On October 29, 1969, a UCLA graduate student read letters aloud over a phone line to a programmer 350 miles north, trying to log into a computer he had never reached before; the machine crashed on the third one.

Charley Kline had a phone headset on, a colleague 350 miles away confirming each keystroke, and a five-letter word to type into a machine he'd never reached before. What he did not have, about forty seconds in, was a computer that was still listening.

What was this network even for?

By 1969 the Advanced Research Projects Agency, a Pentagon office better known as ARPA, had a problem that had nothing to do with missiles or moon shots. Its funded researchers were scattered across the country, and each university's computer was its own island: expensive, incompatible with its neighbors, and idle half the time while some other lab's machine ran out of capacity. Bob Taylor, who ran ARPA's Information Processing Techniques Office, and Larry Roberts, the program manager he recruited to build the fix, wanted one thing: let a researcher in one city run a program on a computer in another.

The answer was a network of small, dedicated switching computers called Interface Message Processors, built under contract by Bolt Beranek and Newman in Cambridge, Massachusetts, with Frank Heart leading the engineering and Bob Kahn shaping the design. Each IMP would sit next to a host computer and handle the actual routing, so the fussy, incompatible mainframes never had to speak to each other directly. UCLA, chosen partly because Leonard Kleinrock's lab there specialized in analyzing network traffic, got the first one. Stanford Research Institute, in Menlo Park, got the second.

What happened on the night of October 29?

UCLA's IMP had arrived over Labor Day weekend; SRI's showed up a month later, on October 1. By late October the two machines were finally ready for a real test: could a person sitting at UCLA's SDS Sigma 7, in Los Angeles, actually operate a program running on SRI's SDS 940, up in Menlo Park?

Kline, a UCLA graduate student working under Kleinrock, took the keyboard. On the other end, in Menlo Park, sat Bill Duvall, a programmer supporting Douglas Engelbart's Augmentation Research Center (the same lab that, less than a year earlier, had unveiled the computer mouse and on-screen windows to a stunned audience in San Francisco). Kline and Duvall kept a phone line open between them so Duvall could confirm, letter by letter, that each keystroke had actually crossed the new link.

Kline typed L. Duvall watched it arrive and said so. Kline typed O. Duvall confirmed that too. He typed G, and the SRI machine, mid-word, stopped responding entirely.

"And he had a bug, and it crashed," Kline later said of the moment, with the flatness of a man describing a Tuesday rather than a birth. The full word he'd been sending was LOGIN, and only its first two letters made it through before the system gave out. Duvall tracked down the problem and reset the machine; about an hour later, the two of them tried again, and this time the whole command went through cleanly. The entry Kline logged in his notebook that night read simply: "Talked to SRI Host to Host."

Why did it crash on the third letter?

The cause wasn't cleverness gone wrong; it was speed. SRI's host expected input arriving the way a person types, a few characters a second, with room built in for the pauses between them. ARPANET didn't pause. It moved data at whatever rate the wire allowed, thousands of characters a second, and the terminal software's echo buffer was sized for the slower world it had been written for. Duvall later compared the mismatch to "filling a glass of water with a fire hose": the buffer overran before the program could catch up, and the connection went down mid-word. It wasn't a network failure. It was software built for human fingers, meeting a machine that no longer needed to wait on one.

Was ARPANET built to survive a nuclear war?

It's the detail most people think they know, and it isn't quite right. The idea that ARPA designed the network to keep military communications alive after a Soviet strike traces back to a RAND Corporation study on distributed communications that Paul Baran wrote earlier in the decade, and RAND did brief the Air Force on it. But Baran's report wasn't the ARPANET project, and Larry Roberts and Bob Taylor both said for decades afterward that survivability never drove their design. Their stated problem was simpler and pettier: too many expensive, underused computers that couldn't talk to each other. The myth survives because it makes a better headline than "the government wanted researchers to stop buying duplicate mainframes."

How did two computers become the whole internet?

UC Santa Barbara's IMP arrived on November 1; the University of Utah's followed in early December, completing the four-node network ARPA had planned from the start. Growth from there was fast by the standards of 1970s computing: Ray Tomlinson, a programmer at the contractor BBN, sent the first networked email in 1971, and by 1972 the ARPANET had grown to 37 connected computers. The Sigma 7 in UCLA's Boelter Hall and the 940 in Menlo Park are both long gone, but the addressing scheme that eventually replaced ARPANET's original protocol, Vint Cerf and Bob Kahn's TCP/IP (switched on across the whole network on January 1, 1983), is still, with updates, what carries a text message or a video call today. ARPANET itself was formally shut down in 1990, its remaining sites already migrated to the growing academic network that became the modern internet's backbone.

None of the people in that room on October 29 described what they'd done as a beginning. Kleinrock's own account of the night spends more words on the crash than on anything that came after it, and the only surviving record is a single line in a lab notebook, not a photograph or a recording. Fifty years on, the routers that move a search query or a bank transfer trace their lineage, however indirectly, back to a UCLA basement and a login command that made it two letters before it died. The internet's actual first word was an accident nobody thought worth writing down more carefully, which might be the most honest thing about how the internet got built at all: piece by piece, by people debugging a crash rather than making a speech.

Sources

• Leonard Kleinrock, "The Day the Infant Internet Uttered Its First Words," UCLA, lk.cs.ucla.edu/internet_first_words.html
• Computer History Museum, "History of the Future: October 29, 1969 — Fifty Years of a Connected World"
• SRI International, "ARPANET," sri.com/hoi/arpanet
• ICANN, "The First Message Transmission" (blog post, October 29, 2019)
• Wikipedia, "ARPANET"