Much as it had done for timesharing and interactive real-time computing, the Q-32 in Santa Monica foreshadowed the 1977 dawning of the ARPANET and Internet at UT Austin and in Texas. The Q-32 was retired around 1970 and did not itself serve as a node in the operational ARPANET, but it played a critical role as a direct technical ancestor. In the sixties, computer timesharing and computer networking, packet switching in particular, evolved quickly as associated areas of research. Computing resources were scarce and expensive, and the overall motivation was to share the available resources among more users, with both timesharing and packet switching slicing the resources into increasingly granular pieces for better distribution both temporally and spatially.
The Q-32 was used for early long-range networking experiments. In late 1963, researchers established a 300-mile link between the Q-32’s PDP-1 front-end in Santa Monica and a CDC 160A minicomputer at the Stanford Research Institute. This connection utilized two full-duplex telephone lines. The defining architectural feature of this experiment was the strict separation of control and data. One telephone line was dedicated to sending standard executive commands while the second line was dedicated to raw data. Because the data line bypassed the Q-32's executive command parser, the data routed directly to the active program. This setup allowed remote SRI users to interactively perform full-text searches on bibliographic databases stored on the Q-32's high-speed magnetic drums.
By 1969, the early experiments with the Q-32 had led onwards to UT Austin and the NSF establishing a large-scale regional network, the Southwest Region Educational Computer Network. Its goal was to provide computing power to smaller colleges and universities across Texas. This regional network was a socio-technical experiment funded largely by the NSF to mitigate the prohibitive upfront costs of mainframe computing. By leveraging a hub-and-spoke model, the southwest network provided the computational plumbing necessary for remote batch processing and timesharing, which in turn created a fertile environment for computer-based education to flourish. It functioned as the vital infrastructure, the hardware backbone and telecommunications architecture, that linked the central processing power and software of the UTCC to a distributed web of smaller colleges, junior colleges, and secondary schools.
The southwest network was centered on the UTCC CDC 6600 and 6400. Schools such as Southwest Texas State in San Marcos, Rice in Houston, Trinity in San Antonio, Austin High and McCallum High, and various junior colleges connected to this central hub via remote terminals, usually teletypes. It expanded from nine initial institutions in 1969 to twenty-three in 1972, with UT Austin serving as the central host. It democratized access to high-level processing power for both administrative tasks and, more importantly, classroom instruction. While the network was a technical success and demonstrated that communication lines and terminals could be stabilized across a distributed institutional landscape, it reached a definitive structural limit. The project successfully navigated the connectivity bottleneck but failed to address the more elusive problem of pedagogical effectiveness. The network proved that data could be transported, but it lacked the content necessary to justify the technology’s presence in the classroom. This highlighted a critical gap. Technical availability did not equate to educational utility, necessitating a shift toward the production of high-quality instructional materials.
In 1965, ARPA funded a project proposed by Thomas Marill and overseen by Larry Roberts to link the Q-32 directly to the TX-2 computer at MIT's Lincoln Laboratory in Massachusetts. This experiment proved that two independent, geographically distant time-sharing operating systems could exchange digital data and invoke programs remotely. It also exposed severe flaws in the era's networking capabilities. These technical frustrations directly convinced Roberts, who would subsequently become the program manager and principal architect of the ARPANET, that building a robust, large-scale computer network would require abandoning circuit-switching in favor of packet-switching technology.
Marill and Roberts devised what they called the "elementary approach". The primary goal was to bridge the gap between the two structurally incompatible operating systems, the TX-2's APEX and the Q-32's TSS, without having to modify or rewrite their complex core kernels. Under this protocol, the local user executed an application that intercepted terminal inputs, repackaged them, and directed them over the communications link so that the remote host monitor treated the connection exactly as if it were a local user terminal. A successful demonstration of this setup involved a researcher at the TX-2 in Massachusetts utilizing a program called the Algebraic Translator to automatically dial the Q-32 in California. The program bypassed standard human login prompts to gain administrative access, loaded a Lisp compiler on the Q-32, transmitted a complex Lisp program across the country, and had the Q-32 execute the calculations before returning the results to the local TX-2 console.
The experiment was a technical success but a practical headache, proving two major things. A user on the TX-2 in Massachusetts could log into the Q-32 in California and run a time-shared program. It showed that computers didn't just have to talk to human typists. They could talk directly to each other and share workloads. And it highlighted the fatal flaw of using standard telephone infrastructure for computing. Because the connection used circuit switching, with a dedicated analog line, it was unreliable and inefficient. Because human-computer interactions and data exchanges happen in short bursts, the line remained idle for most of the session, resulting in extremely low bandwidth utilization and proving that circuit-switched computer networks would be prohibitively expensive. Furthermore, the analog lines were highly susceptible to electrical noise and signal attenuation, which frequently caused data corruption through bit-flipping. Since the remote software did not have automated, system-level error detection and recovery, a single flipped bit meant the user’s local software had to abort the session, clear remote memory buffers, and retransmit the entire data block.
Bob Taylor, the director of ARPA's Information Processing Techniques Office and a UT Austin graduate [1], recruited Roberts from Lincoln Laboratory to become the program manager and principal architect of the ARPANET in Washington. Remembering the lessons of the 1965 Q-32 and TX-2 hookup, Roberts realized that instead of dedicated phone circuits, the new network had to use packet switching to handle the data efficiently. Some researchers advocated for the dual-line model used in the 1963 Q-32 and SRI experiment, arguing that separating command and data channels simplified hardware and eliminated processing overhead. Roberts argued that leasing dual transcontinental telephone lines was economically unsustainable for large-scale networks. He insisted that any viable network must use a single physical channel, with the operating system dynamically multiplexing and parsing commands and data. The 1965 experiment's severe telecommunications bottlenecks had decisively proven that scaling a wide-area network required abandoning circuit-switched telephone lines in favor of packet-switching technology. By breaking data into small, self-contained packets, the network could maximize line utilization and automatically handle error recovery through intermediate switches, rather than forcing the user's application program to handle all error checking.
Drawing on his experience connecting the TX-2 and Q-32, Roberts initially proposed at a 1967 meeting that all host mainframes connect directly to one another and manage their own network administration, terminal routing, and error checking. Mainframe operators, termed principal investigators in the ARPA context, fiercely opposed this plan. They were highly protective of their expensive mainframes and refused to sacrifice 10% to 15% of their limited computing power to handle network overhead.
In response to this resistance, computer scientist Wesley Clark told Roberts, "You've got the network inside out," and proposed a decoupled architecture. Clark suggested installing a small, standardized minicomputer at each site to act as a universal interface. These minicomputers would all speak a uniform language and handle all the routing, packet buffering, and error-checking tasks, completely insulating the main host computers from the network overhead. Roberts adopted the idea and named the specialized communications processors Interface Message Processors or IMPs. By separating communication logic from application processing, the IMPs allowed structurally incompatible mainframes to easily join a unified wide-area network, successfully establishing the foundational architecture of the ARPANET and the modern Internet.
Thanks to Clive Dawson, it's now known how the first ARPANET IMP in Texas arrived at UT in 1977, and in some sense the Southwest Network began fusing into the ARPANET. The first IMP was initially installed alongside the HRC DEC-10. Because the standard TOPS-10 operating system did not yet support interfacing with the IMP, the DEC-10 did not become the first Internet computer in Texas. The many PDP-10s across the ARPANET from its earliest days in 1969 were running custom operating systems, and the HRC machine was kept factory stock. Instead, the first actual host to connect was a PDP-11/45 running a hacked version of UNIX, located across campus in the Daily Texan composing room. Accessible via dial-up modems by a small group of users, this machine became the UTEXAS host on the net and served as the UT's pioneering gateway.
Charles Warlick discussing the Southwest Network in 1973 Southwest Network operations in the UTCC machine room circa 1973 Computation Center subterracean heart of the Soutwest Network circa 1973
[1] Bob Taylor earned a masters in psychoacoustics from UT in 1959. Interestingly, acoustics research at UT during the forties and fifties was connected with Bolt Beranek and Newman in Boston, and BBN would become central to the ARPANET from 1969 onward when it won the contract to create and operate the IMPs. UT Austin and BBN frequently collaborated on major federal and commercial acoustics projects. Researchers at UT Austin have conducted sponsored research and acted as co-chief scientists on defense-oriented ocean acoustic initiatives via BBN and Raytheon. There are also quite likely ties with the Linguistics Research Center at UT from the forties and fifties. It's reasonable to say that Taylor arrived in his ARPA role because of Joseph Licklider, a psychoacoustice researcher, BBN employee, and ARPA appointee.
