the net deliver the instructions to do the recording to the VCR."

SCHINDLER: Toasters on the Internet were not just jokes?
CERF: Indeed. I have already spoken to a gentleman who is in the process of building TCP/IP on a chip. One expects those devices to be so cheap, in quantity, that you'd put them in all kinds of appliances. Water heaters, VCRs, for example. Wouldn't it be nice to have a VCR on the Net, pull up a Web page, click on the programs you want, and have the net deliver the instructions to do the recording to the VCR. For that matter, deliver the time of day as well, so you don't have to do anything about that flashing "12:00" that drives everyone crazy after a power outage. There are lots of reasons appliances might be put up on the net, not just for power control.

SCHINDLER: Do you envision a time when there will be a priority packet service available? When you can pay for better response?
CERF: There are a number of efforts afoot to create premium or priority services that distinguish the run-of-the-mill best efforts traffic -- the kind of Internet service you get today -- from services where some packets do get priority over others. That can be very important for an interactive application such as game playing, or even voice, where even modest delays can be quite disturbing to the participants. That is all quite doable. We are seeing changes in the router software already to offer that kind of differentiated service.

The more responsive services might assure bandwidth, which is another area where you might need premium service, in order to make large file transfers in a short amount of time. All of those services consume more resources than best efforts, so it is quite reasonable to assume that they might cost more. You wouldn't need them all the time; you'd only need them for certain applications for a period of time.