Tubes: A Journey to the Center of the Internet

by Andrew Blum

  But where to go? At that point, the options were limited. Few organizations had websites—only universities, a few computer companies, the National Weather Service. And how did one know where to find them? There was no Google, Yahoo!, MSN, or even Ask Jeeves. Unlike the walled enclosure of AOL, unlike any other computer I’d ever used, it felt as open-ended as the world. The sensation was distinct: it felt like travel. I wasn’t alone in the thrill. That was a heady season for the Internet. Netscape released its web browser in October, as Microsoft ramped up the advertising campaign for its own “Internet Explorer.” The Internet was about to go mainstream, once and for all. The roof was about to blow off.

  But which roof, really? The boom would strain the Internet’s existing infrastructure to the breaking point. So who would save it? How did it expand? To where? I’d heard the business stories of the dot-com boom, about how Jim Clark and Marc Andreessen founded Netscape, and Bill Gates battled to keep Internet Explorer an integral part of Microsoft’s Windows operating system. But what about the networks themselves, and their places of connections? In a business that’s always been obsessed with the next new thing, who was still around to tell that story?

  I went back to California—only to hear about Virginia.

  On a characteristically damp and gray San Francisco day, I met a network engineer named Steve Feldman at a café a few blocks from his office, in the heart of the cluster of Internet companies located south of Market Street. He looked like a high school math teacher, with khaki pants, sturdy brown walking shoes, and a big beard. His office ID hung around his neck from a lanyard embroidered with NANOG, the North American Network Operators’ Group—the clubby association of engineers who manage the biggest Internet networks, and whose steering committee Feldman chairs. These days his job is to run the data network for CBS Interactive, making sure, among other things, that the latest episode of Survivor or the NCAA basketball game is properly streamed to your screen. (Even if he wasn’t a fan himself.) But for a time in the ’90s, Feldman ran the Internet’s single most important place, a global crossroads improbably located in the parking garage of an office building in a suburb of Washington, DC. It was a thrilling moment in the Internet’s evolution—for a while. By the end, things had gotten out of control.

  We sat down between two young guys pecking away at laptops, their heads in the cloud. Our conversation must have sounded strange to them; it was all such ancient history. In 1993, Feldman—a graduate of the computer science department at Berkeley—went to work for a young networking company called MFS Datanet, which had started out laying fiber in Chicago’s coal tunnels, and had more recently been building private networks linking corporate offices together, mostly piggybacking on existing phone lines. MFS wasn’t providing Internet access itself, only helping companies with their internal networks, but it had gotten good at doing so across a city—which was exactly what the handful of companies that were providing Internet access needed. They had a problem. At the time, the de facto backbone of the Internet was run by the National Science Foundation and known as NSFNET, but technically the commercial companies were prohibited from using it by the “acceptable use policy,” which in theory limited traffic to academic or educational purposes. To grow, the commercial providers had to find a way to exchange traffic across their own private roads, in order to avoid traversing that government-run highway. That meant connecting to one another—physically. But where?

  Business was booming for everybody, but the whole endeavor was threatened by an absence of real estate. Where could they connect? Quite literally: Where was a cheap place with plenty of electricity where the engineers could string a cable from the router of one network to the router of another?

  The Virginia suburbs west of Washington, DC, were already a hot spot for many of the early commercial Internet providers, mainly due to the concentration of military contractors and high-tech companies in the area. “It was the technology center,” Feldman told me. For a time, a few of the early Internet providers interconnected their networks inside a Sprint building in northwest Washington, but it was an imperfect solution. Sprint didn’t like its competitors setting up shop inside its building (especially when Sprint didn’t have a business setup to properly charge them for it). And for the Internet providers themselves—companies like UUNET, PSI, or Netcom—it was expensive to be there, because of the cost of leasing local data lines back to their own office or network POP (or “point of presence”).

  MFS offered a solution: it would turn its offices into a hub. The company already had plenty of existing local data lines, which it would use to tether each of the Internet providers, like dancers around a maypole. MFS would then provide a switch, called a Catalyst 1200, that would route traffic between the networks. It wasn’t merely a local road; it was a roundabout. By plugging into this hub, each network would have immediate and direct access to all the other participating networks, no highway tolls required. But for the plan to be viable a handful of the Internet providers had to commit simultaneously—or else it would be a roundabout in the middle of nowhere. A group of them made the decision over lunch one day in 1992 at the Tortilla Factory in Herndon, Virginia. At the table were Bob Collet, who ran Sprint’s network; Marty Schoffstall, cofounder of PSI; and Rick Adams, founder of UUNET (who would later make hundreds of millions of dollars taking it public). Each of these networks operated independently, but they knew full well they were useless without one another. The Internet was still for hobbyists—an eccentric subset of the population, composed mostly of people who had used the network in college and wanted to keep going. (In the United States, the percentage of households with access to the Internet wasn’t measured at all until 1997.) But the growth trend was clear: for the good of the Internet—really if there were to be a functioning nonacademic Internet at all—the networks had to act as one. MFS called its new hub a “Metropolitan Area Exchange.” To indicate its ambition to build several of them around the country, it nicknamed the hub MAE-East.

  It took off immediately. “MAE-East was so popular that we were outgrowing the technology faster than we could upgrade it,” Feldman said. When a new Internet service provider sprang into existence, its customers would mainly call in over a regular telephone line using a modem. But then the provider had to connect to the rest of the Internet (as Jon Auer did in Milwaukee). And for a time, MAE-East was it. “If you connected to MAE-East, you’d have the entire Internet at your doorstep,” Feldman said. “It was the de facto way into the Internet business.” Within a couple years, MAE-East was the crossroads for fully half of all the world’s Internet’s traffic. A message from London to Paris most likely went through MAE-East. A physicist in Tokyo querying a website in Stockholm went through MAE-East—on the fifth floor of 8100 Boone Boulevard in Tysons Corner, Virginia.

  It was a portentous location. The intersection of the Leesburg Pike and Chain Bridge Road may have been the crossroads of the digital world, but it was also conspicuously close to the crossroads of American espionage—overhanging MAE-East with an ongoing fog of mystery, suspicion, even conspiracy.

  Tysons Corner is one of the highest points in Fairfax County, at five hundred feet above sea level. During the Civil War, the Union army took advantage of its views back toward Washington and out toward the Blue Ridge Mountains, and erected a signal tower there, pillaging timber for its construction from nearby farms. A century later, at the start of the Cold War, the US Army built a radio tower on the same spot and for the same reason: to relay communications between headquarters in the capital and distant military posts. A military tower still stands on the site, a red-and-white steel skeleton looming above a busy suburban crossroads, ringed by a protective fence with a sternly worded sign prohibiting photographs. Heightening the place’s mysteriousness, shortwave radio hobbyists have fingered the tower as a source of the “numbers stations”—radio broadcasts of an endless cadence of spoken digits. If the professional spook commentators are to be believed, far-flung spies tune in at sp
ecified times to receive coded communiqués from headquarters. According to Mark Stout, a historian at the International Spy Museum, the single-use codebooks the system uses are uncrackable. “You really truly cryptanalytically have no traction getting into a one-time pad system,” he says. “None at all.”

  Indeed, if counterespionage were your gig, the rest of Tysons Corner might pose similar challenges. MAE-East isn’t there anymore—or rather, whatever networking equipment that still is there is no longer a significant center of the Internet—but the neighborhood remains the same. Circling the parking lots, the buildings themselves seem sealed, with perfectly flat glass façades, as if conceived by their architects to be as anonymous as they are impenetrable. The buildings are mostly unmarked, in accordance with the wishes of their low-profile tenants. When they do have signs, they reveal the identities of military contractors: Lockheed Martin, Northrop Grumman, BAE. Many were built with special rooms, known as Sensitive Compartmented Information Facilities, or “skiffs,” designed to meet government criteria for handling classified information.

  The most paranoid network engineers—the “tin-foil hat guys,” as they’re known, in reference to the conviction that the only way to keep the government from reading your mind is to wear a helmet made of aluminum foil—took MAE-East’s location as proof of its malevolent government control. Why else would it be down the road from the CIA? And if it wasn’t the CIA listening in, then it must have been the supersecret National Security Agency systematically tapping everything passing through—a claim repeated in James Bamford’s bestselling 2008 book about the NSA, The Shadow Factory. Even today, do any idle googling about MAE-East and the information seems oddly sketchy—written in the present tense, even though its importance is long past; marked in red on satellite photographs with its relation to a nearby CIA facility highlighted; somehow frozen in time. MAE-East remains an international woman—or an international something—of mystery.

  Alas, it’s all a little overblown. MAE-East’s importance may have begun spontaneously, but it continued bureaucratically. In 1991, the US Congress had passed the High Performance Computing and Communication Act, better known as the “Gore Bill,” named after its original sponsor, then-senator Al Gore. It’s to this that Gore owed his purported claim of having “invented the Internet”—which isn’t as far-fetched as it sounds. Invent is undoubtedly the wrong word, but the push from government was crucial in getting the Internet out of its academic ghetto. Among the bill’s provisions was a piece of policy best known by its popular name: the “information superhighway.” But rather than putting shovels in the ground to build it, government policymakers catalyzed private companies to do it for them, by funding the construction of “on-ramps.” A network access point, or NAP, as they called it, would be “a high-speed network or switch to which a number of networks can be connected via routers for the purpose of traffic exchange and interoperation.” It would be funded with federal dollars, but operated by a private company. An access point, in other words, would be a network that connects networks: a copycat MAE-East.

  Feldman responded to the government request for bids with an idea for a fancy new exchange—but the National Science Foundation, which ran the process, said they’d rather just give MFS money to keep MAE-East going. Contracts were eventually awarded for four access points, run by four major telecom players: the Sprint NAP in Pennsauken, New Jersey, just across the Delaware River from Philadelphia; the Ameritech NAP in Chicago; the Pacific Bell NAP in San Francisco; and MAE-East. But Feldman likes to say there were really only three and half, “because we already existed.” (And MFS would soon open MAE-West, at 55 South Market Street, in San Jose, California, to compete with the Pacific Bell NAP.) That geography was deliberate. The National Science Foundation knew that to succeed the network hubs needed to serve distinct regional markets, spread evenly across the country. Distance mattered. The original solicitation accordingly identified “California,” Chicago, and New York City as “priority locations.” The decision to locate the Sprint NAP in a bunker of a building in Pennsauken, ninety miles from New York, was because of the existing facility’s links to the transatlantic undersea cables that landed on the New Jersey shore; it was the gateway to Europe.

  The opening of the network access points also marked an important philosophical shift, one that would have ramifications for its physical structure. In a clear departure from its original roots, the Internet was no longer structured as a mesh, but rather was entirely dependent on a handful of centers. As the urban theorist Anthony Townsend has pointed out, “The reengineering of the Internet’s topology that was implemented in 1995 was the culmination of a long-term trend away from the idealized distributed network … envisioned in the 1960s.” As the number of networks increased, their autonomy was best served by centralized meeting points.

  But for Feldman the meeting point felt more like a choke point. By 1996, MAE-East was overstuffed with belching, blinking machines, and growing somewhat out of control, however profitably. The original concept had been that each network would house its own router and link into MAE-East over its data lines. A machine evocatively called a FiberMux Magnum would act like the can on one end of a string telephone, changing the signals coming over the line into a form MAE-East’s router could understand. But as you might guess, FiberMux Magnums themselves take up space, and the fifth-floor suite at 8100 Boone that housed MAE-East—or you might say that was MAE-East—quickly filled up. The situation deteriorated further when networks discovered that they could increase their performance if they chucked the FiberMuxes and installed their actual routers at MAE-East too, in effect making it their new technical office. And it got even more crowded when they discovered performance increased again if they put their servers there as well, so that MAE-East wasn’t just the transit point for data, but often its source. Web pages loaded faster for its customers, and it reduced the costs of moving the bits around. But with those changes, MAE-East had transformed from a crossroads to a depot.

  It fell to Feldman to find a way to expand. The landlord at 8100 Boone had become impatient with its power-sucking tenant, so soon the many-tentacled apparatus moved into a plasterboard enclosure carved out of the basement parking garage of the building across the street, at 1919 Gallows Road. Air-conditioning units surrounded the bare white walls, hard up against the underground parking spots. A generic hardware store ACCESS RESTRICTED sign marked the door. The undisputed capital of the Internet was decidedly humble, the kind of space where you’d expect to find floor-polishing machines and toilet paper stocks, not the spinal cord of a global information network. MAE-East’s location in a parking garage may have seemed like something out of a spy movie—the one where an anonymous door in a dingy corridor opens up to reveal a huge, glistening, high-tech lair. But the high-tech lair was a hovel.

  It drove Feldman to distraction. When he wasn’t selecting and installing new equipment, managing the connections between networks, and trying to figure out what everybody needed, he was apologizing. Traffic had been doubling every year, far outpacing what the router technology could handle, not to mention the real estate. The Internet was clogged. At every meeting of the North American Network Operators’ Group, Feldman would be asked to stand up in front of his colleagues and explain why the crossroads of the Internet, his crossroads, was perpetually jammed. It wasn’t an easy crowd. “People in the NANOG community say what they’re thinking,” Feldman said. “And they don’t pull punches.” At one meeting, ultimately exasperated by the complaints, Feldman taped a paper bull’s-eye to his chest before taking the stage. There was no getting around it: the model was broken. The Internet needed a new kind of place.

  By 1997, 20 percent of American adults were using the Internet—up from nearly zero a few years before. The Internet had proven its usefulness. But it was unfinished, unrealized. Some of the needed pieces were obvious: there had to be new high-capacity long-distance lines between cities; software tools that would enable “e-commerce” and
online videos; and new devices that could connect to the Internet faster and more flexibly. But beneath all that was an unmet mechanical need, an unbuilt room in the Internet’s basement: Where could all the networks connect? They came up with the answer down the road, in the heart of Silicon Valley—in a basement, in fact.

  3

  * * *

  Only Connect

  For a couple of years at the beginning of the millennium—during the quiet time after the Internet bubble burst but before it inflated again—I lived in Menlo Park, California, a supremely tidy suburb in the heart of Silicon Valley. Menlo Park is a place rich in a lot of things, Internet history among them. When Leonard Kleinrock recorded his first “host-to-host” communication—what he likes to call “the first breath of the Internet’s life”—the computer on the other end of the line was at the Stanford Research Institute, barely a mile from our apartment. A few blocks past there is the garage where Larry Page and Sergey Brin first housed Google, before they moved into real offices above a Persian rug store in nearby Palo Alto. On the morning of Google’s public offering, in August 2004, the crowd at the café on our corner was electrified—not, presumably, because they themselves were getting richer by the moment (although maybe), but because it suddenly made everything seem possible again. Indeed, it was that same summer when Mark Zuckerberg moved his fledgling company, then known as The Facebook, from his dorm room at Harvard to a sublet house in Palo Alto. It wasn’t big news at the time—the only person I knew on Facebook then was my sister-in-law, still in college—but it was clear that it made perfect sense. As E. B. White said of New York, this was the place you came if you were willing to be lucky. Just as Wall Street, Broadway, or Sunset Boulevard each contain a dream, so too does this corner of Silicon Valley. Most often, that dream is to build a new piece of the Internet, preferably one worth a billion dollars. (Facebook, by the way, recently moved into a fifty-seven-acre campus, back in Menlo Park.)