America braced itself last week for the release of Kenneth Starr's report on the relationship between President Clinton and Monica Lewinsky. The anticipation was palpable, and much of it had to do with the content of the report: What was Starr claiming the president had done? But some of the anticipation had nothing to do with the content of the report, but rather with the way it was being released over the Internet. The House Judiciary Committee threw the 445-page report up on the World Wide Web in glorious HTML. Though it was just over a day between the Committee's acceptance of the Starr report and its posting on the Web, a day is forever in Web time — long enough for millions of Internet users to consider the technical difficulties of putting on the Web a document of that size, wanted by so many people. Would the Internet, much less President Clinton, be able to survive the release of the Starr report?

Obviously, the 'Net survived and many people are calling the relative ease with which the Internet apparently handled the massive distribution a watershed event. This was the moment, some said, when the Internet came of age. Nonsense. Distribution of the Starr report only demonstrated the deep architectural problems faced by the 'Net.

But wait, you say, the report was read and downloaded by millions of people within hours of its release. How could this be anything but a success?

The Starr report was read and downloaded by millions of people because of an ad hoc network of mirror sites. The House Web site, itself, which its administrators said was ready to handle most any load, was completely overwhelmed. Congress was no more ready to handle the crush of interest in the Starr report than you are capable of keeping your dog from running into the road. And here's the pathetic part: the House Web site was so smug because they'd survived a surge of hits following the death of representative Sonny Bono. These people who work everyday in the political world thought public interest in the possible end of a presidency might be comparable to the interest in a former entertainer turned Congressman skiing into a tree. Ignoring the fact that there was no 445-page HTML document associated with Sonny's death, what could these people have been thinking?

Site statistics suggest the Monica-to-Sonny hit ratio was at least 20:1, and probably would have been even higher had the House Internet connection not maxed-out.

But everyone who wanted the report got it, right? Every Internet Portal worth its port, every news site worth its news, and nearly every ISP in the mix had the report ready for downloading within minutes of its release. The system works, right? Wrong.

The system didn't work, it was made to work. A giant kludge powered by prurient self-interest got those many mirror sites in operation. The Internet really should have been able to handle the Starr report without any special effort. Had the infrastructure been designed correctly there should have been no need for all the heroics.

Here are two facts: 1) the widespread concern that the Starr report would bring the 'Net to its knees (no pun intended) shows by itself that the Internet is not perceived as a reliable medium, and ; 2) history has already shown us that whatever is exceptional on the 'Net today is the norm six months from today. So what are we doing to make the Internet run even more smoothly six months from now? We are doing almost nothing.

The Internet was never designed with tens of millions of users in mind. More specifically, it wasn't designed with the idea that a significant subset of those millions would want to receive the same 445-page document within a 24 hour period. The Internet wasn't designed with the idea of it becoming a mass medium, either, but that's exactly what it has become. It's major strength — point-to-point communication — becomes a major weakness when we begin to talk about sending the same thing to millions of people at the same time.

There are Internet technologies intended to handle these problems, but these technologies are imperfect. We see this already in what has come to be called the "World Wide Wait." Why does Web browsing have to be as slow as it is? Because the system has no foreknowledge of what information people will actually want. On network TV, at least we know enough to expect that people will want Seinfeld on Thursday night. Compared to network TV, the Internet is anarchy — glorious anarchy sometimes, but horrible anarchy in those times when what you and nearly every other person would really like is just Seinfeld.

The fact is that we are a tribal culture with large groups of people who like to watch, read, and consider exactly the same thing, yet this Internet invention doesn't handle those needs very well.

We have cache servers, but they generally don't know what we want before we want it. And they aren't prepared to serve up more of one thing than another. This is changing, slowly, with outfits that are loading cache servers based on statistical samplings of many ISPs. The idea here is that the cache at MIT ought to be a good predictor for the cache at Caltech with the advantage that it's also three hours ahead. This is progress, but we have to go further.

We have a major plumbing problem with 29 commercial Internet backbones, 11 major interconnects and hundreds of peering points. For some things like the dispersal of realtime streaming data, the system just doesn't work. For proof, look at the poor performance of streaming video. Okay, so 56 kilobits- per-second isn't much, but try using two modems and one phone line for an IP video conference compared to doing the same video conference over the Internet. The one phone line solution is vastly superior even though the modems and protocols are all the same. It's the Internet architecture that's at fault AND IT ISN'T GOING TO CHANGE UNLESS WE CHANGE IT. Bigger pipes alone is not the answer because traffic growth will always fill those bigger pipes.

There are movements, of course, toward guaranteeing certain quality of service over the 'Net. For a fee, perhaps, you'll get a lane all to yourself while the rest of us will be crowded a bit closer together. This is no more the sole answer to Internet congestion than is bigger pipes.

It's this point-to-point communication that's killing us, running a million Web sessions in parallel to ask for the same damned report. The Internet has ways of handling this, but they don't work. IP multicast is the buzzword most often heard, but it has scalability problems, isn't supported at all by some routers, isn't supported well by some operating systems (did I remember to say Windows?), and for the most part just plain doesn't work. Multicasting isn't the answer either.

What is the answer is a combination of all these techniques with the addition of a new resolution and an ancient solution. The resolution is to use point- to-point communication for what it is really good for (communication between individual people or devices) and use other techniques for everything else. The ancient solution is the broadcast radio network, circa 1923.

In the early days of radio, pre-1923, the way to reach more people was to build a higher antenna and increase the power of your transmitter. That's the way we tend to do it today on the Internet, where we buy a bigger pipe and add servers to reach people all over the world. But in the early days of radio there was a clear limitation to this technique, a limitation based on the physics of groundwave signaling. Beyond a certain point, it didn't matter how much power you added: reaching Pittsburgh sometimes couldn't be done. The same can be said for the Internet, where a combination of slow NAPs and MAEs, too many routers, too many backbones, and just too darned many users can make Pittsburgh seem impossible, too.

Then AT&T, which owned some radio stations, decided to link them by telephone, for the first time allowing the same content to be carried on several stations at the exact same time. AT&T's 1923 experiment brought scalability to radio, matching high-quality content to listener demographics, taking the signal to where audiences were and not to where they weren't. We don't do this today on the Internet. It's not even a factor in the design of the system.

What General Sarnoff knew when he started NBC in 1926 is repugnant to the Internet architects of today. Caught up in the purity of point-to-point they've failed to create a network suitable for true mass communication just at the time the world is declaring the Internet to be just that. They don't even like the idea. But the Internet now belongs to the world, so adding these capabilities is inevitable and the sooner the better. And it's not hard. Even I know how to do it and I'm an idiot.

Why does Yahoo have only a single mirror site? Why does Excite live only in Redwood City and Virginia? If we have 100 major media markets, why not 100 Yahoos and Excites, placing their content a hop or two away from you and me? These two companies alone, through their profligate wasting of bandwidth, are keeping the Internet from reaching its potential. But it's not just them. Every other portal, news agency, and ISP is the same. We have so embraced virtuality that we've forgotten that most major Web servers and news organizations physically handle the whole world from a single site, which pisses me off. That's my bandwidth they are wasting, and yours.

NBC, CBS, and ABC learned this lesson 60 years ago. It's time the Internet did too.