ARPANET
Packet Switching and the Network That Became the Internet
- 16 chapters
- 28m
- Networking
- Free · no sign-up
The book traces how ARPANET evolved from its early days through network performance improvements and the development of protocols like 1822 and the Network Control Program. TCP/IP eventually replaced NCP, enabling the modern internet. The transition included the arrival of email as an unplanned but crucial application, along with security considerations and rules for network etiquette.
ARPANET's story shows how military research led to today's global communications system. The book covers everything from technical implementation to the debate about original design goals. Anyone interested in how the internet began will find this detailed account worth their time.
Listen
-
Read this chapter
Voice and data communications used to work like the old telephone network, where a dedicated connection was set up for the whole length of a call. When you made a phone call, switching systems would link multiple parts of the line together for as long as the conversation lasted. That method is called circuit switching, and it was how things worked before packet switching came along to change everything.
In the early 1960s, Paul Baran at the RAND Corporation began questioning how telecommunication networks might survive partial destruction, like from nuclear war. He developed a model for a distributed adaptive message block switching system. His ideas were rejected by the established telecommunication leaders who preferred existing methods. Meanwhile, in 1965, Donald Davies at the United Kingdom’s National Physical Laboratory reached a similar conclusion independently.
In April 1963, computer scientist J. C. R. Licklider of Bolt Beranek and Newman wrote memoranda about an “Intergalactic Computer Network” that would let users communicate across computers. Those early ideas included many features found in today’s Internet. By October of that year, Licklider had become head of the Behavioral Sciences and Command and Control programs at ARPA, the Defense Department’s Advanced Research Projects Agency. There, he made his case to Ivan Sutherland and Bob Taylor that this network concept was important enough to develop. Licklider left ARPA before any contracts were assigned for the work.
Sutherland and Taylor advanced their network concept to let ARPA-funded researchers share computers and software. Taylor operated three terminals in his office connected to different machines: one for the System Development Corporation's Q-32 in Santa Monica, one for Project Genie at UC Berkeley, and another for Multics at MIT. He recalled switching between these terminals was frustrating, requiring him to get up from one terminal to log into another to communicate with colleagues elsewhere. "Oh Man!" he said, realizing the solution: "If you have these three terminals, there ought to be one terminal that goes anywhere you want to go." That concept became ARPANET.
Donald Davies' work drew attention from ARPANET developers at a symposium in October 1967. He gave the first public talk on packet switching in August 1968, introducing the concept while building the NPL network in England. That network, followed by ARPANET, became the world’s first two systems to use packet switching. Later, Roberts said the computer networks of the 1970s resembled Davies’ original 1965 design “in nearly all respects.”
-
Read this chapter
In February 1966, Bob Taylor convinced ARPA's Director Charles M. Herzfeld to fund a network project by redirecting one million dollars from a ballistic missile defense program to his budget. The following January, Taylor appointed Larry Roberts as program manager in the ARPA Information Processing Techniques Office to lead the ARPANET effort. Roberts met Paul Baran in February 1967, though they didn’t discuss networks at that time.
In April 1967, ARPA convened a design session to address technical standards for the network. Roberts proposed that all mainframe computers connect directly, but other investigators resisted dedicating computing resources to network administration. Frank Westervelt was tasked with exploring message size and contents and writing a position paper on intercomputer communication protocol, which included conventions for character and block transmission, error checking, retransmission, and identification of both computers and users. After the session, Wesley Clark suggested using minicomputers as an interface to build a message switching network. Roberts revised the ARPANET plan accordingly and named those minicomputers Interface Message Processors, or IMPs.
The plan for what would become ARPANET was first shown at the very first Symposium on Operating Systems Principles, in October 1967. That’s when Donald Davies’ work on packet switching and the NPL network came to the attention of ARPA investigators. A colleague, Roger Scantlebury, presented Davies’ ideas, along with those of Paul Baran. Later, Roberts took Davies’ concept and applied it to the ARPANET project, while also reaching out to Baran for advice on dynamic routing. The NPL network was already running at 768 kbit/s, but the proposed speed for ARPANET was raised from 2.4 kbit/s to 50 kbit/s.
By mid-1968, Roberts and Barry Wessler finalized the IMP specification after a Stanford Research Institute report detailed the ARPANET communications network. Roberts presented the work to Taylor on 3 June, and he approved it on 21 June. Once ARPA gave its approval, they issued a Request for Quotation with 140 potential bidders. Most computer science companies thought the proposal was too ambitious, submitting only twelve bids. Of those, ARPA considered only four top contractors. By year’s end, they had narrowed it down to two and awarded the contract to build the network to BBN in January 1969.
The BBN team, led by Frank Heart and including Bob Kahn, Dave Walden, Severo Ornstein, and William Crowther, built the first working ARPANET system in just nine months. They followed Roberts’ ARPA plan closely, designing a network of small computers called IMPs, which acted as gateways between local resources. These IMPs handled store-and-forward packet switching and connected via leased lines using modems with data rates of 50kbit/s. Host computers linked to the IMPs through custom serial interfaces. The team worked closely with the NPL group, meeting in both the U.S. and the U.K. The system included all hardware and software, designed and installed quickly to fulfill ARPA’s needs.
The first-generation IMPs, like those on the NPL network, were built using a sturdy version of the Honeywell DDP-516 computer. This system had 24KB of magnetic-core memory that could be expanded, and a 16-channel Direct Multiplex Control unit to connect with host computers and modems. There were indicator lamps showing the status of each communication channel, and each IMP supported up to four local hosts and could link with six remote IMPs over early Digital Signal 0 leased phone lines. The network initially connected one computer in Utah with three in California. Later, universities were allowed to join the network so they could share hardware and software resources.
-
Read this chapter
The first four nodes became a testbed for the 1822 protocol, a major technical challenge. Even though they were electronically connected in 1969, actual network use didn’t begin until the Network Control Program was put in place in 1970. That allowed the first host-host protocols—Telnet for remote login and FTP for file transfer—to work, although these were developed between 1969 and 1973. The network was officially declared operational in 1971. By around 1973, email had spread to most sites, and network traffic started increasing.
-
Read this chapter
The very first ARPANET connected four locations: UCLA, ARC, UCSB, and the University of Utah School of Computing. UCLA hosted the network’s first node, where Leonard Kleinrock tested his ideas about message delay and studied how the system performed. The sites were chosen not just to cut down on leased line costs, but because each brought special knowledge that helped with this early stage of building the network.
At the University of California, Los Angeles, Kleinrock had set up a Network Measurement Center, or NMC, where the first computer attached was an SDS Sigma 7. This was the beginning of what would grow into the network that became the internet.
The Augmentation Research Center at Stanford Research Institute, now known as SRI International, was home to Douglas Engelbart and his work on the NLS system, an early hypertext system. That center would also host the Network Information Center, or NIC, using an SDS 940 computer running NLS, which was nicknamed "Genie." This machine became the first host connected to what would later grow into the internet.
At the University of California, Santa Barbara, a machine called the IBM 360/75 sat in the Culler-Fried Interactive Mathematics Center. This computer ran OS/MVT, and it was connected to what would become ARPANET. It was one of the first four hosts on the network. The system had been set up to support interactive mathematics research, and it played a key role in testing early networking protocols. This host was part of the initial group that helped define how the internet would work. It was not just any machine — it was specially configured for this experimental network. That setup marked an important step in building what we now call the internet.
On 29 October 1969, at 10:30 pm PST, Bill Duvall at Stanford Research Institute and Charley Kline at UCLA made the first successful host-to-host connection on what would become the ARPANET. Kline typed the command “login” from a computer at UCLA, but the Stanford Research Institute’s machine crashed after just two characters. An hour later, after Duvall adjusted parameters, Kline tried again and successfully logged in. The first two characters transmitted were “lo.” The first permanent link between the network’s IMPs was set up on 21 November 1969, and by 5 December 1969, the initial four-node ARPANET was established.
Elizabeth Feinler created the first Resource Handbook for ARPANET in 1969, and that work led to the development of the ARPANET directory. The directory was built by Feinler and a team, and it made it possible to navigate the ARPANET.
-
Read this chapter
In 1968, Roberts hired Kleinrock to assess how well the network was working and identify where it could improve. Building on his earlier research into queueing theory and message delay in communication systems, Kleinrock created mathematical models that described how packet-switched networks performed. These models became essential as ARPANET grew quickly during the early 1970s. While analytic methods were useful, they had limits, so computer simulations also played a role—ARPA used them, as did the NPL in the United Kingdom.
-
Read this chapter
Roberts brought in Howard Frank to help design the network’s structure, and Frank suggested ways to boost performance while cutting costs as the system expanded. By March 1970, the ARPANET had reached the East Coast, connecting an IMP at BBN in Cambridge, Massachusetts. The growth continued rapidly: nine IMPs by June 1970, then thirteen by December. By September 1971, there were eighteen IMPs and twenty-three hosts at universities and government agencies. Eighteen more IMPs joined by August 1972, and by September 1973, the count reached forty. The network hit forty-six IMPs by June 1974, and fifty-seven by July 1975. By 1981, there were 213 host computers, with a new one connecting roughly every twenty days.
In 1970, support was added for inter-IMP circuits running at up to 230.4 kbit/s. But because of concerns over cost and the processing power of the IMPs, this higher speed capability was never actively used.
Larry Roberts saw ARPANET and the NPL project as partners, and in 1970 he set out to join them with a satellite link. By 1971, Peter Kirstein’s team at University College London had been selected over NPL to handle the UK connection. Then, in June 1973, a transatlantic satellite linked ARPANET to the Norwegian Seismic Array, or NORSAR, using the Tanum Earth Station in Sweden and continuing on to a TIP at UCL via a landline. That setup made UCL a gateway for connecting ARPANET with British academic networks, marking the first international resource-sharing network and hosting some of the earliest experiments in internetworking.
In 1971, the Honeywell 316 began to be used as an IMP, replacing the earlier, more rugged models with a lighter version that was still capable of handling network traffic. This processor could also function as a Terminal Interface Processor, connecting up to sixty-three ASCII serial terminals through a multi-line controller instead of one of the main hosts. Compared to the 516 model, the 316 offered greater integration, which made it cheaper and easier to maintain. When configured as a TIP, it came with 40 kB of core memory, but by 1973, the amount was increased—first to 32 kB for IMPs and then to 56 kB for TIPs.
In 1975, BBN introduced IMP software that ran on the Pluribus multi-processor, and a few of these systems were installed at various sites. Then in 1981, BBN released new IMP software designed to run on its own C/30 processor product.
-
Read this chapter
The ARPANET was designed so that only BBN’s IMPs would work together, which meant there was no standard protocol for IMPs made by other companies. Still, these IMPs did communicate among themselves to handle routing, make sure messages got through reliably, and support remote monitoring from the Network Control Center. Each IMP started with a 6-bit identifier and could connect up to four hosts, each labeled with a 2-bit index. So an address combined both the host’s index on its IMP and the IMP’s identifier, written either as port/IMP or as one byte—for example, MIT-DMG’s address was shown as either 1/6 or 70. In early 1976, the system was upgraded to use 8-bit host numbers and 16-bit IMP numbers.
The IMP handled more than just routing and forwarding packets; it also ran several background programs like TTY, DEBUG, PARAMETER-CHANGE, DISCARD, TRACE, and STATISTICS. These programs were assigned their own host numbers so they could be addressed directly, allowing them to function independently of any connected host. For instance, the TTY program let an operator send ARPANET packets manually using a teletype that was connected straight to the IMP.
-
Read this chapter
In 1969, the ARPANET started connecting hosts using the 1822 protocol, which set the rules for sending messages to an IMP. The design made sure it would work with many different kinds of computers, using a message structure that included a type, a numeric address, and data. When one host needed to send information to another, it created a message containing the destination address and the data itself. This was passed through a hardware interface built for the 1822 protocol to the IMP. That device then sent the message either to a local host or to another IMP. Once the message reached its final stop, the receiving IMP responded with an RFNM signal to confirm delivery.
-
Read this chapter
The ARPANET was built to reliably send 1822 messages and tell the host computer if any were lost, something modern IP doesn’t do—it’s unreliable—while TCP makes it reliable. Still, the 1822 protocol couldn’t handle several connections at once from different apps on the same machine. To fix this, they created the Network Control Program, or NCP. The NCP made it possible to set up dependable, controlled communication between processes on different computers. It let application programs connect across the network using higher-level protocols, showing an early version of how protocol layers would later be organized in the OSI model.
Steve Crocker, a graduate student at UCLA, led the creation of NCP. He headed the Network Working Group, or NWG, which included other graduate students from universities and research labs. Jon Postel was among them, working at UCLA. The group received support from ARPA to build the ARPANET and the software needed for host computers to run applications.
-
Read this chapter
In 1972, Louis Pouzin and Hubert Zimmermann at IRIA in France introduced a streamlined end-to-end networking method called a catenet, presented at the ICCC. That same year, Bob Kahn left BBN to become President of Telenet before joining DARPA, first as program manager for the ARPANET under Larry Roberts, then later as director of the IPTO after Roberts moved to lead Telenet. Kahn explored both satellite and ground-based radio packet networks, recognizing the importance of linking them. At DARPA, Steve Crocker worked with British and French network leaders to form the International Network Working Group (INWG), and on Crocker's suggestion, Vint Cerf, now a Stanford faculty member, took over as its chair. Meanwhile, Bob Metcalfe developed the theory and practice behind Ethernet and the PARC Universal Packet. These efforts focused on how to connect different packet-switching networks—what came to be known as internetworking. Peter Kirstein demonstrated this concept in practice at University College London in 1973.
Research by Kahn and Cerf led to the creation of the Transmission Control Program in 1974. Cerf, along with Yogen Dalal and Carl Sunshine, wrote its specification at Stanford in December of that year, producing RFC 675. A group of researchers then began publishing Internet Experiment Notes, or IENs. The following year, testing started with concurrent implementations at Stanford, BBN, and University College London. Originally a monolithic design, the software was later restructured as a modular protocol stack in Version 4 in 1978, drawing on ideas from the French CYCLADES project and Bob Metcalfe’s work at Xerox Parc. It was installed in ARPANET for production use in January 1983, replacing NCP. By 1989, the full Internet protocol suite was developed, as outlined in RFC 1122 and RFC 1123, setting the stage for TCP/IP to become the core of the growing Internet.
-
Read this chapter
ARPA aimed to support advanced research, and the ARPANET was built with communications rather than user needs in mind. Yet by the summer of 1975, the network came under the operational control of the Defense Communications Agency. During this period, the first encryption devices for the ARPANET were introduced to manage sensitive data. A report on the project’s completion, authored in 1978 and released in 1981 by BBN and DARPA, concludes that:
The ARPANET program’s influence extended far beyond its original purpose, shaping the field of computer science in ways that continue to resonate. From the foundation of this network, the discipline of computer science itself grew stronger, supported and strengthened by the very technology it helped create. The feedback loop between the network and the science was direct and powerful, showing how closely intertwined the development of the internet and the study of computers had become. This relationship underscores the profound impact that ARPANET had on the future of computing and communication.
In 1981, access to the ARPANET expanded when the National Science Foundation funded the Computer Science Network, known as CSNET. This development marked a key step in growing the network that would eventually become the internet. The NSF's support helped extend connectivity beyond its original military and academic users, paving the way for broader scientific collaboration and data sharing across the United States.
In 1977, the ARPANET connected with SATNET, which had grown from earlier links to Norway and University College London. That year also saw PRNET join the network. The Department of Defense adopted TCP/IP as the standard military protocol in 1980. By 1982, Norway and UCL had left the ARPANET and switched to using TCP/IP over SATNET. Then on January 1, 1983—known as Flag Day—TCP/IP replaced the old Network Control Protocol across the entire ARPANET.
In September 1984, the work was finished on restructuring the ARPANET. That’s when the Military Network, or MILNET, was created for unclassified communications within the U.S. military. Both networks carried unclassified information and were connected at a few controlled gateways, which allowed them to be completely separated if needed. MILNET became part of the Defense Data Network, or DDN. The split reduced the original 113-node ARPANET by 68 nodes. After MILNET was separated, the ARPANET continued to serve as an Internet backbone for researchers, though it would eventually be phased out slowly.
-
Read this chapter
NCP set up a standard group of network services that many applications could use on one computer. Because of this, application protocols began to develop more or less on their own, separate from the basic network support. This made it possible for the network part and the application part to improve without blocking each other. The result was progress in both areas happening at the same time.
In the 1980s, early email systems used File Transfer Protocol, or FTP, to send messages, which was later replaced by the Simple Mail Transfer Protocol. As time went on, protocols like POP and IMAP came into use. These application protocols, including TELNET for remote access to time-sharing systems, were originally developed to work over different networking systems before eventually being adapted to run on TCP/IP. FTP, in particular, was key in enabling basic electronic mail functions during this transition period.
Abhay Bhushan wrote the original specification for the File Transfer Protocol, which he published as RFC 114 on April 16, 1971. By 1973, the File Transfer Protocol had been fully defined in RFC 354 and implemented, allowing users to transfer files over the ARPANET.
In 1971, Ray Tomlinson at BBN sent the first message over the ARPANET. That same year, a study found that three-quarters of all traffic on the network was made up of email messages. E-mail quickly became the dominant form of communication across the ARPANET, remaining a major part of its overall use for years to come.
In 1977, the Network Voice Protocol, or NVP, was laid out in a document called RFC 741, and it was built and put into use. But even though it existed, the system never worked well for making conference calls over ARPANET. The technology wasn’t ready for real-time voice communication, and it would be decades before something like today’s Voice over Internet Protocol, or VoIP, became possible.
-
Read this chapter
In 1971, Larry Roberts, who led ARPA at the time, asked for a way to protect passwords on the ARPANET. That request led to the creation of the Purdy Polynomial hash algorithm. It worked by computing a polynomial of degree 224 + 17, using a 64-bit prime number p = 264 − 59. The algorithm was later adopted by Digital Equipment Corporation for use in their VMS operating system and is still being used today for hashing passwords.
-
Read this chapter
Leonard Kleinrock says he did something that was technically illegal on what would become the internet. In 1973, after a meeting in England, he sent a message back through ARPANET asking for the return of his electric razor. At the time, using the network for personal matters was not allowed.
In 1978, Gary Thuerk of Digital Equipment Corporation broke the network's rules by sending what became the first mass email, targeting around 400 potential clients through ARPANET. He said that campaign brought in about $13 million in sales for DEC products and showed how powerful email marketing could be.
Using the ARPANet for anything not directly supporting government work is against the rules. Personal messages to other users—like arranging a meet-up or just saying hello—are usually seen as harmless. But sending email for profit or political reasons is both rude and illegal. That kind of activity can upset many people and could get MIT into serious trouble with the government agencies that run the network.
-
Read this chapter
In 1985, the NSF funded the creation of national supercomputing centers at several universities. The following year, 1986, they provided network access and interconnectivity through the NSFNET project. That network became the backbone of the Internet for government agencies and universities.
The ARPANET project came to an end in 1990 when it was officially decommissioned. The original IMPs and TIPs that powered the network were gradually taken offline as the newer NSFNet began to replace it. Still, some of the older IMPs kept running until July 1990, after the switch was made.
After the ARPANET was officially shut down on February 28, 1990, Vinton Cerf reflected on its legacy with a piece he wrote called "Requiem of the ARPANET."
-
Read this chapter
The ARPANET wasn’t created to build a Command and Control System that could survive a nuclear attack, even though that sounds like what many people think today. That kind of system would have been a major military need, but it wasn’t ARPA’s job to do that — in fact, they would’ve faced serious criticism if they had tried. Instead, the network came from frustration over how few large, powerful research computers existed in the country. Many researchers who needed access to them were separated from them by geography.
The ARPANET relied on distributed computing and regularly updated its routing tables, which was technically difficult back then. This approach made the network more likely to keep working even if parts of it were damaged or went down. It was built to handle losses in smaller networks without failing completely. The Internet Society agrees with Herzfeld’s view, as noted in a footnote to their online article titled A Brief History of the Internet.
A false idea took hold that the ARPANET was built to resist nuclear war, but that rumor came from the RAND study. The ARPANET itself wasn’t designed for that purpose, though the earlier RAND work on secure communication included such considerations. Later developments in internetworking placed strong emphasis on robustness and survivability, ensuring the system could handle the loss of major parts of the network.
Paul Baran first proposed a model for communication using packet switching and led the RAND study on the topic. Although the ARPANET didn’t fully match his original goals, he believed his work helped shape it. At a design meeting on 9–10 October 1967, notes by Elmer Shapiro show that a version of Baran’s routing method, called “hot potato,” was considered. This idea aligned with a proposal from the NPL team presented at the Symposium on Operating System Principles in Gatlinburg.
In the 1970s, ARPA placed emphasis on the goal of “command and control.” Stephen J. Lukasik served as deputy director from 1967 to 1970 and later became Director of DARPA, holding that position from 1970 to 1975.
The objective was to enhance military command and control in the face of nuclear dangers, ensure the survival of U.S. nuclear forces, and boost tactical and management decisions within the armed forces. This effort aimed at leveraging emerging technologies to support national defense strategies during a time of heightened Cold War tensions. The focus remained on creating systems that could withstand potential attacks while maintaining effective communication and control over critical military operations. These goals shaped early developments in network architecture and laid groundwork for what would later become the Internet.
Read
Free to download, keep and share. For general information only — not professional medical, legal or financial advice. Please consult a qualified professional.