July 15, 1997

HOW THE INTERNET DOESN'T WORK (PART 1)

Edited by Dennis Berman

What's wrong with the Internet? Among other things, says Bob Metcalfe -- founder of 3Com, inventor of Ethernet, and technology VP for International Data Group -- it's about to collapse. "Gigalapses loom," Metcalfe predicts, referring to massive Internet blackouts. Such was the first of many provocations offered during Metcalfe's June 26 talk to a nationwide satellite audience of Massachussetts Institute of Technology Enterprise Forums -- a network of MIT alumni groups.

Metcalfe's speech is an insider's take on the sometimes haphazard, oftentimes rewarding possibilities of this new medium. Excerpts of the speech, which will appear at Business Week Online in three daily installments, cover a wide range of topics -- from the Internet's collapse, to its greatest enemies (phone companies), to the four breakthrough technologies of the future.

Whatever the subject, Metcalfe's mix of computer history, technological passion, and political advocacy makes his opinions one way -- at least as he puts it -- of getting "the truth about the Internet."

For more, view "How the Internet Doesn't Work (Part 2)" or "How the Internet Doesn't Work...and Beyond (Part 3)."

HOW THE INTERNET DOES NOT WORK

The Internet is bogging down. The number one way you know that the Internet is bogging down is that everybody knows where the "Stop" button is in their Web browser. And why do you think Netscape and Microsoft put that stop button right in the middle of your browsing screen? Because you use it a lot. Why? Are you fickle? No, it's because the Internet is bogging down. You spend too much time waiting for Web downloads and give up too often on what you really want. The number two way you know the Internet is bogging down is how often you get inexplicable error messages, like being accused of mistyping names of sites you go to every day.

THE INTERNET IS COLLAPSING

I got into some serious trouble last year by predicting that the Internet would collapse. Actually, I predicted collapses, plural -- getting bigger and longer -- from various causes. Unfortunately, I got credit for predicting that the Internet would meltdown and go away, once and for all. Even today, people ask me when the Internet is going to collapse, as if it has not already, as if it isn't as we speak.

Anyway, by collapses I mean large outages, measured by multiplying the number of people denied Internet access times the number of hours they were denied it. In late 1995, noting reports of increasing packet losses along Internet backbones, and worrying about the press of increasing routing complexity on a fragile routing infrastructure, I began warning that kilolapses -- Internet outages measured in the thousands of lost user hours -- were looming. Perhaps even megalapses.

The so-called ISP [Internet service provider] community defensively scoffed. They made me promise, in front of 2,000 people at the 4th International World Wide Web Conference here in Boston, to eat my Infoworld column if the Internet did not collapse during 1996. Little did I know, nor would ISPs admit, that kilolapses where already then a daily event and megalapses imminent. In early 1996, one of the larger ISPs, Netcom, crashed entirely, losing Internet access for all of its then 400,000 users for 13 hours. That's a 5.2 megalapse. The cause was a typo, an ampersand wrongly typed by a Netcom engineer into a routing table. The typo, buried in complexity and unchecked, propagated to all of Netcom's routers, bringing them down.

As you might expect, I immediately crowed, "I told you so." The ISP community, however, said the outage was small, limited by the Internet's resilient design and their fast action. Only one ISP was hurt, so that was not a collapse. And it would not likely happen again.

Then, on Aug. 7, while I was defending my collapse theories at an ISP convention in San Francisco and on C-Span, by far the word's largest ISP, America Online, lost its routers, denying Internet access to all of its then 6.2 million users for 19 hours -- a 118 megalapse. Immediately I crowed, and the ISP community countered, America Online is not really part of the Internet.

Now, if only I'd just crowed, "but no." Following the AOL megalapse, I ran some numbers and warned, in August, of the strong possibility of a gigalapse in 1996. A gigalapse would be about 10 times worse than the AOL collapse, a billion lost user-hours in a single Internet outage. Next, skipping over many kilolapses, the story goes that a rat chewed through a power cable at Stanford, bringing down for most of a day one of the major West Coast POPs [points of presence] of another large ISP, the first ISP, BBN, here in Cambridge. The next day BBN fired off a preemptive press release stating that their outage wasn't so bad, as only 400 users were affected. BBN's release failed to mention that among those 400 "users" were Stanford, Berkeley, Hewlett-Packard, Sun ... And Infoworld, where my column appears every week to the delight of over a half million readers.

See, ISPs are just like vendors everywhere, God bless them, and so you've got be... Optimistic, but not naive.

The next big event in my collapse saga again involved AOL which, against my advice, changed its by-then 8 million users over to flat-rate billing. This caused its users to stay online 25% more than when they were paying by the minute. This caused millions of AOL users to get busy signals. This caused AOL, belatedly, to start investing in adding many more modem ports. And this caused our telephone monopolies -- our beloved telopolies -- to start whining in Washington that ISPs should be made to pay extra for all those modem ports with long Internet holding times, not the short voice call holding times that telephone switches are engineered to carry. Soon everyone was in a tizzy about a collapse of the telephone system, and elderly shut-ins not getting dial tone for their 911 calls.

But anyway, while 1996 saw all this, and packet losses routinely reaching 30% during busy hours, and a steady bogging down of the Internet -- everybody found out where the stop button was on their browsers -- no gigalapse occurred, at least as far as I could tell. And so, on April 10, 1997, at the Sixth World Wide Web Conference, in Santa Clara, after hearing my animated arguments and pathetic pleadings, an audience of 1,000 Internet enthusiasts demanded that I eat, literally eat, yuk! my December, 1995, Internet collapse column. After the press coverage of that sad event, I am now better known for eating my Internet collapse column than for inventing Ethernet.

Of course, eating my column did not stop the Internet from continuing to bog down and intermittently collapse. Just 15 days later, on April 25, 1997, there was a really scary incident in which a tiny ISP inadvertently created some bogus routing information, which you'll recall is easy. The scary part was that the bogus info propagated to a larger ISP upstream, and to an even, larger ISP further upstream, and eventually into the Internet's major inter-ISP exchange points (NAPS). Reports vary, but between 0% and 40% of the Internet's traffic was lost for 0 to 7 hours during the North American business day. The worst part of the Apr. 25 incident was, however, that hardly anybody complained. Sadly, Internet users are used to losing their access for hours at a time and simply coming back later, not knowing that millions of them might all be disconnected at the same time. Rest assured that the Internet has not yet been fixed, and gigalapses loom, although I'm not promising to eat anything if they don't.

(Click here for part two)