When you talk on the phone, your words are converted into bits that are sent down a wire. When you surf the Web, your computer sends out bits that ask for yet more bits to be sent back. When you go to the store and buy a Japanese VCR or an article of clothing with a Made in Thailand label, you're touching off a cascade of information flows that eventually leads to transpacific faxes, phone calls, and money transfers.

If you get a fast busy signal when you dial your phone, or if your Web browser stalls, or if the electronics store is always low on inventory because the distribution system is balled up somewhere, then it means that someone, somewhere, is suffering pain. Eventually this pain gets taken out on a fairly small number of meek, mild-mannered statisticians - telecom traffic forecasters - who are supposed to see these problems coming.

Like many other telephony-related technologies, traffic forecasting was developed to a fine art a long time ago and rarely screwed up. Usually the telcos knew when the capacity of their systems was going to be stretched past acceptable limits. Then they went shopping for bandwidth. Cables got built.

That is all past history. "The telecoms aren't forecasting now," Mercogliano says. "They're reacting."

This is a big problem for a few different reasons. One is that cables take a few years to build, and, once built, last for a quarter of a century. It's not a nimble industry in that way. A PTT thinking about investing in a club cable is making a 25-year commitment to a piece of equipment that will almost certainly be obsolete long before it reaches the end of its working life. Not only are they risking lots of money, but they are putting it into an exceptionally long-term investment. Long-term investments are great if you have reliable long-term forecasts, but when your entire forecasting system gets blown out of the water by something like the Internet, the situation gets awfully complicated.

The Internet poses another problem for telcos by being asymmetrical. Imagine you are running an international telecom company in Japan. Everything you've ever done, since TPC-1 came into Ninomiya in '64, has been predicated on circuits. Circuits are the basic unit you buy and sell - they are to you what cars are to a Cadillac dealership. A circuit, by definition, is symmetrical. It consists of an equal amount of bandwidth in each direction - since most phone conversations, on average, entail both parties talking about the same amount. A circuit between Japan and the United States is something that enables data to be sent from Japan to the US, and from the US to Japan, at the same rate - the same bandwidth. In order to get your hands on a circuit, you cut a deal with a company in the States. This deal is called a correspondent agreement.

One day, you see an ad in a magazine for a newfangled thing called a modem. You hook one end up to a computer and the other end to a phone line, and it enables the computer to grab a circuit and exchange data with some other computer with a modem. So far, so good. As a cable-savvy type, you know that people have been hacking cables in this fashion since Kelvin. As long as the thing works on the basis of circuits, you don't care - any more than a car salesman would care if someone bought Cadillacs, tore out the seats, and used them to haul gravel.

A few years later, you hear about some modem-related nonsense called the World Wide Web. And a year after that, everyone seems to be talking about it. About the same time, all of your traffic forecasts go down the toilet. Nothing's working the way it used to. Everything is screwed up.

Why? Because the Web is asymmetrical. All of your Japanese Web customers are using it to access sites in the States, because that's where all the sites are located. When one of them clicks on a button on an American Web page, a request is sent over the cable to the US. The request is infinitesimal, just a few bytes. The site in the States promptly responds by trying to send back a high-resolution, 24-bit color image of Cindy Crawford, or an MPEG film of a space shuttle mission. Millions of bytes. Your pipe gets jammed solid with incoming packets.

You're a businessperson. You want to make your customers happy. You want them to get their millions of bytes from the States in some reasonable amount of time. The only way to make this happen is to purchase more circuits on the cables linking Japan to the States. But if you do this, only half of each circuit is going to be used - the incoming half. The outgoing half will carry a miserable trickle of packets. Its bandwidth will be wasted. The correspondent agreement relationship, which has been the basis of the international telecom business ever since the first cables were laid, doesn't work anymore.

This, in combination with the havoc increasingly being wrought by callback services, is weird, bad, hairy news for the telecom monopolies. Mercogliano believes that the solution lies in some sort of bandwidth arbitrage scheme, but talking about that to an old-time telecrat is like describing derivative investments to an old codger who keeps his money under his mattress. "The club system is breaking down," Mercogliano says.

Somewhere between50° 54.20062' N, 1° 26.87229 W and50° 54.20675' N, 1° 26.95470 WCable Ship Monarch, Southampton, England

John Mercogliano, if this is conceivable, logs even more frequent-flier miles, to even more parts of the planet, than the cable layers we met on Lan Tao Island. He lives in London, his office is in Amsterdam, his territory is Europe, he works for a company headquartered in Bermuda that has many ties to the New York metropolitan area and that does business everywhere from Porthcurno to Miura. He is trim, young-looking, and vigorous, but even so the schedule occasionally takes its toll on him, and he feels the need to just get away from his job for a few days and think about something - anything - other than submarine cables. The last time this feeling came over him, he made inquiries with a tourist bureau in Ireland that referred him to a quiet, out-of-the-way place on the coast: a stately home that had been converted to a seaside inn, an ideal place for him to go to get his mind off his work. Mercogliano flew to Ireland and made his way overland to the place, checked into his room, and began ambling through the building. The first thing he saw was a display case containing samples of various types of 19th-century submarine cables. It turned out that the former owner of this mansion had been the captain of the Great Eastern, the first of the great deep-sea cable-laying ships.

The Great Eastern got that job because it was by a long chalk the largest ship on the planet at the time - so large that its utter uselessness had made it a laughingstock, the Spruce Goose of its day. The second generation of long-range submarine cables, designed to Lord Kelvin's specifications after the debacle of 1857, were thick and heavy. Splicing segments together in mid-ocean had turned out to be problematical, so there were good reasons for wanting to make the cable in one huge piece and simply laying the whole thing in one go.

It is easier to splice cables now and getting easier all the time. Coaxial cables of the last few decades took some 36 to 48 hours to splice, partly because it was necessary to mold a jacket around them. Modern cables can be spliced in more like 12 hours, depending on the number of fibers they contain. So modern cable ships needn't be quite as great as the Great Eastern.

Other than the tank that contains the cable, which is literally nothing more than a big round hole in the middle of the ship, a cable ship is different from other ships in two ways. One, it comes with a complement of bow and stern thrusters coupled to exquisitely sensitive navigation gear on the bridge, which give it unsurpassed precision-maneuvering and station-keeping powers. In the case of Monarch, a smaller cable repair ship that we visited in Southampton, England, there are at least two differential GPS receivers, one for the bow and one for the stern - hence the two readings given at the head of this section. Each one of them reads out to five decimal places, which implies a resolution of about 1 centimeter.