The Actually Existing Internet: Opening the Internet (1969-1991)

Liam Mullally

1st February 2024

Figure 1: Visualising the global topology of the MBone (1996)


Before the internet, there was ARPANET. This is the second of my blogs looking at the changing shape of the actually existing internet, focusing on its early years, first as a military research project, later expanding to a number of adjacent academic fields and finally becoming a publicly accessible network. This is the period in which many of the basic terms of today’s networked communication emerge, as do the first major public infrastructure projects over which today’s internet would be built. 

ARPANET was an American communications project intended to consolidate limited technological resources between the many R&D projects being funded by DARPA (Defence Advanced Research Projects Agency) during the Cold War (see Ben Tarnoff’s book, Internet for the People for a concise account of this relationship). While not the first computer network, ARPANET was to become the centre of an interconnected group of networks we know today as the internet. The ARPANET also contained some innovations: it differed from prior networks through its use of packet switching, a technique which transmits data across the network in small packets. This has the effect of greatly increasing transmission rate, especially in cases such as DARPA’s computation sharing initiative (and the internet of the day) where many actors send messages on the same network simultaneously.

Figure 2: An early ARPANET node at UCLA.

Networking has always been as much an issue of infrastructures as of protocols, and very early on – while the internet was still just a research project – it became necessary for the research apparatus to adapt itself to serious construction. Much of this work was done by BBN, a private research company spun off from MIT in the late 1940s (and which still exists to this day, as Raytheon BBN). While BBN was a private research organisation, it was almost entirely funded by state research funding. In fact, almost all the work on and around ARPA was paid for by the state, mainly via DARPA itself. It is important to stress, given what will come later, that for the first 25 years or so of its life ARPANET and the internet comprised almost entirely of publicly-funded research and publicly-owned infrastructures. Of course, this was also a military project, even if potential military applications of the work were seen as very far off (again, see Internet for the People). It would be far too simplistic to suggest that the internet was ever a solely military endeavour, but one can’t avoid the military orientation of this early work. If it was not the case for all the internet’s developers, its funders certainly had this purpose in mind, and if networking promised no military applications, the work would never have gone ahead.

Figure 3: Two of the earliest maps of the ARPANET, drawn by hand and with just a few nodes.

Beyond the United States

At least, this was the case in the US. One consequence of ARPANET’s centrality is that the early history of the American internet is often taken as that of the internet in general; there is some important work to be done undoing this impression, and parallel efforts were taking place around the globe: the economic management project Cybersyn in Chile, the National Physical Laboratory Network in the UK and Akademset in the USSR are just a few examples. Unlike in the USA, much of this work was not explicitly funded or integrated into military research structures. Some such networks remained entirely distinct projects from that of ARPANET, with their own goals, communications protocols and logics. This was the case for Cybersyn (violently destroyed in 1973 during Pinochet’s coup) and Minitel (the most successful of a number of web-like systems developed by state owned telecommunications organisations in Europe) which ran in parallel to the internet until 2012. Most of these networks, however, were not displaced but subsumed into the particular set of networking protocols which came out of ARPANET and became what we call the internet.

In 1973, SATNET (or the Atlantic Packet Satellite Network), linked computers from ARPANET to Europe for the first time via the existing NORSAR (seismic research) satellite communication link between North America and Norway. As with much of the existing ARPANET, the system was built by BBN technologies. At the same time, PRNET (Packet Radio Network) tested using radio connections to network with a mobile vehicle.

Figure 4: This 1973 map of the ARPANET shows the SATNET connection to London in the bottom right (you can also see Hawaii’s ALOHA network connected similarly on the left hand side of the diagram).

OSATNET was one of a few early experiments in the interlinking of networks in which a fundamentally political, rather than technical problem, became apparent. While these were all packet-switching networks, they operated with different packet sizes, protocols, labelling conventions, etc; in effect, they held different rules for communication. The transmission of data between networks required some set of mutual rules, but what policies ought the network to adopt to manage the interface between these rulesets?

These protocols embody the concerns of an experimental research network, not of the later private network, and so they contain a certain kind of openness: they do not discriminate between users or kinds of data.

In 1977 SATNET was to become the first test-case (along with ARPANET and PRNET) for the Transmission Control Protocol (or TCP), a transport protocol responsible for coordinating the transmission of data across a network. With TCP came a number of basic features of today’s internet architecture, including routing addresses (today IP addresses). To this day, TCP/IP, which (along with UDP) became known as The Internet Control Suite, remains the core set of communications protocols used on the internet; all information transmitted via the internet is organised by them. And it is not alone in this longevity, many protocols designed around the needs of ARPANET and the early internet remain in use. These protocols embody the concerns of an experimental research network, not of the later private network, and so they contain a certain kind of openness: they do not discriminate between users or kinds of data. While TCP and UDP differ in their prioritisation of fidelity (TCP) and real-time transmission (UDP), the priority of these protocols is always the effective distribution of bandwidth between all traffic on the network. In the context in which these protocols were developed, the internet even operated on an assumption of full transit – that all participating networks would transmit information for one another – but such an arrangement has obviously not endured into today (and would in fact break down by the early 1990s). One could of course imagine a situation in which a similar offer might return via state infrastructure.

Figure 5: Diagram of early TCP/IP tests, which involved satellite connections to both the RCNET mobile van and a SATNET satellite station across the Atlantic in Norway.

The birth of the Internet

Early networking efforts which would come together to form the British internet came from a few locations. This would include the NPL Network (developed by the National Physical Laboratory, an industrial standards organisation) and a number of regional university networks including the South West Universities Computer network. It’s notable that this – not so overtly a military endeavour – was being joined up to a US military communications project. Indeed, while ARPANET and related work in computing up to the 1980s remained ostensibly military in origin, and funded by the US Department of Defense, contradictions between this imperative and much of the actual work being done were already apparent – classified military communications were being transmitted on the same network that now carried Usenet messages! This was evidently not sustainable, and culminated in a split between ARPANET and MILNET (literally “Military Network”) in 1983.

While the “birth of the internet” is often located in this split between ARPANET and MILNET, it might also be placed with SATNET and the introduction of TCP/IP a few years earlier. Thus, the arrival of the internet was a consequence of both the joining up of separate fledgeling networks, and the bifurcation of what was then still the largest network into separate civic and military infrastructures – ARPANET and MILNET. If SATNET signals the arrival of an international internet, MILNET (by exorcising the military priorities of ARPANET) signals the birth of the internet as a civic commons (albeit one still only accessible to a relatively small group of experts/professionals – only government agencies, universities and a small number of connected private enterprises could access it. A number of communication projects which were not clearly academic in character began to be distributed over the internet, first email, then USENET (not a network as such, but a discussion and filesharing system which prefigured later internet forums) and later Internet Relay Chat (IRC) which similarly prefigured direct messaging.

This backbone [NSFNET] was also an early indication that the internet could no longer model itself as just the horizontal linking up of networks, that its distribution could not be even and that certain networks would be more important for data carriage than others. Necessarily, this question of shape would come to take on a political character.

From the early 1980s there began a massive proliferation of computer networks, both in and outside the US. Notable among these was the Computer Science Network (CSNET), founded in 1981 by the National Science Foundation, a US Federal Government agency with a mandate to support scientific research. There is a break here from the almost exclusively military support for such work in the 60s and 70s towards a different understanding of the internet, as an information technology fit for resolving the kind of academic information crisis described by Vannevar Bush in the 1940s: “The investigator is staggered by the findings and conclusions of thousands of other workers – conclusions which he [sic] cannot find time to grasp, much less to remember, as they appear.” The CSNET was superseded by the National Science Foundation Network (NSFNET) in 1985. For much of its history the NSFNET maintained strict rules about who could use it (research, not commercial activity), and so helped to maintain the internet as a limited and elite venture. At the same time, however, it was host to a series of important infrastructure projects that would make it the first example of what today is called an “internet backbone”, core infrastructure that acts as a route for vast amounts of data to be transmitted, rather than a discrete network onto itself. This backbone was also an early indication that the internet could no longer model itself as just the horizontal linking up of networks, that its distribution could not be even and that certain networks would be more important for data carriage than others. Necessarily, this question of shape would come to take on a political character.

Figure 6: Each of the nodes in this 1985 map by BBN represent individual networks in a wider internet.

The Exterior Gateway Protocol

And so, in 1982 the first decentralised gateway protocol, the Exterior Gateway Protocol (EGP) was published by researchers at BBN with the intention of standardising this interface between networks. Paul Dourish has written about how protocols become locations of political contestestation in decentralised networks, yet remain largely unexamined. Data travelling across the internet will need to pass between a number of intermediary networks to reach its destination – how this information travels is determined by the gateway protocol (which gives individual networks information about adjacent networks). In the change between simple routing protocols like the Routing Information Protocol (RIP) and the EGP Dourish finds, for instance, evidence of the centralisation of internet traffic. There is a parochial politics embodied in such protocols, which handle the interchange of information between ASes, but this intersects with more conventionally conceived political concerns. As Dourish demonstrates, the choice of which protocol to use has important consequences for the efficacy of implementation in different communication networks; the US is a massive geographic space, with big gaps between significant nodes and networks, whereas North-West Europe is comparatively densely populated, and so the infrastructures of the US and Europe were very different (operating different packet sizes, for instance). Interface, required negotiation between these different geographies – and compromise which remains in the protocols used to this day.

The EGP brought with it an early definition of autonomous systems (ASes) as discrete networks within the internet, defining them as groups of interconnected computers which share a routing policy. This is a definition which spans a number of scales, from small experimental networks of just one or two machines to vast backbone infrastructures. Nonetheless, the EGP began the process of assigning each of these networks an AS number (or ASN). In practice, ASes are mostly a function of routing protocols and IP address allocation, but they also result in the creation of large volumes of metadata. Since ASes can be thought of as collectively making up the material routes of the internet, tracing this metadata is one method of visualising the internet’s structure. Such an account is complicated somewhat by the fact that autonomous systems do not describe servers, computers or processing power, but network routes. You can think of ASes as constituting the “paths” of the internet, but not the homes, warehouses, offices, or the traffic itself.

ASNs are assigned (along with IP addresses) by the Internet Assigned Numbers Authority (IANA) and a number of regionally devolved authorities; RIPE (Réseaux IP Européens Network Coordination Centre) maintains a list of all ASNs, but crawling the BGP (Border Gateway Protocol, successor to the EGP) offers the clearest view of AS level information. Looking at the lowest AS numbers can tell us something about this earliest period. Several early networks claim to have been registered on 1st January 1970, 00:00:00 in the Unix time standard (the beginning of time in many computational systems). This indicates an error in the data, but we can fairly safely assume that they were allocated in between the previous and following ASes. More troubling are registry dates which appear to be much later – in these cases AS numbers have been reallocated, and so will not accurately reflect the original owners of the number. Nonetheless, the data shows that the first 100 ASNs were allocated between early 1984 and February 1987. After another three years, AS561 was allocated to HP on 1st January 1990, showing a marked acceleration, but this was nothing beside what was to follow; by the 1st January 1991 there were at least 1000 ASes. Shortly after, with the network growing more complex and global in nature (including the reservation and allocation of ASNs in large blocks), and an increase in the number of reallocated ASes, the relationship between AS number and age of network begins to break down.

Nonetheless, there’s plenty to learn from this early period of growth. Of the first 100 ASes, the vast majority fit into one of three categories: universities and state funded research institutes, military and defence organisations, and a network of think tanks and not-for-profit organisations. In addition to these, there is a smaller number of private computing and networking companies, many of which spun directly off from computing labs within universities, especially MIT. BBN (via its service BBN Planet) was the first organisation to receive an AS number, early in 1984. Just over a decade earlier, BBN had spun off a subsidiary, Telenet. Led by an ex-employee of DARPA, Telenet was conceived of as a commercial extension to the ARPANET. Telnet and BBN entered into a particularly messy (and poorly documented) set of acquisitions, split-offs and mergers from the 1980s into the 2000s, with the service running under several names including BBN Planet, GTE Interworking and Genuity during the period. In its many iterations throughout this period, the service formed into the first of what we today call an Internet Service Provider (ISP), a private company that sells internet access to end-users.

Figure 7: By the mid-1980s, the internet was connecting all major research institutions and most major population centres in the US.

Going submarine: TAT-8

Of these earliest AS allocations only a few are not owned and operated in the US: in Britain the Defense Science and Technology Laboratory, a Canadian research initiative. This lack of low ASNs from outside the US doesn’t suggest a lack of activity elsewhere, only that other networks were not yet directly interfacing with the early American internet; a direct consequence of the fundamental topographical barrier of the Pacific and Atlantic oceans. While experiments like SATNET had allowed some interchange between European and American networks (and been vital to the development of TCP/IP), they represented very little bandwidth. The real breakthrough in this regard came in 1988, with the installation of the first fibre-optic submarine cable (TAT-8), linking New Jersey to Cornwall in England and Brittany in France.

TAT-8 was a collaborative project between major telecommunications organisations on both sides of the Atlantic (British Telecom, France Télécom and AT&T), and was intended to act as a channel for telephone signals. The new capacity it brought also acted as a new route for internet connections between Europe and America. Not only did this high speed connection allow Tim Berners-Lee to access the NSFNET while demonstrating the World Wide Web in following years, it also became a major factor in the widespread adoption of TCP/IP across Europe. From this moment on submarine cables would also become a major vector of global internet connection; today they represent 99% of all transoceanic digital communication. As Nicole Starosielski describes compellingly in her book the undersea network, if these vectors did not exist, “the Internet would effectively be split between continents.”

After TAT-8 and the adoption of TCP/IP in Europe, the ISP model also proliferated. CompuServe, Pipex and Demon Internet all offered commercial internet access in the UK during the late 1980s or early 1990s, with a huge number of commercial services emerging within a few years. While the ISP was the most widespread model of internet connection that emerged from the 1980s, it was not the only one. In 1986 a community network was established in Cleveland, the Cleveland Free-Net. Curiously, both Bell and AT&T donated resources to the project, although its funding largely came from Case Western Reserve University. The Cleveland Free-Net gave internet access along with a number of services (although these varied over time), including email, newsgroups, chat rooms, file-sharing, community archives and web access to anyone in Cleveland with a PC or access to a terminal in a library; this model of locally governed municipal internet provision proliferated across several cities in North America, including Detroit, Ottawa, Dillon and Austin, as well as Helsinki in Finland.

… this was the moment the internet became publicly accessible, but the privatisation of its infrastructures had already begun.

Following the birth of the commercial internet, ARPANET was officially discontinued in 1990, with DARPA instead pursuing partnerships with the private telecommunications and the nascent information services industry. A year later NSFNET would drop restrictions to its infrastructure, opening its backbone to commercial services; a race was sparked between a growing number of ISPSs to integrate more and more users into the network. The number of users with access to the internet began to grow rapidly; this was the moment the internet became publicly accessible, but the privatisation of its infrastructures had already begun.

Selected bibliography

  • Ben Tarnoff, Internet for the People: The Fight for Our Digital Future (London: Verso, 2022)
  • Paul Dourish, “Protocols, Packets, and proximity: The Materiality of Internet Routing”, Signal Traffic: Critical Studies of Media Infrastructures (Urbana: University of Illinois Press, 2015)
  • Benjamin Peters, How Not to Network a Nation: The Uneasy History of the Soviet Internet (London: MIT Press, 2016)
  • Eden Medina, Cybernetic Revolutionaries: Technology and Politics in Allende’s Chile (London: MIT Press, 2014)

Liam Mullally is a CHASE-funded PhD candidate in Cultural Studies at Goldsmiths, University of London. His research interests include digital culture, the history of computing, information theory, glitch studies, and the politics and philosophy of noise. Previously he has worked as a copywriter and tech journalist. He is working on several projects with Autonomy, from skills commissions to policy strategy.