ARPANET's First Message: Why 'LO' Reached SRI Before the System Crashed

At 22:30 Pacific time on October 29, 1969, programmer Charley Kline sat at UCLA’s SDS Sigma 7 and attempted to log in to Bill Duvall’s SDS 940 at Stanford Research Institute. Kline typed an L; Duvall confirmed by telephone that it arrived. He typed O; that arrived too. When the next character was sent, the SRI host failed. The surviving “LO” became famous precisely because a mundane debugging failure interrupted an otherwise practical remote-login test.

The event is well documented, but its usual one-sentence retelling compresses months of engineering and decades of later internetworking into a creation myth. It was the first host-to-host message over ARPANET—not the first packet ever switched, the first computer network, the first email, the World Wide Web, or a fully formed Internet speaking TCP/IP.

Two incompatible hosts, two IMPs, one test

UCLA was ARPANET’s first node. Bolt Beranek and Newman had won ARPA’s contract to construct Interface Message Processors, dedicated packet switches placed between host computers and the leased-line subnetwork. The first IMP arrived at UCLA around the end of August 1969; SRI received the second in early October.

The IMP architecture mattered because ARPANET did not require every host to be the same model. Each institution had to build host software and a hardware interface to its local IMP, while the IMP network handled packet movement between sites. UCLA’s Sigma 7 and SRI’s SDS 940 were different systems with different local operating environments. Successful communication required both host interfaces, both IMPs, the line, and immature host protocol software to agree.

This was not a consumer “router setup.” Teams were implementing a new system while specifications were still being discussed. Steve Crocker’s RFC 1, dated April 7, 1969, described host software ideas months before the first four-node network existed and explicitly presented tentative conventions rather than a finished Internet standard.

What Kline was actually trying to send

The operational goal was remote login, one of ARPANET’s central resource-sharing use cases. UCLA accounts commonly describe the intended command as LOGIN. Leonard Kleinrock’s account adds a useful implementation detail: UCLA would type LOG, and the SRI system was expected to complete the command with IN.

That distinction explains why the famous message consists of only two characters. L crossed successfully and was acknowledged over the phone. O crossed successfully. On the next step, the SRI host software crashed before the login could complete. The network was repaired and the full login succeeded roughly an hour later, according to Kleinrock’s reproduction of the UCLA IMP log.

“LO” was therefore not a planned salutation. The later “lo and behold” interpretation is a memorable joke, not evidence of authorial intent. The experimenters were entering an ordinary command and checking delivery character by character while talking on a separate telephone connection.

What the surviving log proves

The UCLA IMP log records the date, 22:30 time, and activity surrounding the transmission. It anchors the event more strongly than later anniversary slogans. The source and destination hosts, operators, and failure sequence are corroborated by UCLA, DARPA, SRI/Computer History Museum accounts, and participant recollections.

It proves that two attached hosts could begin an interactive exchange across the operational two-node ARPANET. It also proves how incomplete the system still was: a remote host/software fault stopped the application-level operation after two characters. That is not a contradiction. Early network milestones are often partial demonstrations whose failures expose the next engineering task.

The log does not prove that the Internet as presently defined began in one indivisible instant. Historical labels such as “the first words of the Internet” are retrospective shorthand. The physical/subnetwork event belongs to ARPANET in 1969; the open internetwork protocol architecture associated with today’s Internet emerged through later packet-radio, satellite, gateway, TCP, and IP work.

Why the IMP was more than a modem

The IMPs formed a packet-switching subnetwork. Hosts handed messages to their IMP; the system divided and transported data through standardized switching nodes rather than dedicating a host-to-host circuit for the entire session. BBN’s machines also isolated much network-control complexity from heterogeneous research computers.

Calling an IMP simply “the first router” is understandable but incomplete. The device and protocols belonged to an ARPANET-specific design, including host/IMP interfaces and message formats that evolved substantially. Modern routers operate in the later Internet architecture. The family resemblance is real; identity is not.

The architecture also refutes the popular image of a single UCLA computer somehow “sending the Internet.” The working path depended on ARPA program management and funding, BBN engineering, telephone facilities, UCLA and SRI host teams, protocol discussions across the research community, and prior packet-switching theory and experiments.

The four-node network came after the two-letter message

On October 29 only UCLA and SRI could participate in this host-to-host test. UC Santa Barbara and the University of Utah joined by the end of 1969, producing the canonical first four-node ARPANET map. It is anachronistic to place all four behind the “LO” exchange.

The larger public demonstration arrived in 1972. Email became ARPANET’s unexpectedly important application after the first message. Host-to-host Network Control Program conventions matured before TCP/IP. Each is a separate milestone, and collapsing them into October 1969 erases the engineering sequence that made the network useful.

ARPANET was not yet running TCP/IP

The 1969 test did not use today’s TCP or IP. Early ARPANET host communication developed into NCP, which provided services within that network. Work by Robert Kahn, Vint Cerf, and many others on connecting unlike packet networks came later in the 1970s.

The ARPANET transition plan set January 1, 1983 as the date to move from NCP services to IP/TCP. That transition—more than thirteen years after “LO”—is a better boundary for the operational TCP/IP Internet, although even it was a managed process with dual-protocol hosts and relay arrangements before the deadline.

The World Wide Web is later still: it is an application/information system built on Internet infrastructure, not another name for ARPANET. Nothing resembling a browser or web page was transmitted during the UCLA test.

The crash is part of the achievement

Retellings often treat the failure as comic decoration. Technically it is the most honest part of the story. The link and IMP path carried characters successfully; the end-to-end application failed; operators diagnosed it and completed the login the same night. That sequence resembles real network commissioning more than a ceremonial switch-on.

The corrected historical claim is deliberately narrow: at 22:30 on October 29, 1969, the first recorded host-to-host ARPANET message began from UCLA’s Sigma 7 to SRI’s SDS 940. The characters received before the SRI system failed were LO. A successful login followed after repair.

That narrow claim is important enough. It captures the moment a planned packet network stopped being only specifications, switches, and local interfaces and began carrying an interactive host exchange between institutions—without pretending every later layer of the Internet was already present.

Related:

Sources: