"LO": The First Message Sent on the ARPANET

Abstract

The first message sent on the ARPANET, on the evening of October 29, 1969, was “LO.” Charley Kline, a student programmer at UCLA, was trying to log in to the SDS 940 at the Stanford Research Institute (SRI) in Menlo Park, about 350 miles away. Bill Duvall at SRI watched the characters arrive and confirmed them over a telephone line. When Kline typed the “G” of “LOGIN”, the SRI system stopped: an echo buffer was too small for the word the 940 tried to send back. Duvall fixed it, and about an hour later the full login went through. The UCLA log records only the success, in one line: “Talked to SRI host to host.”

The UCLA IMP log, with Charley Kline’s 22:30 entry of October 29, 1969. Image: Andrew “FastLizard4” Adams from United States, CC BY-SA 2.0, via Wikimedia Commons.

The Two-Node Network

A BBN Interface Message Processor at the Computer History Museum. Image: ArnoldReinhold, CC BY 4.0, via Wikimedia Commons.

On the night of the first message, the ARPANET had exactly two nodes. UCLA’s IMP had been installed at the end of August 1969 and SRI’s arrived on October 1; the other two sites came later that year, UCSB in November and the University of Utah in December, completing the four-node map by December 5, 1969. The node numbers followed that order, from UCLA as node 1 to Utah as node 4.

Each node had an Interface Message Processor (IMP), a Honeywell DDP-516 minicomputer programmed by Bolt Beranek and Newman (BBN) that handled packet routing and error detection. The host computers at each site connected to their IMP, and the IMPs connected to each other over leased telephone lines.

Larry Roberts managed the ARPANET project at ARPA (renamed DARPA in 1972); it grew out of the theoretical work of J.C.R. Licklider. The full engineering story of the network’s construction is in ARPANET: Building the Network.

Two Programmers, Two Programs

The IMPs could pass packets between them, but the host computers still had to be taught to use them. At each end, one programmer wrote that software.

At UCLA, Charley Kline, a student in Leonard Kleinrock’s Network Measurement Center, wrote the client program on the lab’s SDS Sigma 7, which let a user at UCLA act as a terminal on a remote machine. At SRI, Bill Duvall modified the time-sharing system of an SDS 940 so that it would accept such terminal connections from the network. The 940 was the machine on which Douglas Engelbart’s lab at SRI ran its NLS system. Its time-sharing system was written in assembler, and the whole machine had at most 65K words of 24-bit memory, shared by the operating system, its drivers, and active data. Every buffer was kept as small as possible.

For the test, each of them put on a telephone headset. Besides the network line, there was a separate, dedicated voice line, so the two could tell each other what their machines were doing.

The Login Attempt

Kline sat in Room 3420 of Boelter Hall at UCLA. The SRI system expected the word “LOGIN” to start a session. “The first thing I typed was an L,” Kline recalled for the Computer History Museum’s 50th-anniversary account. Duvall saw the “L” arrive and start a login process on the 940 and said so on the phone. Kline typed “O”, and Duvall confirmed the “O”. Then Kline typed “G”. “And he had a bug and it crashed.”

So the first characters ever carried host to host over the ARPANET were “L” and “O.”

Why It Crashed

The SDS 940 ran full duplex: every character it received, it echoed back to the terminal, here across the network to UCLA. The protocol Kline and Duvall had implemented assumed one character in, one character echoed out. But the 940’s command interpreter had a completion feature. After “L”, “O”, and “G”, it recognized the command and completed it, echoing “GIN” in one go. Three characters did not fit the echo buffer, which had been sized for one.

Duvall found the problem quickly, enlarged the buffer, and restarted the system. Kline typed “LO” again, received the completed “LOGIN” back from SRI, pressed RETURN, and was logged in. In 2019, Duvall and the network pioneer Steve Crocker described the fault in a joint account as a buffer “sized too small”; other retellings call it a configuration problem or simply a bug, which all describe the same thing at different levels of detail.

The Log Entry

The UCLA lab kept a handwritten IMP log on a printed customer-service form. The page for October 29, 1969, has an entry at 17:50, written by another operator, about testing the line to UCSB, followed by the note “Charley please call Ben at SRI!” At 21:00 Kline (initials “CSK”) loaded an operating program for Ben Barker of BBN. At 22:30 comes the line that made the page famous: “Talked to SRI host to host.” Below it, Kline added that he had left the IMP program running after sending a “host dead” message to the IMP.

The log does not mention “LO,” the crash, or the phone calls. Those details come from the recollections of Kline, Duvall, and Kleinrock. The timing is also less exact than the 22:30 entry suggests. That entry records the success; the crash came about an hour before it, and the Computer History Museum’s 2009 anniversary release placed Kline’s first attempt at around nine in the evening and the working connection at 10:30. Kleinrock’s own web page compresses the two and dates the “LO” itself to 22:30.

What the Network Was For

The ARPANET’s official purpose was resource sharing: allowing researchers at one institution to use computational resources at another. UCLA had a powerful computer; SRI had different software and data. Remote login meant a researcher at UCLA could use SRI’s machine without traveling, and the first test was a login for that reason.

In practice, the network went elsewhere. Ray Tomlinson added networked email informally in 1971, with no central planning, and a usage study commissioned under ARPA director Stephen Lukasik found that it made up about three quarters of the network’s traffic. Hafner and Lyon date the study to 1973; Lukasik himself later recalled a 1974 study by MITRE.

The night of the first message showed the same gap between plan and practice: the network could not yet recover on its own, and it took two people on a telephone line to restart a crashed host.

“Lo and Behold”

The two letters are often read as a greeting: “Lo,” as in “lo and behold.” Nobody read them that way at the time. To the people involved, it was a failed login, fixed within the hour. “When we sent that first message, there weren’t any reporters, cameras, tape recorders or scribes,” Kleinrock said for the 50th anniversary in 2019; the team had no idea how much it would matter.

The poetic reading came decades later, in retrospective accounts, and Kleinrock himself took it up when he told the story. In 2016, Werner Herzog named his documentary about the internet Lo and Behold, Reveries of the Connected World after the message; it opens with Kleinrock recounting the evening in Boelter Hall.

The Room

Room 3420 was an ordinary lab for decades after 1969. Brad Fidler, a UCLA doctoral student in the history of science, identified it from photographs and first-hand accounts and restored it to its 1969 appearance, with replicas of the machines and reproductions of documents, including the IMP log. The Kleinrock Internet Heritage Site and Archive opened there on October 29, 2011, the 42nd anniversary of the first message. Fidler described the result as “an immersive environment” with “nothing to break the illusion.” On the evening it commemorates, three people were online.