At 10:30 p.m. on October 29, 1969, Charley Kline typed a lowercase l into a computer at UCLA. Before he typed the next letter, he asked Bill Duvall, 350 miles away at the Stanford Research Institute, whether it had arrived.
He asked by telephone.
Duvall confirmed the l. Kline typed o and received another spoken confirmation. He typed g, and the remote system crashed. The intended command was login, so the first host-to-host message sent over ARPANET became the accidental fragment lo.
The crash became the famous part of this story. The phone call explains how experimental the network still was.
Two networks, one conversation
Kline and Duvall wore telephone headsets while testing the connection. Kline would type a character, then ask whether Duvall's machine had received it. The new packet-switched network carried the experiment; the old voice network supplied its human-readable delivery receipts.
This was sensible, not quaint. ARPANET had only two connected sites that night. UCLA had an SDS Sigma 7 host, SRI had an SDS 940, and the machines did not share a common design. Each connected through a special-purpose Interface Message Processor, or IMP, built by Bolt Beranek and Newman. The IMP was a refrigerator-sized packet switch whose job was to give unlike computers a standard doorway into the network.
The Computer History Museum's surviving early IMP contains a Honeywell 516 minicomputer with 12,000 words of memory. It is a useful antidote to the word network, which now suggests something airy and invisible. The first ARPANET connection required large computers, large interface boxes, dedicated communication lines and two people talking through every keystroke.
The call also reveals what the test was really proving. Sending two characters was not impressive by itself. Telegraph systems had moved text over long distances for more than a century. The difficult part was making one kind of computer interact with another through packet switches, then extending that arrangement without redesigning every machine.
According to the Internet Society's history written by several of the network's architects, UCLA's Network Measurement Center became ARPANET's first node in September 1969. SRI followed as the second. UC Santa Barbara and the University of Utah joined by the end of the year, creating the original four-node network. The fully developed Internet came later; ARPANET was one of its principal ancestors, not simply today's Internet with fewer websites.
The successful message was less memorable
The crash was fixed quickly. UCLA's account says the team completed the full login about an hour later. That success did not produce a neat two-letter origin story, so history kept the failure.
Leonard Kleinrock later enjoyed the accidental drama of lo, reading it as "lo and behold." But his longer account of the evening preserves the more revealing exchange: "Did you get the l?" "Got the l." The machine-to-machine connection was so new that the humans verified it one character at a time through a system they already trusted.
New infrastructure rarely begins by replacing the old world all at once. It borrows power, cables, terminology and confidence from whatever is already working. On ARPANET's first night, packet switching carried the letters, but the telephone call told everyone when they had arrived.