The Advanced Research Projects Agency Network (ARPANET) was the first large network that sent data in small pieces and used many computers to control it. It was also one of the first computer networks to use the TCP/IP protocol suite. These technologies became the base for the Internet. ARPANET was created by the Advanced Research Projects Agency (now called DARPA), which is part of the United States Department of Defense.

The idea for ARPANET came from J. C. R. Licklider. Bob Taylor started the project in 1966 to allow computers far apart to share resources. Taylor chose Larry Roberts to manage the project. Roberts made important decisions about asking companies to build the network. He used Donald Davies' ideas about sending data in small pieces and asked Paul Baran for help. In 1969, the U.S. government gave a contract to Bolt Beranek & Newman (BBN) to build the Interface Message Processors (IMPs) for the network. Bob Kahn led the design and created the first protocol for the network. Roberts worked with Leonard Kleinrock at UCLA to develop math methods for studying the packet network.

The first computers were connected in 1969. In 1970, the Network Control Program was created, led by Steve Crocker at UCLA and other students, including Jon Postel. The network became fully operational in 1971. Later software allowed people to log in from far away and send files, which helped create early email. The network grew quickly, and control was handed to the Defense Communications Agency in 1975.

Bob Kahn moved to DARPA, and with Vint Cerf at Stanford University, they created the Transmission Control Program for connecting different networks. This work led to a way to join separate networks into one big network. It used ideas from the French CYCLADES project, led by Louis Pouzin, and Bob Metcalfe's work at Xerox Parc. In 1983, version 4 of TCP/IP was used on ARPANET after the Department of Defense made it the standard for military networks.

In 1981, the National Science Foundation (NSF) helped create the Computer Science Network (CSNET). In the early 1980s, the NSF funded supercomputing centers at universities and started the NSFNET project in 1986 to connect networks. ARPANET was officially shut down in 1990 after partnerships with companies ensured the private sector could expand and commercialize the Internet.

Inspiration

Historically, voice and data communications used a method called circuit switching, as seen in traditional telephone networks. In this system, each phone call is given a dedicated electronic path between the two people talking. This path is created by switching systems that link several parts of the call together for the entire conversation.

The traditional circuit-switched network faced challenges in the early 1960s when Paul Baran, working at the RAND Corporation, studied systems that could keep functioning even if parts of them were damaged, such as during a nuclear war. He proposed a theoretical model called distributed adaptive message block switching. However, the telecommunication industry preferred existing models instead of Baran’s idea. Around the same time, Donald Davies at the United Kingdom’s National Physical Laboratory (NPL) independently developed a similar concept in 1965.

In April 1963, computer scientist J. C. R. Licklider, working at Bolt Beranek and Newman (BBN), wrote documents describing the idea of a "Intergalactic Computer Network." These ideas included many features found in today’s Internet. In October 1963, Licklider became head of the Behavioral Sciences and Command and Control programs at the Defense Department’s Advanced Research Projects Agency (ARPA). He encouraged Ivan Sutherland and Bob Taylor to support the network concept, though Licklider left ARPA before any development contracts were made.

Sutherland and Taylor continued working on the network to help ARPA-funded researchers at different locations share computer resources and quickly share new software and research results. Taylor had three computer terminals in his office, each connected to a different computer funded by ARPA: one at the System Development Corporation (SDC) in Santa Monica, one at the University of California, Berkeley, and one at the Massachusetts Institute of Technology (MIT). He described the problem: "Each terminal had different commands, so I had to move between terminals to talk to people at different locations. I realized we needed one terminal that could connect to any computer. That idea became ARPANET."

Donald Davies’ work was noticed by ARPANET developers during a meeting in October 1967. He introduced the term "packet switching" in August 1968 and used it in the NPL network in England. The NPL network and ARPANET were the first two networks in the world to use packet switching. Roberts noted that networks built in the 1970s closely resembled Davies’ 1965 design.

Creation

In February 1966, Bob Taylor asked ARPA's Director Charles M. Herzfeld to support a network project. Herzfeld moved one million dollars from a missile defense program to Taylor's budget. In January 1967, Taylor hired Larry Roberts as a program manager in the ARPA Information Processing Techniques Office to work on the ARPANET. Roberts met Paul Baran in February 1967 but did not talk about networks.

Roberts asked Frank Westervelt to study message size and content for the network and to write a report about how computers could communicate, including rules for sending characters, checking for errors, and identifying users. In April 1967, ARPA held a meeting to discuss technical standards. Topics included how to identify users, send characters, and handle errors. Roberts suggested that all mainframe computers should connect directly. However, other researchers were hesitant to use these computers for network tasks. After the meeting, Wesley Clark proposed using smaller computers as a link to create a message-switching network. Roberts changed the ARPANET plan to include Clark’s idea and called these smaller computers Interface Message Processors (IMPs).

The plan was shared at the first Symposium on Operating Systems Principles in October 1967. At this meeting, Donald Davies’ work on packet switching and the NPL network, presented by Roger Scantlebury, and Paul Baran’s ideas were introduced to ARPA researchers. Roberts used Davies’ packet-switching concept for ARPANET and asked Baran about dynamic routing. The NPL network used line speeds of 768 kbit/s, and ARPANET’s speed was increased from 2.4 kbit/s to 50 kbit/s.

By mid-1968, Roberts and Barry Wessler completed the final IMP specification based on a report from the Stanford Research Institute (SRI) that ARPA had asked to write. Roberts presented the plan to Taylor on June 3, and Taylor approved it on June 21. After approval, ARPA issued a Request for Quotation (RFQ) to 140 companies. Most computer companies thought the proposal was unusual, and only twelve submitted bids. ARPA considered only four as top contractors. By the end of the year, two companies were chosen, and the contract was given to BBN in January 1969.

The BBN team, which had seven members, was helped by the detailed specifications in their response to the RFQ. They quickly built the first working system. The team was led by Frank Heart for the IMPs and Bob Kahn for the network design. Other members included Dave Walden, Severo Ornstein, William Crowther, and others. The BBN network followed Roberts’ plan: small computers (IMPs) connected local resources as gateways. The BBN team developed routing, flow control, software, and network control. At each site, IMPs stored and forwarded packets through leased lines using modems, with initial speeds of 50 kbit/s. Host computers connected to IMPs through custom serial interfaces. The system, including hardware and software, was completed in nine months. The BBN team met with the NPL team in the U.S. and the U.K.

Like the NPL network, the first IMPs used a rugged version of the Honeywell DDP-516 computer with 24 KB of memory and a 16-channel Direct Multiplex Control (DMC) unit. The DMC connected to host computers and modems. The DDP-516 had 24 indicator lamps showing the status of communication channels. Each IMP could connect to four local hosts and six remote IMPs through early Digital Signal 0 lines. The network first linked one computer in Utah with three in California. Later, the Department of Defense allowed universities to join the network to share hardware and software resources.

Implementation

The first four nodes were used as a test area for creating and fixing the 1822 protocol, which was a large project. These nodes were connected electronically in 1969, but network applications were not possible until the Network Control Program was added in 1970. This program allowed the first two host-to-host protocols, remote login (Telnet) and file transfer (FTP), which were developed and used from 1969 to 1973. The network was officially declared operational in 1971. Network traffic increased after email was set up at most sites around 1973.

The initial ARPANET setup connected UCLA, ARC, UCSB, and the University of Utah School of Computing. The first node was built at UCLA, where Leonard Kleinrock tested network performance and studied his ideas about message delays. Locations were chosen to lower costs and because each site had special skills helpful for the project:

• University of California, Los Angeles (UCLA), where Kleinrock had a Network Measurement Center (NMC) with an SDS Sigma 7, the first computer connected to the network;
• The Augmentation Research Center at Stanford Research Institute (now SRI International), where Douglas Engelbart developed the NLS system, an early hypertext system, and ran the Network Information Center (NIC) using an SDS 940 computer named "Genie," the first host attached;
• University of California, Santa Barbara (UCSB), with the Culler-Fried Interactive Mathematics Center's IBM 360/75 running OS/MVT;
• A sketch of the ARPANET in December 1969 showed nodes at UCLA and SRI. The University of Utah School of Computing, where Ivan Sutherland worked, used a DEC PDP-10 running TENEX.

The first successful host-to-host connection on the ARPANET was between Stanford Research Institute (SRI) and UCLA, made by SRI programmer Bill Duvall and UCLA student Charley Kline at 10:30 pm PST on October 29, 1969 (6:30 UTC on October 30, 1969). Kline connected from UCLA's SDS Sigma 7 computer to SRI's SDS 940 computer. He typed "login," but the SDS 940 crashed after two characters. After Duvall adjusted settings, Kline tried again and logged in successfully. The first two characters sent over the ARPANET were "lo." The first permanent ARPANET link was made on November 21, 1969, between UCLA and SRI. By December 5, 1969, the initial four-node network was complete.

Elizabeth Feinler created the first ARPANET Resource Handbook in 1969, leading to the development of an ARPANET directory. This directory, built by Feinler and her team, helped users navigate the network.

In 1968, Roberts hired Kleinrock to measure network performance and find ways to improve it. Using his earlier work on queueing theory and optimizing message delays in communication networks, Kleinrock created mathematical models that helped design the ARPANET as it expanded in the 1970s. Since analytic methods had limits, ARPA and the UK’s NPL also used computer simulations.

Roberts worked with Howard Frank to design the network’s structure. Frank suggested ways to increase speed and lower costs in a larger network. By March 1970, an IMP at BBN in Cambridge, Massachusetts was connected to the network, reaching the East Coast. The ARPANET grew rapidly: 9 IMPs by June 1970, 13 by December 1970, 18 by September 1971 (with 23 university and government hosts), 29 by August 1972, and 40 by September 1973. By June 1974, there were 46 IMPs, and by July 1975, 57 IMPs. By 1981, there were 213 host computers, with a new host added every 20 days.

Support for inter-IMP circuits up to 230.4 kbit/s was added in 1970, though cost and IMP processing power limited its use.

Larry Roberts saw the ARPANET and NPL projects as helpful together and tried to connect them via satellite in 1970. In 1971, Peter Kirstein’s group at University College London (UCL) was chosen instead of NPL for the UK connection. In June 1973, a transatlantic satellite link connected ARPANET to the Norwegian Seismic Array (NORSAR) via Sweden and then to UCL. UCL acted as a gateway for ARPANET to connect with British academic networks, creating the first international resource-sharing network and conducting early research on internetworking.

In 1971, the non-ruggedized (lighter) Honeywell 316 was used as an IMP. It could also act as a Terminal Interface Processor (TIP), supporting up to 63 ASCII serial terminals. The 316 was more integrated than the 516, making it cheaper and easier to maintain. TIPs had 40 kB of core memory, later increased to 32 kB for IMPs and 56 kB for TIPs in 1973.

The ARPANET was shown at the International Conference on Computer Communications (ICCC) in October 1972.

In 1975, BBN introduced IMP software for the Pluribus multi-processor, used in a few locations. In 1981, BBN launched IMP software for its own C/30 processor.

Networking evolution

The ARPANET was not designed to support IMPs from companies other than BBN, so the IMP-to-IMP protocol and message format were not standardized. However, IMPs still communicated with each other to perform tasks such as link-state routing, reliable message forwarding, and remote monitoring and management for ARPANET’s Network Control Center. At first, each IMP had a 6-bit identifier and could support up to 4 hosts, which were identified with a 2-bit index. An ARPANET host address included both the port index on its IMP and the IMP’s identifier, written as either port/IMP notation or a single byte. For example, the address of MIT-DMG (known for hosting the development of Zork) could be written as 1/6 or 70. In early 1976, the host and IMP numbering systems were upgraded to 8-bit and 16-bit, respectively.

In addition to routing and forwarding messages, IMPs ran background programs such as TTY, DEBUG, PARAMETER-CHANGE, DISCARD, TRACE, and STATISTICS. These programs were assigned host numbers to allow direct communication and performed tasks independently of connected hosts. For example, the TTY program enabled on-site operators to manually send ARPANET packets through a teletype connected directly to an IMP.

In 1969, the starting point for host-to-host communication on ARPANET was the 1822 protocol, which defined how messages were sent to an IMP. The message format was designed to work with many different computer systems. An 1822 message included a message type, a numeric host address, and a data field. To send a data message, a host formatted the message with the destination address and data, then transmitted it through the 1822 hardware interface. The IMP delivered the message to the destination host directly or to another IMP. When the message reached its destination, the receiving IMP sent a Ready for Next Message (RFNM) acknowledgment to the sending host’s IMP.

Unlike modern Internet datagrams, ARPANET was designed to reliably transmit 1822 messages and notify hosts when messages were lost. Modern IP is unreliable, while TCP is reliable. However, the 1822 protocol could not handle multiple connections for different applications on a single host. This issue was solved with the Network Control Program (NCP), which provided a standard method for establishing reliable, flow-controlled, bidirectional communication between processes on different host computers. The NCP interface allowed application software to connect across ARPANET by using higher-level communication protocols, an early example of protocol layering later used in the OSI model.

NCP was developed under the leadership of Steve Crocker, a graduate student at UCLA. Crocker led the Network Working Group (NWG), a group of graduate students from universities and research labs, including Jon Postel at UCLA. They were supported by ARPA to develop ARPANET and the software for host computers that supported applications.

Starting in 1972, Louis Pouzin and Hubert Zimmermann at IRIA in France introduced a simplified end-to-end networking approach called a catenet, discussed at the ICCC. Bob Kahn left BBN in 1972, briefly worked as President of Telenet, then joined DARPA as a program manager for ARPANET under Larry Roberts and later became director of the IPTO when Roberts left. Kahn worked on satellite and ground-based radio packet networks and recognized the importance of communication across both. Steve Crocker, now at DARPA, and leaders of British and French network projects founded the International Network Working Group (INWG). On Crocker’s recommendation, Vint Cerf, then a faculty member at Stanford University, became INWG’s chair. Bob Metcalfe developed the theory and practice behind Ethernet and the PARC Universal Packet. These groups explored how to connect packet-switching networks with different specifications, known as internetworking. Peter Kirstein implemented internetworking at University College London in 1973.

Research by Kahn and Cerf led to the creation of the Transmission Control Program in 1974. The specification was written by Cerf, Yogen Dalal, and Carl Sunshine at Stanford in December 1974 (RFC 675), and researchers began publishing Internet Experiment Notes (IENs). The following year, testing began with simultaneous implementations at Stanford, BBN, and University College London. Initially a single, unified design, the software was redesigned as a modular protocol stack in Version 4 in 1978, influenced by the French CYCLADES project and Bob Metcalfe’s ideas at Xerox PARC. This version was installed in ARPANET for production use in January 1983, replacing NCP. By 1989, the complete Internet protocol suite, outlined in RFC 1122 and RFC 1123, and partnerships with the telecommunications and computer industries laid the foundation for TCP/IP becoming the core protocol of the emerging Internet.

Operation

ARPA was created to support research into advanced technologies. The ARPANET was a research project focused on communication systems rather than user needs. In the summer of 1975, the Defense Communications Agency took control of ARPANET's operations. Around the same time, the first encryption devices for ARPANET were installed to protect classified information. A report titled The ARPANET Completion Report, written by BBN and DARPA in 1978 and published in 1981, stated:

In 1981, the National Science Foundation (NSF) funded the Computer Science Network (CSNET), which expanded access to ARPANET.

Connections between ARPANET and Norway (through NORSAR and NDRE) and University College London (UCL) later became part of the SATNET. ARPANET, SATNET, and PRNET were connected in 1977. In 1980, the Department of Defense (DoD) made TCP/IP the standard protocol for all military computer networks. Norway and UCL left ARPANET in 1982 and began using TCP/IP over SATNET. On January 1, 1983, known as "flag day," TCP/IP replaced the older Network Control Protocol as the standard for ARPANET.

In September 1984, ARPANET was restructured to create a separate network called MILNET for unclassified military communications. Both ARPANET and MILNET carried unclassified information and were connected through limited gateways that could be used to separate the networks during emergencies. MILNET was part of the Defense Data Network (DDN). Separating the networks reduced ARPANET from 113 nodes to 45 nodes. After MILNET was created, ARPANET continued to serve as an Internet backbone for researchers but was gradually phased out.

Applications

NCP offered a standard set of network services that multiple applications on a single computer could use. This development led to the creation of application protocols that functioned independently of the basic network service, allowing improvements in network protocols without affecting the applications.

Early application protocols, such as TELNET for remote access and the File Transfer Protocol (FTP), were created and later adapted to work with the TCP/IP protocol suite. In the 1980s, FTP was replaced by the Simple Mail Transfer Protocol for email, followed by POP and IMAP.

TELNET was first developed in 1969, starting with RFC 15 and later expanded in RFC 855.

The original File Transfer Protocol (FTP) was written by Abhay Bhushan and published as RFC 114 on April 16, 1971. By 1973, the FTP specification (RFC 354) was finalized and used to transfer files over the ARPANET.

In 1971, Ray Tomlinson of BBN sent the first network email using RFC 524 and RFC 561. A study by ARPA in 1973, one year after email was introduced to the ARPANET community, found that three-quarters of ARPANET traffic was email. Email remained a major part of ARPANET activity.

The Network Voice Protocol (NVP) was defined in 1977 (RFC 741) and implemented. However, due to technical problems, conference calls over ARPANET did not work well. Modern Voice over Internet Protocol (packet voice) technology was not available for many years.

Security

The Purdy Polynomial hash algorithm was created for the ARPANET in 1971 at the request of Larry Roberts, who was the head of ARPA at that time. It used a special math formula called a polynomial of degree 2 + 17, calculated using a method called modulo with a 64-bit prime number, which is 2^64 − 59. Later, the algorithm was used by Digital Equipment Corporation (DEC) to hash passwords in the VMS operating system, and it is still used today for this purpose.

Rules and etiquette

Because the network was funded by the government, some types of communication were not allowed or stopped.

Leonard Kleinrock is said to have done the first action that broke the rules on the Internet. In 1973, he sent a message asking for his electric razor back after a meeting in England. At that time, using the ARPANET for personal purposes was against the law.

In 1978, Gary Thuerk from Digital Equipment Corporation (DEC) sent the first large email message to about 400 possible customers through the ARPANET. He says this action led to $13 million in sales for DEC products and showed how useful email marketing could be.

A 1982 guide about computing at MIT's AI Lab included rules for proper behavior on the network:

Decommissioning

In 1985, the NSF helped create national supercomputing centers at several universities. In 1986, the NSFNET project connected these centers through a network, allowing them to share information. This network became the main part of the Internet used by government agencies and universities.

The ARPANET project was officially shut down in 1990. The original IMPs and TIPs, which were used to connect computers on ARPANET, were gradually removed after NSFNET was introduced. However, some IMPs continued to work until July 1990.

After ARPANET was shut down on February 28, 1990, Vinton Cerf wrote a sad piece of writing called "Requiem of the ARPANET."

Legacy

The ARPANET helped create modern computer networks, including the Internet. Ideas like packet switching, decentralized networks, and communication protocols were developed and used on ARPANET. These ideas formed the basis for a worldwide network that changed how people communicate, share information, and work together on research.

ARPANET was connected to many other research projects. Some of these projects influenced ARPANET’s design, were related to ARPANET, or came from ARPANET.

Senator Al Gore wrote the High Performance Computing and Communication Act of 1991, often called "The Gore Bill," after hearing about a 1988 idea for a National Research Network. This idea was presented to Congress by a group led by Leonard Kleinrock. The bill passed on December 9, 1991, and led to the National Information Infrastructure (NII), which Gore called the "information superhighway."

The ARPANET project was recognized with two IEEE Milestones, both added in 2009.

On May 17, 2011, Arlington County, Virginia, placed two historical markers in its Rosslyn neighborhood at 1400 Wilson Boulevard. These markers honored the origins of the Internet, as the area was ARPANET’s home from 1970 to 1975.

Charles Herzfeld, who led ARPA from 1965 to 1967, said ARPANET used distributed computation and updated routing tables often. These features helped the network stay functional during major disruptions. ARPANET was also designed to keep working even if parts of it failed. The Internet Society agrees with Herzfeld, as noted in a footnote on their website, A Brief History of the Internet.

Paul Baran created the first theoretical model for sending data using packet switching. Though ARPANET’s goals were different from Baran’s project, he said his work helped shape ARPANET. Notes from a 1967 ARPANET design meeting show that a version of Baran’s routing method, called "hot potato," might have been used. This matches a proposal by the NPL team at the Symposium on Operating System Principles in Gatlinburg.

In the 1970s, ARPA focused more on "command and control." Stephen J. Lukasik, who was deputy director of ARPA from 1967 to 1970 and later director of DARPA from 1970 to 1975, said this was a key goal during that time.