office and have a clear written message sent to any large city in the United States?1
Now almost exactly a century later, at&t’s astonishing lack of interest in running the arpanet meant that arpa had to search for an alter- native party to do so. Ironically, it turned to the Defense Communica- tions Agency (dca), whose involvement had been regarded as the kiss of death for packet networking only a decade before.
Military administration
dca had become convinced of the importance of packet-switched digital networks in the interim, and concluded an agreement with arpain 1975 to take over day-to-day responsibility for the operation of the arpanet for a period of three years. After this three-year period arpa expected to have found a private sector company to assume responsibility, yet in reality dca’s administration of arpanet would last until 1983. Thus was the hand of military administration thrust into the community of computer scientists – not a cold and steely mitt perhaps, but certainly something different to the informal and often haphazard procedures that had held sway under arpa.
In July 1980 Major Joseph Haughney, the manager that the Defence Communications Agency had assigned to oversee the arpanet, explained why controls would be tighter:
Now that the network has grown to over 66 nodes and an esti- mated four to five thousand users, flexibility must be tempered with management control to prevent waste and misuse . . . We just want to ensure that we can verify proper resource utilization and prevent unauthorized penetrations.2
dcaattempted to prevent unauthorized access, limit leisurely use of the network and control the sharing of files. At one point an exasper- ated dca officer warned arpanet not to participate in a chain letter that had been widely circulated across the network. This was a waste of net- work resources for which dca would hand out severe penalties ‘up to and including the possibility of removing entire facilities from the net if appropriate’.3
Sparks such as these were inevitable at the juncture of cultures so different as the computer research community and the
military. Nor were they without precedent. The Second World War Army Air Force General ‘Hap’ Arnold, though himself a historic figure in the development of us military research and development, referred to the techies as ‘long-hair boys’.4
Had dca continued to administer the arpanetfor a period of decades, the culture of the Internet might have been marked less by the long-haired eccentrics and more by hierarchi - cal, disciplined military convention.
dca’s administration, however, proved to be temporary. In 1983 the network was split into two. The military facilities that had been con- nected to arpanet were now segregated within a new, secure network called milnet. arpanet and milnet were both connected through gateway machines using tcp/ip, but these gateways could limit access from arpanet to milnet if necessary. With the separate milnet pur- suing rigorous operating procedures under dca’s watchful eye, arpanetwas free to continue as a research network where invention trumped order. So much the better for the Internet.
The longer-term consequence of dca’s tenure as manager of the arpanetwas that the protocols that today knit together the Internet, tcp/ip, were imposed as standard across the entire network. Most facilities connected to the arpanet were reluctant to devote the effort required to convert from the first arpanet protocol, ncp, to the new internetworking protocol, tcp/ip. They did not need to change from the arpanet’s original ncp protocol because they were not networking with other networks beyond the arpanet. Moreover, tcp/ip would be difficult to implement, as ncp had initially been, and would inevitably undergo further adjustments as the technology matured. Despite this reluctance events that were about to unfold within dca would see the Agency make tcp/ip mandatory for all connected, entrenching the internetworking protocol and establishing a basis for the later expan- sion of the Internet.
What follows, the reader is warned in advance, is a steady flow of acronyms. dca decided to develop a second generation of its ‘automatic digital network’ (autodin), which had from 1966 provided commu- nications with us military bases around the globe. autodin ii would become dca’s military communications network while arpanet would be maintained as a space separate for research and testing. tcp/ip would connect the two networks through gateway machines. To allow dcato use tcp/ip in this capacity, the Secretary of Defense made the protocol a military standard in 1980. This meant that when the autodin
iiproject ran into serious difficulties the tcp/ip protocol was available as an alternative option for wider use in military networking. In April 1982Colonel Heidi Heiden, who had been appointed by the Assistant Secretary of Defense for Communications for Command and Control to report on the matter, recommended to a Department of Defense review board that the proposed Defense Data Network (ddn) should be based on the arpanet. tcp/ip became not only the bridge between arpanetand autodin ii but the standard across both arpanet and the new ddn. The military was adopting tcp/ip and stragglers would be cut off.
Two months later a dca newsletter circulated on the arpanet announcing the decision on the ddn and admonished the facilities that had not yet ordered the equipment required to make the switch from ncpto tcp/ip.5In March 1981 Major Haughney had already announced that the deadline for every node to cease using ncp was January 1983.6
After a number of setbacks the entire network adopted tcp/ip, the internetworking rather than networking protocol, in June 1983. The foundations for the Internet had been laid.
The next step
Much had changed in the two decades since arpa initiated the arpanet. In the beginning, when Lawrence Roberts had met with the principal investigators of the research centres funded by arpa in early 1967, few had been enthusiastic about networking. Yet arpa had forced the project through and access to the arpanet proved to be of enor- mous value to the facilities that had the privilege of being connected. Participation was open only to arpa contractors since the network was intended to support us government work rather than to compete with commercial offerings.7
While other networks existed, such as the proprietary decnet and the physics research network mfenet, the arpanetand its exclusivity created demand among America’s young computer science departments. As the eighties dawned and arpanet’s first decade of operation began to draw to a close, those without access jealously regarded their connected peers.
One of those who felt that demand most keenly was Lawrence Landweber, a professor at the University of Wisconsin. Landweber approached the National Science Foundation (nsf) with proposals for a computer science network. His second proposal, in which arpa
cooperated, was approved by the nsf in January 1981.8
nsfcommitted to support the project for five years after which the network would need to pay for itself. The new Computer Science Network (csnet) con- nected to arpanet using tcp/ip and was novel in a number of respects. First, the central role played by nsf marked the beginning of a transi- tion from military to civilian use of the Internet. Second, the network was open to all computer researchers in the United States and eventu- ally beyond. Annual fees ranged from $5,000 for university to $30,000 for industry facilities in contrast to the $90,000 price of a single imp (the machine that each facility connected to arpanet was required to install).9
csnetnot only prompted an explosive growth in the number of people with connections to the Internet but also proved that those without arpa support were willing to pay for the service. Moreover, csnetcontributed to an open approach to Internet billing across net- works. Robert Kahn at arpa made an agreement with Landweber that csnetsubscribers would not be charged for traffic that crossed the arpanet– an agreement, according to Landweber, that ‘broke new ground with respect to the principle of access and openness’.10
Centres with limited budgets could connect to csnet using a telephone dial-up system called Phonenet.
The same year that Landweber’s proposal for csnet was accepted, two academics from Yale and the City University of New York began to develop an alternative network for academics in the humanities. Using the remote spooling communications system (rscs) that the ibm machines in their respective universities operated, the two established a connection using a leased line and modems to pass messages back and forth between their mainframe computers in May 1981. They named the network ‘Because it’s there’ (bitnet), in reference to the ibm equipment that was ubiquitous in academic facilities. After some development bit- net could run on some dec and Unix machines. Over a ten-year period from 1982, it provided a cheap and efficient means for academic institutions around the world to connect to one another. By 1991, shortly before its popularity declined, bitnet connected 1,400 organ- izations in 49 countries, including the earn European research net- work.11
A new system called ‘Listserv’ performed functions similar to the message groups on the arpanet, circulating group e-mail discussions to a list of discussion participants. The bitnet operated similarly to Usenet using low-bandwidth leased telephone connections from one node to another.12
Corporation for Research and Educational Networking (cren). From 1993, bitnet began to decline as superior alternatives became more readily available.
Three years after its agreement to support csnet, in 1984, the National Science Foundation initiated a programme supporting supercomputing research at American universities. Research centres in the same region would be linked by regional networks, connected in turn to a high-speed national ‘backbone’ network. In addition, any uni- versity could connect to the new national high network, nsfnet. This was an important choice on the part of nsf. Both nasa and the Department of Energy would have preferred that nsfnet include only government research centres.13
The new nsfnet essentially continued the work of csnet in expanding connections among academic insti- tutions. It also used tcp/ip, which meant that the internetworking protocol would now be implemented on all connected machines, whether mighty supercomputers or humble pcs.14
In 1987 the Michigan Educated Research Information Triad (merit), a body with an acronym worthy of its intentions, was commissioned to build nsfnet. The arpanet would act as a temporary backbone for nsfnettraffic while work was in progress. The merit nsfnet backbone became operational on 24 July 1988. Its introduction enabled further growth of Internet connections. At the end of 1987 over 28,000 machines were connected to the Internet. Two years later almost 160,000 were online.15Traffic on the backbone doubled every seven months.16
Between 1989 and 1991 merit upgraded the backbone to handle far higher data speeds, from 1.5 Mbps to 45 Mbps. J.C.R. Lick- lider had lamented that the arpanet project in the early 1970s had been ‘forced to taper off without really attacking the problems of use by non- programmers’.17
This was now being corrected. The massive expansion of the community of users connected by nsfnet tested networking in a new way, connecting a plethora of new systems and users of varying degrees of expertise and experience. The internetworking protocol was put through its paces. Gradually merit administrators ironed out problems as they arose.
As nsfnet grew, another momentous event occurred. arpa’s man- date was to support new technologies and it was keen to unburden itself of a network that started operation in 1969 and was now ‘slow and expensive’.18
The arpanet was an order of magnitude slower than the nsfnet backbone. It was decommissioned on 28 February 1990. So
ended the military era of the Internet. In its place, a rapidly expanding and increasingly reliable civilian network was growing. Across the us and beyond, connections to nsfnet began to grow exponentially. In a single year the number of connected machines doubled. By October 1990almost a third of a million computers were connected. The next year the figure doubled again. Just short of 620,000 computers were online by October 1991. In October 1992 over one million computers were connected. By October 1993 more than two million computers were connected to the Internet.19
E-mail traffic was gradually sur- passed by new and more sophisticated information-finding services such as Gopher, Archie and Veronica. The stage was now set for a ‘killer ap’ to use on this new Internet, something that people other than specialist computer scientists could use, something revolutionary.
The end of the beginning
For a brief moment during the 2000 presidential election in the United States the history of the Internet became an issue of much debate. Al Gore, the Democratic Party’s candidate, came under attack because, it was reported, he had claimed to have invented the Internet. According to one estimate, more than 4,800 television, newspaper and magazine items made reference to the purported claim during the campaign.20
In the 1980s Gore had been among a cohort of so-called ‘Atari Democrats’, a group of Democratic Party politicians who believed that computing held the prospect of future prosperity. Gore raised the idea that the us, despite its early-mover advantage in networking, was losing this edge as other countries developed their own national net- works. In 1986 Gore, then in the Senate, sponsored a bill requiring the Office of Science and Technology Policy (ostp) to undertake a study of the challenges and opportunities for networking and computing research including supercomputers at university and federal research centres. In response, Gordon Bell, one of the senior figures at dec, chaired a committee on computing and digital infrastructure of the Federal Coordinating Council on Science, Engineering and Technology. The resulting report, released in November 1987, contained the main features that would appear in the 1991 High Performance Computing and Communication Act. These included the need to connect the net- working efforts of nasa, Department of Energy, nsf and arpa; and to establish a 1.5-Mbps network that would connect between two and three
hundred us research institutions, and in the longer term, to develop a 3-gigabyte (Gbps) network within fifteen years.21 The 1991 Act was informally known as the Gore Act, and it set the conditions for the next phase of the Internet.22
Nine years later, however, Gore’s work returned to haunt him. The controversy during the 2000 presidential election arose from an inter- view that he participated in with the television programme, cnn Late Edition, on 9 March 1999. Had any one of the writers of the almost 5,000 pieces on the incident consulted the transcript of Gore’s statement they would perhaps have learned that he had never, in fact, used the word ‘invent’ at all. The transcript reports that Gore said: ‘During my service in the United States Congress, I took the initiative in creating the Internet.’23
Vint Cerf and Robert Kahn, two individuals uniquely placed to comment on the matter, defended his record. Gore, they said, ‘deserves significant credit . . . No other elected official, to our knowledge, has made a greater contribution over a longer period of time’.24
A global Internet
In 1851 the first international submarine cable was laid across the Eng- lish Channel, and the first transatlantic transmission occurred seven years later.25
The first message transmitted along the Atlantic Ocean floor, from Queen Victoria to President James Buchanan, read, ‘England and America are united. Glory to God in the highest and on Earth, peace, goodwill toward men.’ Though brief, the transmission took thirty hours.26
(For reasons that Tom Standage relays with no small humour, the transatlantic cable that had relayed the message had failed within a month.)27
Far less ceremony attended the first international connection of the arpanet to the norsar seismic array near Oslo in 1973. The year before, at the public demonstration of the arpanet at the Washington Hilton Hotel in October 1972, a group of researchers had taken the first steps towards formal international cooperation on the Internet. They formed the International Network Working Group (inwg), initially comprising arpanet researchers, representatives from the National Physics Laboratory (npl) in the uk, and from the French Cyclades networking project. npl and arpa already had a history of cooperation on networking. Donald Davies at npl had invented packet-switched
networking at almost the same time as Paul Baran had at rand, and npl contributed ideas to the arpanet project from the beginning. It had been npl’s Roger Scantlebury, attending a conference at Gatlinburg in 1967, who introduced Lawrence Roberts to Paul Baran’s neglected work on packet networking. In addition, the team at Stanford working on the tcpprotocol from the spring of 1973 included a member of the French Cyclades network team. The new inwg agreed to operate by an infor- mal agreement of standards and the first draft of the tcp protocol developed at Cerf ’s laboratory at Stanford was reviewed by the group in September 1973. The inwg would go on to have regular meetings on both sides of the Atlantic and grow from an initial group of 25 parti - cipants to include several hundred participants on the mailing list by the 1990s.28
A number of other international cooperative bodies emerged. In 1981 the International Cooperation Board (icb) was established to coordi- nate work on satnet, arpa’s satellite network. The icb expanded into a forum for nato members to cooperate on networking research. In addition, International Academic Networkshop (ianw) meetings were held annually from 1982 to 1989 and brought together researchers from the us, Europe and eventually Asia-Pacific countries. At the ianw meeting at Princeton in November 1987 a new body was established called the Coordinating Committee for Intercontinental Research Net- works (ccirn). The ccirn involved Canada, the us, Europe, the Internet Advisory Board (iab), icb and, from 1992, representatives of the Asia - Pacific, Africa from 1995 and Latin America from 1997. ccirn produced a further offshoot called the Intercontinental Engineering Planning Group (iepg). A torrent of acro nyms though all this is, it demon- strates the gathering pace of international cooperation on networking.29
Since the late 1960s, private proprietary networks such as the sita airline reservation system had spanned the globe. Thereafter networks using various protocols including bitnet, tcp/ip, x.25 and dec’s ‘dec- net’ proprietary system proliferated. Many countries, including several European countries, Canada, Brazil, Japan and Australia were devel- oping internal networks, many of which used x.25. Much as arpanet had, these networks tended to serve specific academic and research purposes. Scientific collaborations required interconnectivity and international traffic. International connections evolved in the form of networks like the High Energy Physics Network (hepnet), which used dec’s decnet protocol; the European Research Network (earn), which
used bitnet; and the nasa Science Internet (nsi), which used decnet and tcp/ip.30
European networks also began to consolidate and inter- network under the coordination of various bodies (the Réseaux Associés pour la Recherche Européenne (rare) coordinated the development of x.25 networks and the Réseaux ip Européens (ripe)