The Advanced Research Projects Agency Network (ARPANET) was an early packet-switching network and the first network to implement the TCP/IP protocol suite. Both technologies became the technical foundation of the Internet. The ARPANET was initially founded by the Advanced Research Projects Agency (ARPA) of the United States Department of Defense.^{[1][2][3][4][5]}

The packet-switching methodology employed in the ARPANET was based on concepts and designed by Leonard Kleinrock, Paul Baran, Donald Davies, and Lawrence Roberts.^[6] The TCP/IP communications protocols were developed for the ARPANET by computer scientists Robert Kahn and Vint Cerf, and incorporated concepts from the French CYCLADES project directed by Louis Pouzin.

As the project progressed, protocols for internetworking were developed by which multiple separate networks could be joined into a network of networks. Access to the ARPANET was expanded in 1981, when the National Science Foundation (NSF) funded the Computer Science Network (CSNET). In 1982, the Internet protocol suite (TCP/IP) was introduced as the standard networking protocol on the ARPANET. In the early 1980s the NSF funded the establishment of national supercomputing centers at several universities and provided interconnectivity in 1986 with the NSFNET project, which also created network access to the supercomputer sites in the United States from research and education organizations. The ARPANET was decommissioned in 1990.

Historically, voice and data communications were based on methods of circuit switching, as exemplified in the traditional telephone network, wherein each telephone call is allocated a dedicated, end to end, electronic connection between the two communicating stations. The connection is established by switching systems that connected multiple intermediate call legs between these systems for the duration of the call.

The traditional model of the circuit-switched telecommunication network was challenged in the early 1960s by Paul Baran at the RAND Corporation, who had been researching systems that could sustain operation during partial destruction, such as by nuclear war. He developed the theoretical model of distributed adaptive message block switching.^[7] However, the telecommunication establishment rejected the development in favor of existing models. Donald Davies at the United Kingdom's National Physical Laboratory (NPL) independently arrived at a similar concept in 1965.^[8][9]

The earliest ideas for a computer network intended to allow general communications among computer users were formulated by computer scientist J. C. R. Licklider of Bolt, Beranek and Newman (BBN), in April 1963, in memoranda discussing the concept of the "Intergalactic Computer Network". Those ideas encompassed many of the features of the contemporary Internet. In October 1963, Licklider was appointed head of the Behavioral Sciences and Command and Control programs at the Defense Department's Advanced Research Projects Agency (ARPA). He convinced Ivan Sutherland and Bob Taylor that this network concept was very important and merited development, although Licklider left ARPA before any contracts were assigned for development.^[10]

Sutherland and Taylor continued their interest in creating the network, in part, to allow ARPA-sponsored researchers at various corporate and academic locales to utilize computers provided by ARPA, and, in part, to quickly distribute new software and other computer science results.^[11] Taylor had three computer terminals in his office, each connected to separate computers, which ARPA was funding: one for the System Development Corporation (SDC) Q-32 in Santa Monica, one for Project Genie at the University of California, Berkeley, and another for Multics at the Massachusetts Institute of Technology. Taylor recalls the circumstance: "For each of these three terminals, I had three different sets of user commands. So, if I was talking online with someone at S.D.C., and I wanted to talk to someone I knew at Berkeley, or M.I.T., about this, I had to get up from the S.D.C. terminal, go over and log into the other terminal and get in touch with them. I said, "Oh Man!", it's obvious what to do: If you have these three terminals, there ought to be one terminal that goes anywhere you want to go. That idea is the ARPANET".^[12]

Donald Davis' work caught the attention of ARPANET developers at a conference in Gatlinburg, Tennessee, in October 1967.^[13] He gave the first public demonstration, having coined the term packet switching, on 5 August 1968 and incorporated it into the NPL network in England.^[14] Elizabeth Feinler created the first Resource Handbook for ARPANET in 1969 which led to the development of the ARPANET directory.^[15] The directory, built by Feinler and a team made it possible to navigate the ARPANET.^[16][17] The NPL network followed by the ARPANET were the first two networks in the world to use packet switching,^[18][19] and were themselves connected together in 1973.^[20][21]

In February 1966, Bob Taylor successfully lobbied ARPA's Director Charles M. Herzfeld to fund a network project. Herzfeld redirected funds in the amount of one million dollars from a ballistic missile defense program to Taylor's budget.^[22] Taylor hired Larry Roberts as a program manager in the ARPA Information Processing Techniques Office in January 1967 to work on the ARPANET. In April 1967, Roberts held a design session on technical standards. The initial standards for identification and authentication of users, transmission of characters, and error checking and retransmission procedures were discussed. At the meeting, Wesley Clark proposed minicomputers called Interface Message Processors (IMPs) should be used to interface to the network rather than the large mainframes that would be the nodes of the ARPANET. Roberts modified the ARPANET plan to incorporate Clark's suggestion. The plan was presented at the ACM Symposium in Gatlinburg, Tennessee, in October 1967.^{[23][24][25][26]} Donald Davies' work on packet switching and the NPL network, presented by a colleague (Roger Scantlebury), came to the attention of ARPANET developers at this conference.^[27][28] Larry Roberts at ARPA applied Davies' concept of packet switching for the ARPANET,^[29][30] and sought input from Paul Baran and Leonard Kleinrock. Building on his earlier work on queueing theory, Kleinrock modelled the performance of packet-switched networks, which underpinned the development of the ARPANET.^[27] The NPL network was using line speeds of 768 kbit/s, and the proposed line speed for the ARPANET was upgraded from 2.4 kbit/s to 50 kbit/s.^[31]

By mid-1968, Roberts had prepared a complete plan for the computer network and gave a report to Taylor on June 3, who approved it on June 21. After approval by ARPA, a Request for Quotation (RFQ) was issued for 140 potential bidders. Most computer science companies regarded the ARPA proposal as outlandish, and only twelve submitted bids to build a network; of the twelve, ARPA regarded only four as top-rank contractors. At year's end, ARPA considered only two contractors, and awarded the contract to build the network to Bolt, Beranek and Newman Inc. (BBN) on 7 April 1969. The initial, seven-person BBN team were much aided by the technical specificity of their response to the ARPA RFQ, and thus quickly produced the first working system. This team was led by Frank Heart and included Robert Kahn. The BBN-proposed network closely followed Roberts' ARPA plan: a network composed of small computers called Interface Message Processors (or IMPs), similar to the later concept of routers, that functioned as gateways interconnecting local resources. At each site, the IMPs performed store-and-forward packet switching functions, and were interconnected with leased lines via telecommunication data sets (modems), with initial data rates of 56kbit/s. The host computers were connected to the IMPs via custom serial communication interfaces. The system, including the hardware and the packet switching software, was designed and installed in nine months.^[24][32][26] The BBN team continued to interact with the NPL team.^[33][34]

The first-generation IMPs were built by BBN Technologies using a rugged computer version of the Honeywell DDP-516 computer, configured with 24KB of expandable magnetic-core memory, and a 16-channel Direct Multiplex Control (DMC) direct memory access unit.^[35] The DMC established custom interfaces with each of the host computers and modems. In addition to the front-panel lamps, the DDP-516 computer also features a special set of 24 indicator lamps showing the status of the IMP communication channels. Each IMP could support up to four local hosts, and could communicate with up to six remote IMPs via early Digital Signal 0 leased telephone lines. The network connected one computer in Utah with three in California. Later, the Department of Defense allowed the universities to join the network for sharing hardware and software resources.

According to Stephen J. Lukasik, who as Deputy Director and Director of DARPA (1967–1974) was "the person who signed most of the checks for Arpanet's development":

The ARPANET incorporated distributed computation, and frequent re-computation, of routing tables. This increased the survivability of the network in the face of significant interruption. Automatic routing was technically challenging at the time. The ARPANET was designed to survive subordinate-network losses, since the principal reason was that the switching nodes and network links were unreliable, even without any nuclear attacks. Resource scarcity supported the creation of the ARPANET, according to Charles Herzfeld, ARPA Director (1965–1967):

The ARPANET was operated by the military during the two decades of its existence, until 1990.^[38][39]

The Internet Society agrees with Herzfeld in a footnote in their online article, A Brief History of the Internet:

Paul Baran, the first to build a theoretical model for communication using packet switching, conducted the RAND study referenced above.^[7] Baran confirmed that though the ARPANET did not exactly share his project's goal, his work had greatly contributed to the development of the ARPANET.^[41] Minutes taken by Elmer Shapiro of Stanford Research Institute at the ARPANET design meeting of 9–10 October 1967 indicate that a version of Baran's routing method and suggestion of using a fixed packet size was expected to be employed.^[42]

The initial ARPANET consisted of four IMPs:^[43]

The first successful message on the ARPANET was sent by UCLA student programmer Charley Kline, at 10:30 pm PST on 29 October 1969 (6:30 UTC on 30 October 1969), from Boelter Hall 3420.^[44] Kline transmitted from the university's SDS Sigma 7 Host computer to the Stanford Research Institute's SDS 940 Host computer. The message text was the word login; on an earlier attempt the l and the o letters were transmitted, but the system then crashed. Hence, the literal first message over the ARPANET was lo. About an hour later, after the programmers repaired the code that caused the crash, the SDS Sigma 7 computer effected a full login. The first permanent ARPANET link was established on 21 November 1969, between the IMP at UCLA and the IMP at the Stanford Research Institute. By 5 December 1969, the entire four-node network was established.^[45]

In March 1970, the ARPANET reached the East Coast of the United States, when an IMP at BBN in Cambridge, Massachusetts was connected to the network. Thereafter, the ARPANET grew: 9 IMPs by June 1970 and 13 IMPs by December 1970, then 18 by September 1971 (when the network included 23 university and government hosts); 29 IMPs by August 1972, and 40 by September 1973. By June 1974, there were 46 IMPs, and in July 1975, the network numbered 57 IMPs. By 1981, the number was 213 host computers, with another host connecting approximately every twenty days.^[43]

In 1973, a transatlantic satellite link connected the Norwegian Seismic Array (NORSAR) to the ARPANET, making Norway the first country outside the US to be connected to the network. At about the same time a terrestrial circuit added a London IMP.^[46] This connectivity later evolved into the SATNET.

In 1975, the ARPANET was declared "operational". The Defense Communications Agency took control since ARPA was intended to fund advanced research.^[43] At about this time, the first ARPANET encryption devices were deployed to support classified traffic.

In September 1984 work was completed on restructuring the ARPANET giving U.S. military sites their own Military Network (MILNET) for unclassified defense department communications.^[47][48] Controlled gateways connected the two networks. The combination was called the Defense Data Network (DDN).^[49] Separating the civil and military networks reduced the 113-node ARPANET by 68 nodes. The MILNET later became the NIPRNet.

Because of its government funding, certain forms of traffic were discouraged or prohibited. A 1982 handbook on computing at MIT's AI Lab stated regarding network etiquette:^[50]

Support for inter-IMP circuits of up to 230.4 kbit/s was added in 1970, although considerations of cost and IMP processing power meant this capability was not actively used.

1971 saw the start of the use of the non-ruggedized (and therefore significantly lighter) Honeywell 316 as an IMP. It could also be configured as a Terminal Interface Processor (TIP), which provided terminal server support for up to 63 ASCII serial terminals through a multi-line controller in place of one of the hosts.^[51] The 316 featured a greater degree of integration than the 516, which made it less expensive and easier to maintain. The 316 was configured with 40 kB of core memory for a TIP. The size of core memory was later increased, to 32 kB for the IMPs, and 56 kB for TIPs, in 1973.

In 1975, BBN introduced IMP software running on the Pluribus multi-processor. These appeared in a few sites. In 1981, BBN introduced IMP software running on its own C/30 processor product.

The original IMPs and TIPs were phased out as the ARPANET was shut down after the introduction of the NSFNet, but some IMPs remained in service as late as July 1990.^[52][53]

The ARPANET Completion Report, jointly published by BBN and ARPA, concludes that:

In the wake of the decommissioning of the ARPANET on 28 February 1990, Vinton Cerf wrote the following lamentation, entitled "Requiem of the ARPANET":^[55]

Senator Albert Gore, Jr. authored the High Performance Computing and Communication Act of 1991, commonly referred to as "The Gore Bill", after hearing the 1988 concept for a National Research Network submitted to Congress by a group chaired by Leonard Kleinrock. The bill was passed on 9 December 1991 and led to the National Information Infrastructure (NII) which Al Gore called the information superhighway.

The ARPANET project was honored with two IEEE Milestones, both dedicated in 2009.^[56][57]

The starting point for host-to-host communication on the ARPANET in 1969 was the 1822 protocol, which defined the transmission of messages to an IMP.^[58] The message format was designed to work unambiguously with a broad range of computer architectures. An 1822 message essentially consisted of a message type, a numeric host address, and a data field. To send a data message to another host, the transmitting host formatted a data message containing the destination host's address and the data message being sent, and then transmitted the message through the 1822 hardware interface. The IMP then delivered the message to its destination address, either by delivering it to a locally connected host, or by delivering it to another IMP. When the message was ultimately delivered to the destination host, the receiving IMP would transmit a Ready for Next Message (RFNM) acknowledgement to the sending, host IMP.

Unlike modern Internet datagrams, the ARPANET was designed to reliably transmit 1822 messages, and to inform the host computer when it loses a message; the contemporary IP is unreliable, whereas the TCP is reliable. Nonetheless, the 1822 protocol proved inadequate for handling multiple connections among different applications residing in a host computer. This problem was addressed with the Network Control Program (NCP), which provided a standard method to establish reliable, flow-controlled, bidirectional communications links among different processes in different host computers. The NCP interface allowed application software to connect across the ARPANET by implementing higher-level communication protocols, an early example of the protocol layering concept incorporated to the OSI model.^[59]

Networking research in the early 1970s by Robert E. Kahn and Vint Cerf let to the formulation of the Transmission Control Program, and its specification in December 1974 in RFC 675 ^[111] . Still a monolithic design at this time, the software was redesigned as a modular protocol stack. Originally referred to as IP/TCP, it was installed in the ARPANET for production use in January 1983. The development of a complete protocol suite by 1989, as outlined in RFC 1122 ^[112] and RFC 1123 ^[113] , laid the foundation for growth of TCP/IP as a comprehensive protocol suite as the core component of the emerging Internet.^[60]

NCP provided a standard set of network services that could be shared by several applications running on a single host computer. This led to the evolution of application protocols that operated, more or less, independently of the underlying network service, and permitted independent advances in the underlying protocols.

In 1971, Ray Tomlinson, of BBN sent the first network e-mail (RFC 524 ^[114] , RFC 561 ^[115] ).^[61] By 1973, e-mail constituted 75% of the ARPANET traffic.

By 1973, the File Transfer Protocol (FTP) specification had been defined (RFC 354 ^[116] ) and implemented, enabling file transfers over the ARPANET.

The Network Voice Protocol (NVP) specifications were defined in 1977 (RFC 741 ^[117] ), and implemented. But, because of technical shortcomings, conference calls over the ARPANET never worked well; the contemporary Voice over Internet Protocol (packet voice) was decades away.

The Purdy Polynomial hash algorithm was developed for the ARPANET to protect passwords in 1971 at the request of Larry Roberts, head of ARPA at that time. It computed a polynomial of degree 224 + 17 modulo the 64-bit prime p = 264 − 59. The algorithm was later used by Digital Equipment Corporation (DEC) to hash passwords in the VMS operating system and is still being used for this purpose.

The ARPANET was related to many other research projects, which either influenced the ARPANET design, or which were ancillary projects or spun out of the ARPANET.