ARPANET

1.1 Cold War context and the need for resilient communication

The development of ARPANET was influenced by the geopolitical tensions of the Cold War. The U.S. Department of Defense sought communication systems that could survive large-scale disruptions, including nuclear attacks. Traditional circuit-switched telephone networks were vulnerable because they relied on central switching offices; a single point of failure could sever entire regions. In the early 1960s, researchers at the RAND Corporation, notably Paul Baran, proposed a distributed network using packet switching, where messages are broken into small blocks and routed dynamically through multiple paths. This concept promised robustness: even if several nodes were destroyed, the network could reroute traffic around damage. ARPA (the Advanced Research Projects Agency, later renamed DARPA) became the driving force behind turning these theoretical ideas into a practical network.

1.2 J.C.R. Licklider and the concept of "Intergalactic Computer Network"

In 1962, J.C.R. Licklider became the first head of ARPA's Information Processing Techniques Office (IPTO). Licklider envisioned a network that would connect computers not merely for file sharing, but for collaborative interaction and resource sharing. He wrote memos describing an "Intergalactic Computer Network," a global system where users could access data and programs from any connected machine. Though the term was whimsical, the concept laid the philosophical groundwork for ARPANET. Licklider's ideas influenced his successors, including Ivan Sutherland and Robert Taylor, who later launched the ARPANET project.

1.3 Key institutions (ARPA, RAND, MIT, UCLA)

ARPANET's development involved multiple research centers. ARPA provided funding and strategic direction. RAND contributed theoretical studies on survivable networks. MIT, especially through researchers like Leonard Kleinrock (who pioneered queuing theory for packet switching), provided mathematical foundations. UCLA, where Kleinrock moved in the mid-1960s, became the site of the first ARPANET node. Other institutions, including Stanford Research Institute (SRI) and the University of Utah, played key roles in designing protocols and hosting early nodes. The collaboration among these academic and military-contractor organizations set a pattern for open, consensus-driven networking research.

2.1 Early design and specifications (1966–1968)

In 1966, ARPA's director Robert Taylor initiated the ARPANET project, with the goal of connecting ARPA-funded computer centers. He hired Larry Roberts from MIT to lead the effort. Roberts refined the packet-switching concept, drawing on Baran's work and independent research by Donald Davies in the UK. By 1967, the first design specifications were drafted, calling for a subnetwork of dedicated minicomputers called Interface Message Processors (IMPs) that would handle packet routing and relieve host computers of network duties.

2.1.1 Initial proposals (BBN's role)

In 1968, ARPA issued a request for proposals to build the IMPs. The contract was awarded to Bolt, Beranek and Newman (BBN), a Cambridge, Massachusetts consulting firm. BBN's team, led by Frank Heart, designed the IMP hardware and software. They selected the Honeywell DDP-516 as the platform and developed custom software for packet store-and-forward operations. BBN's proposal included a novel distributed routing algorithm that allowed IMPs to adapt to network topology changes without central control.

2.1.2 First packet-switching nodes (Interface Message Processors)

The IMP was a dedicated minicomputer with up to 24 kilobytes of memory, equipped with special interfaces to connect to host computers and other IMPs. Each IMP operated independently, running a real-time operating system that handled packet queuing, error checking, and routing table updates. The first IMP was delivered to UCLA in September 1969, and subsequent units were installed at SRI, UCSB, and the University of Utah.

2.2 First connections (1969)

The initial ARPANET consisted of four nodes, each connecting a host computer to an IMP. The IMPs were linked via 50 kbit/s leased telephone lines.

2.2.1 UCLA, Stanford Research Institute, UCSB, University of Utah

• UCLA: Node 1, with a Sigma-7 host. Leonard Kleinrock's team managed the first IMP.
• Stanford Research Institute (SRI): Node 2, with an SDS-940 host. The Network Information Center (NIC) was later established here.
• University of California, Santa Barbara (UCSB): Node 3, with an IBM 360/75. Researchers focused on visualization and graphics.
• University of Utah: Node 4, with a DEC PDP-10. The site specialized in computer graphics, including work by Ivan Sutherland.

2.2.2 The first message (October 29, 1969)

On October 29, 1969, a UCLA graduate student, Charley Kline, attempted to log into the SRI host from UCLA. He typed "LOGIN" but the system crashed after transmitting the letters "L" and "O". The first message was thus "LO" — a piece of internet lore often retold as a friendly "hello" before the network's teething problems. After debugging, the full login succeeded, marking the birth of packet-switched networking between distant computers.

2.3 Expansion through the 1970s

Following the initial four-node network, ARPANET grew rapidly. By 1972, there were over 20 nodes, and the network was demonstrated publicly at the International Conference on Computer Communications (ICCC). The growth was driven by increasing demand from research institutions for remote access to supercomputers, databases, and collaborative tools.

2.3.1 Growth to dozens of nodes

Throughout the 1970s, ARPANET expanded to more than 100 nodes, including sites at major universities, military research labs, and government contractors. The network employed a hierarchical topology, with IMPs connected by multiple redundant links. The IMP software was updated to support larger routing tables and new features like packet fragmentation.

2.3.2 International connections (e.g., University College London, NORSAR)

The first international connection came in 1973, linking ARPANET to University College London (UCL) via a satellite link to Norway. Soon after, a connection to NORSAR (the Norwegian Seismic Array) was established. These links demonstrated the viability of global packet-switching and inspired similar national networks (e.g., Cyclades in France, NPL in the UK).

3.1 Packet switching

ARPANET was built on the principle of packet switching, where data is divided into small packets that travel independently through the network and are reassembled at the destination. This contrasts with circuit switching, where a dedicated path is reserved for the entire communication.

3.1.1 Store-and-forward operation

Each IMP operated in a store-and-forward manner: it would receive a complete packet, check for errors, store it temporarily in memory, and then forward it to the next IMP toward its destination. This allowed the network to recover from packet loss by retransmission and to use link bandwidth efficiently by interleaving packets from different conversations.

3.1.2 Message vs. packet concepts

An ARPANET "message" could be up to 8095 bits long and was broken into smaller "packets" (typically 1000 bits each) for transmission. The IMPs handled packet-sequencing and reassembly. This layering—splitting messages into packets—was crucial for fair access and error control, and later influenced Internet Protocol design.

3.2 Interface Message Processor (IMP)

The IMP was the heart of ARPANET's physical infrastructure. Each site had at least one IMP, which connected directly to the telephone lines and to the local host computer via a serial interface.

3.2.1 Hardware design (Honeywell DDP-516)

BBN chose the Honeywell DDP-516, a rugged 16-bit minicomputer with a cycle time of about 1 microsecond. It had 12 to 24 kilobytes of magnetic core memory. The IMP used a specialized front panel with indicator lights and switches for debugging. Later versions, such as the IMP-II, used a modified Honeywell 316, which was smaller and cheaper.

3.2.2 Software and routing algorithms

The IMP software was written in assembly language. It implemented a dynamic routing algorithm that periodically exchanged hop-count information with neighboring IMPs. Each IMP maintained a routing table that could be updated every 0.6 seconds. The algorithm was an early form of distance-vector routing, later formalized as the Bellman–Ford algorithm. It allowed the network to adapt to node failures and link congestion.

3.3 Host-to-IMP protocols

Host computers communicated with the local IMP using a defined hardware and software interface. The host sent data in the form of "messages" (up to 8159 bits) to the IMP, which then fragmented them into packets for transmission. The IMP would reassemble packets from the network into a message and deliver it to the destination host. This separation of host and IMP responsibilities simplified host hardware and allowed different computer architectures to participate.

3.4 Network Control Protocol (NCP)

Before TCP/IP, ARPANET used the Network Control Protocol (NCP) to manage host-to-host communication. NCP was implemented in 1970 and provided connection-oriented, reliable data transfer.

3.4.1 Connection establishment and maintenance

NCP used a system of "sockets" and "connections." A host would open a socket (a communications endpoint) and then initiate a connection to a remote socket. NCP handled flow control by using a "ready" signal to prevent buffer overruns. It was simple but designed only for the single ARPANET environment—it could not handle networks with different characteristics.

3.4.2 Limitations leading to TCP/IP

NCP had several limitations: it assumed a reliable underlying network (the IMPs provided error-free delivery), it lacked addressing beyond ARPANET's 8-bit host numbers, and it could not handle packet loss in interconnected networks. As ARPANET began connecting to other networks (e.g., packet radio net, satellite net), these shortcomings became critical, driving the creation of the Transmission Control Protocol (TCP) and later the Internet Protocol (IP).

4.1 Transition from NCP to TCP/IP

The move from NCP to TCP/IP was a major architectural shift, making ARPANET the backbone of an internetwork (hence the term "Internet").

4.1.1 Robert Kahn and Vint Cerf’s vision (1973–1974)

In 1973, Robert Kahn (then at ARPA) and Vint Cerf (at Stanford) began designing a "protocol for packet network interconnection." Their goal was a protocol that could connect heterogeneous networks without assuming reliability at the lower layers. They published a seminal paper in 1974 outlining TCP, which included both transport and internetworking functions. The protocol introduced the concept of a "gateway" (later router) that would fragment packets as needed and route them between networks.

4.1.2 IP addressing and packet encapsulation

TCP/IP introduced a 32-bit IP address to uniquely identify each host across the entire internetwork. IP encapsulated TCP segments into datagrams, which were then forwarded by gateways. This encapsulation allowed ARPANET, radio net, and satellite net to exchange data seamlessly.

4.1.3 TCP split into TCP and IP (1978)

In 1978, to improve modularity, the original TCP was split into separate components: IP (providing unreliable, connectionless packet delivery) and TCP (providing reliable, connection-oriented stream service). This separation allowed other transport protocols (like UDP) to be built on top of IP. The split was codified in RFC 791 (IP) and RFC 793 (TCP), both published in 1981.

4.2 Role of the Internet Engineering Task Force (IETF) precursors

Before the formal IETF was established in 1986, protocol development for ARPANET was managed by the Network Working Group (NWG), which operated through informal meetings and a series of "Request for Comments" (RFCs). The NWG was open to anyone with a technical contribution, and decisions were made by rough consensus and running code—an ethos that later defined the IETF.

4.3 Key RFC documents

4.3.1 RFC 1 (1969) – Host Software

RFC 1, written by Steve Crocker, described the host-to-IMP interface and the initial software requirements. It marked the beginning of the RFC series, which remains the primary documentation channel for Internet standards.

4.3.2 RFC 791, 793 (1981) – IP and TCP specifications

RFC 791 defined the Internet Protocol, including the datagram format, addressing (32-bit addresses), fragmentation rules, and routing. RFC 793 defined the Transmission Control Protocol, specifying connection establishment, flow control, retransmission timeouts, and error recovery. These two RFCs became the foundation of the TCP/IP suite used on the modern Internet.

5.1 Early network applications

ARPANET's primary use was to share scarce and expensive computing resources, but it quickly enabled new communication tools.

5.1.1 Telnet (remote login)

Telnet, specified in RFC 97 in 1971, allowed users to log into a remote host as if they were using a local terminal. It was one of the first applications, enabling researchers to use supercomputers at distant sites.

5.1.2 FTP (file transfer)

The File Transfer Protocol (FTP), defined in RFC 114 (1971) and later updated by RFC 265 (1971) and RFC 354 (1972), provided a way to copy files between hosts. FTP allowed binary and text transfers, directory listing, and authentication.

5.1.3 Email (1972 – Ray Tomlinson, @ sign)

Email emerged as a killer application. In 1971, Ray Tomlinson of BBN wrote a program that could send messages between users on different ARPANET hosts. He chose the @ symbol to separate the user name from the host name (e.g., "user@host"). By 1973, email accounted for most ARPANET traffic. The first email management programs, such as SNDMSG and RDMAIL, were soon replaced by more sophisticated tools like MH and Pine.

5.2 Resource sharing and remote computing

Beyond messaging, ARPANET enabled resource sharing. Researchers at UCLA could run jobs on a supercomputer at Illinois while using local processing. The network also supported shared databases, remote printing, and even early networked games (e.g., "Spacewar!" over the network in 1972). This collaborative access fostered new forms of scientific research.

5.3 Influence on later Internet services (e.g., Usenet, SMTP)

The applications developed on ARPANET directly shaped later Internet services. Email evolved into SMTP (RFC 821, 1982). FTP became the basis for TFTP and later HTTP's file transfer methods. Telnet influenced SSH. And the concept of mailing lists and newsgroups led to Usenet (1979), which used UUCP to propagate messages across networks. ARPANET's applications demonstrated that network value grows with the number of participants—a lesson embodied in Metcalfe's Law and in the explosion of the World Wide Web in the 1990s.

6.1 Network Working Group (NWG)

The Network Working Group (NWG) was an informal body that coordinated the development of ARPANET protocols and standards. It was led by Steve Crocker and Jon Postel, who maintained the RFC series.

6.1.1 Informal consensus-based decision making

The NWG operated on a Bazaar model: anyone could propose a protocol via an RFC, and adoption was driven by demonstration and implementation, not by formal voting. This "rough consensus and running code" approach was later institutionalized by the IETF. The NWG's openness allowed rapid innovation but occasionally led to divergent implementations.

6.2 ARPA's oversight and funding

ARPA (later DARPA) funded ARPANET through its Information Processing Techniques Office (IPTO). The agency set high-level goals (e.g., survivable networking, resource sharing) but left technical details to the researchers. ARPA approved expansion requests and managed contracts with BBN for IMP maintenance and operations. This light-touch management enabled creative freedom while ensuring the network served defense-related research.

6.3 Decommissioning process (1989–1990)

By the late 1980s, ARPANET had been superseded by NSFNET, a faster backbone for academic and research use. The original 56 kbit/s lines were outdated, and maintaining the aging IMP hardware became difficult.

6.3.1 Migration to NSFNET and the commercial Internet

NSFNET, funded by the National Science Foundation, had begun operations in 1986 with a 1.5 Mbit/s backbone (later upgraded to 45 Mbit/s). Most ARPANET traffic had already shifted to NSFNET. ARPANET was officially decommissioned in February 1990. Many ARPANET nodes were simply shut down, while others transitioned to the new Internet backbone. The decommissioning marked the end of an era, but the protocols and practices developed on ARPANET lived on in the global Internet.

7.1 Technological contributions

7.1.1 Packet switching as a paradigm

ARPANET proved that packet switching could work at scale, providing efficient, error-corrected data transport over unreliable lines. This paradigm replaced circuit-switched telephony for data and became the foundation of all modern networks.

7.1.2 Redundancy and distributed routing

The distributed routing algorithms pioneered in ARPANET demonstrated that a network could survive node failures without losing connectivity. This redundancy, along with end-to-end error control in TCP, offered resilience that centralized networks could not match. These design principles are now standard in the Internet.

7.2 Social and cultural influence

7.2.1 Academic collaboration and open standards

ARPANET fostered a culture of open collaboration, where researchers freely shared ideas, software, and data. The RFC series was a key instrument, enabling anyone to contribute to the design process. This spirit of openness later characterized the World Wide Web and the open-source movement.

7.2.2 Birth of the "hacker ethic" and internet culture

The early ARPANET community was small, technically skilled, and informal. Users often tinkered with the system, wrote new programs, and ignored rigid hierarchies. This environment birthed what became known as the "hacker ethic"—the belief that information should be free, that systems should be decentralized, and that creativity should triumph over bureaucracy. Early email mailing lists, such as the "SF-LOVERS" discussing science fiction, and later "netiquette" norms emerged from this culture. The notion of "flaming" and "trolling" also has its roots in ARPANET discussions.

7.3 Relationship with the modern Internet

7.3.1 Direct technical lineage

The modern Internet is a direct descendant of ARPANET. TCP/IP, DNS (Domain Name System), and many basic application protocols were designed and tested on ARPANET. The IMP architecture anticipated routers, and the NCP/TCP/IP evolution set the layered model that continues today.

7.3.2 Lessons for network security and resilience

ARPANET had almost no security: any user could access any host, and passwords were rarely encrypted. This lack of security led to the first "worm" incident (the Morris worm of 1988, which exploited protocol vulnerabilities) and taught painful lessons about authentication, encryption, and firewalls. The Internet's current focus on security—via SSL/TLS, IPsec, and vigilant administration—is in large part a response to ARPANET's naive trust model.

8.1 Early security and privacy concerns

From the start, ARPANET had no built-in authentication or encryption. Anyone with network access could attempt to log into any host. This openness led to incidents of unauthorized access and data breaches. In the 1970s, some researchers warned that the network could be exploited for espionage or vandalism. Later, the 1988 Morris worm exploited flaws in Unix sendmail and finger, bringing down about 10% of the Internet at the time—a direct consequence of the trust-centric design inherited from ARPANET.

8.2 Debates over DARPA's military role

Because ARPANET was funded by the Department of Defense, critics argued that the network promoted military objectives under the guise of academic research. Some academics refused to participate for political reasons. Others saw the network as a tool for the military–industrial complex. The debate continued into the 1980s, when DARPA's involvement in the early Internet sparked discussions about network neutrality and government control. Proponents countered that ARPANET's research was unclassified and open, and that it ultimately benefited civilian science.

8.3 Scale and cost management

ARPANET was expensive to build and maintain, especially the leased phone lines and IMP hardware. Critics pointed out that the network served a relatively small community of researchers—about 200 hosts by the early 1980s—and questioned the return on investment. Defenders noted that the network's value was in enabling new forms of collaboration that eventually led to the global Internet. Later, the move to NSFNET, which was funded by a civilian agency, helped separate the network from direct military control.

• Internet history
• Packet switching
• TCP/IP
• Network Control Program
• Interface Message Processor
• BBN Technologies
• J.C.R. Licklider
• Robert Kahn and Vint Cerf
• NSFNET
• ARPA

• Abbate, Janet. *Inventing the Internet*. MIT Press, 1999.
• Hafner, Katie, and Matthew Lyon. *Where Wizards Stay Up Late: The Origins of the Internet*. Simon & Schuster, 1996.
• Cerf, Vinton, and Robert Kahn. "A Protocol for Packet Network Intercommunication." *IEEE Transactions on Communications*, 1974.
• RFC 1 (1969), RFC 791 (1981), RFC 793 (1981). Request for Comments series.
• Kirstein, Peter. "Early Experiences with the ARPANET and the Internet in the UK." *IEEE Annals of the History of Computing*, 2009.
• Comer, Douglas. *Internetworking with TCP/IP: Principles, Protocols, and Architecture*. Prentice Hall, 1991.