The story of the internet’s first packet begins not with a flash of inspiration but with a memo. In the spring of 1966, Bob Taylor, then a senior official at the Advanced Research Projects Agency (ARPA), sat down with a growing frustration. He watched researchers at various universities waste precious time shuttling decks of punched cards between locations, waiting for results that could have been shared instantly if the machines could talk. Taylor’s irritation was practical, not philosophical; he wanted a way to make the disparate computers under ARPA’s sponsorship work together as a single resource. He convinced his boss, ARPA director Charles Herzfeld, to allocate a million dollars for a networking experiment. The money was modest by today’s standards, but it was enough to launch a project that would eventually rewire the planet.
Taylor’s vision needed a technical lead, and he found one in Lawrence Roberts, a bright-eyed MIT Lincoln Laboratory veteran who had already been tinkering with the idea of connecting computers over telephone lines. Roberts had participated in earlier experiments with the TX-2 and the AN/FSQ-7, and he understood both the promise and the pitfalls of digital communication. When Taylor offered him the job of chief scientist for the new network, Roberts accepted with the condition that he could bring in a team of engineers who shared his belief that packet switching was the right way forward. The concept itself was not new; Leonard Kleinrock’s queueing theory and Paul Baran’s distributed communications work had laid the groundwork, but turning theory into hardware required a different kind of ingenuity.
Roberts’ first step was to define the network’s architecture. He rejected the idea of a simple point‑to‑point link between every pair of machines—a scheme that would have required an impractical number of connections as the network grew. Instead, he proposed a store‑and‑forward model where data would be broken into small blocks, each packaged with addressing information, and sent through a series of specialized computers that would forward them toward their destination. These specialized computers would later become known as Interface Message Processors, or IMPs. The IMP would sit at the edge of each host computer, translating its native data streams into the uniform packets that could travel the network, and vice versa. By standardizing the packet format, the IMPs could be built once and replicated across many sites, a crucial economies‑of‑scale move.
With the architecture sketched, Roberts turned to the challenge of finding a contractor to build the IMPs. After a competitive bid process, the contract went to Bolt, Beranek and Newman (BBN), a Cambridge‑based consulting firm famed for its work in acoustics and, more recently, in computer science. BBN’s team, led by Frank Heart, arrived at the project with a healthy skepticism about the feasibility of packet switching but also with a determination to prove the doubters wrong. They set up a small lab in a former shoe factory, where they began soldering together transistors, diodes, and the newly minted Intel 1103 memory chips that would give the IMP its modest but sufficient processing power. The first IMP was a rugged, rack‑mounted unit roughly the size of a refrigerator, its front panel blinking with a array of LEDs that would later become the heartbeat of the ARPANET.
While BBN hammered out the hardware, Roberts and his ARPA colleagues were busy lining up the first host institutions. The University of California, Los Angeles (UCLA) was a natural choice; its Computer Science Department, under Leonard Kleinrock’s direction, already possessed a Sigma 7 mainframe and a deep familiarity with the theoretical underpinnings of packet routing. The second node was selected at the Stanford Research Institute (SRI) in Menlo Park, where the Augmentation Research Center, led by Douglas Engelbart, housed an SDS‑940 computer and a burgeoning interest in collaborative computing. A third node was added at the University of California, Santa Barbara (UCSB), and a fourth at the University of Utah, home to the pioneering graphics work of Ivan Sutherland. These four sites formed the initial backbone, a modest quadrilateral that would soon stretch across the continent.
The night of October 29, 1969, was unremarkable in the world outside the UCLA computer lab. Inside, a small group gathered around the Sigma 7, its console lights casting a pale glow on the faces of programmers, technicians, and a few curious onlookers. The message they hoped to send was simple: the word “LOGIN.” The plan was for the UCLA IMP to break the word into packets, transmit them over a 50‑kilobit‑per‑second leased line to the SRI IMP, and have the SRI machine reassemble them and display the result on its screen. At 10:30 p.m., UCLA student programmer Charley Kline typed the “L,” hit send, and watched as the first packet left the host. The IMP acknowledged receipt, and the second packet, carrying the “O,” followed shortly after. Then, as the third packet—bearing the “G”—was being prepared, the SRI IMP crashed. The connection dropped, and the letters “LOGIN” appeared on the SRI screen as “LO.” The historic first message was thus incomplete, a truncated greeting that nonetheless proved that packets could travel, be routed, and be reassembled across a distance of nearly 400 miles.
The crash was not a failure of concept but a symptom of the fledgling hardware’s immaturity. The SRI IMP had run out of memory while trying to buffer the incoming packets, a bug that the BBN team would soon patch. Within an hour, the link was restored, and the full “LOGIN” transmission succeeded. The event, though modest, was captured in a log entry that read, “Node to node communication established.” It was a quiet triumph, barely noticed by the press, yet it marked the moment when a collection of isolated computers began to behave as a single, communicative whole.
In the weeks that followed, the ARPANET grew with deliberate caution. Each new IMP was tested in isolation before being linked to the existing nodes, a process that helped isolate bugs without jeopardizing the whole system. By the end of 1969, the network comprised four IMPs and four host computers, exchanging everything from simple text messages to early file transfers. The traffic was light—often just a few packets per minute—but the pattern was clear: data could be chopped, sent, and reconstructed with reliability that surpassed the circuit‑switched telephone lines of the era. Engineers began to joke that the IMPs were the “postal workers of the digital age,” each packet a letter handed off from one sorter to the next until it reached its destination.
The early ARPANET also revealed unexpected social dynamics. Programmers at UCLA and SRI started exchanging jokes and gossip via the network, a precursor to the casual chatter that would later flood email lists and chat rooms. At UCSB, researchers used the link to share aerospace simulation data, while the Utah team transmitted graphical primitives that would later influence the development of computer‑aided design. These exchanges were not grand scientific breakthroughs; they were mundane, everyday uses that demonstrated the network’s utility beyond the confines of defense research. The ARPANET was proving to be a general‑purpose communication tool, a fact that would later justify its expansion beyond the military‑academic nexus.
Funding for the project continued to flow from ARPA, but the agency’s officials were keen to see a return on their investment. They encouraged the host institutions to document their experiences, produce technical reports, and present findings at conferences. The resulting papers, often dense with diagrams of packet formats and flow‑control algorithms, circulated among a small but growing community of computer scientists. This early literature helped spread the packet‑switching gospel, planting seeds that would later take root in commercial ventures and academic curricula worldwide.
One of the most enduring legacies of the first ARPANET packet was the establishment of a culture of open documentation and collaborative problem‑solving. The IMP software was released under a permissive arrangement that allowed any site to examine, modify, and improve it. This openness fostered a feedback loop: bugs discovered at one node were quickly communicated to BBN, patched, and redistributed. The practice foreshadowed the open‑source ethos that would later dominate software development, and it ensured that the network could evolve rapidly without being bottlenecked by a single vendor’s release cycle.
As the year 1970 dawned, the ARPANET prepared for its first major expansion. Plans were already underway to add nodes at MIT, Harvard, and the RAND Corporation, each bringing its own research agenda and technical expertise. The original four‑node core would soon become a rudimentary mesh, testing the network’s ability to handle multiple routes and dynamic re‑routing when links failed—a feature that would prove vital for survivability in the hostile communications environments imagined by Cold War strategists. Yet, even as the network grew in size and complexity, the memory of that first tentative “LO” transmission lingered in the minds of its creators. It reminded them that monumental change often begins with a humble, almost hesitant step, and that the true power of a network lies not in the speed of its links but in the willingness of its participants to try, to fail, and to try again.