TCP/IP: The Protocol That Accidentally Built the Internet

On January 1, 1983, something quiet and consequential happened on a Pentagon-funded research network called ARPANET. Around 400 connected computers simultaneously cut over from an old protocol called NCP (Network Control Program) to a new one called TCP/IP. There was no announcement outside the research community. No press coverage. A button was effectively flipped, a handful of engineers held their breath, and by the end of the day the modern internet existed.

Every website you visit, every message you send, every video you stream, every banking transaction you make — all of it runs on top of the architecture that was designed in the early 1970s to survive a nuclear war and switched on overnight in 1983.

THE PROBLEM ARPANET HAD TO SOLVE

ARPANET, funded by the U.S. Department of Defense's Advanced Research Projects Agency (ARPA, later DARPA), went live in October 1969 with four nodes: UCLA, Stanford Research Institute, UC Santa Barbara, and the University of Utah. It was the first operational packet-switched network. It used a protocol called NCP for host-to-host communication.

NCP had two critical limitations. First, it was tightly coupled to ARPANET's specific hardware — you couldn't easily route packets between ARPANET and another network using different cables, different speeds, or different conventions. Second, NCP assumed the underlying network was reliable. If a node went down, the network's behavior became unpredictable.

By the early 1970s, ARPANET was not the only packet-switched network in the world. Xerox had built its own internal network. The UK had NPL. France had CYCLADES. Hawaii had ALOHANET (radio-based). Satellite packet networks were being tested. Each network spoke its own protocol. None of them could talk to each other.

CERF AND KAHN

In 1973, Vint Cerf (a Stanford professor) and Robert Kahn (a DARPA program manager who had previously worked at BBN, the company that built ARPANET's routers) started working on a solution. Their goal was a protocol that could work across multiple different underlying networks — wired, radio, satellite — without any of them needing to know each other's internal details. They called this an 'internetwork' protocol. The shortened version, 'internet,' eventually became the name of the whole thing.

The paper they published in 1974, 'A Protocol for Packet Network Intercommunication,' in IEEE Transactions on Communications, is effectively the internet's birth certificate. It introduced the core ideas: break data into small self-contained packets, put a standard header on each packet with source and destination addresses, let each network route the packets using whatever mechanism it had, and reassemble the packets in order at the destination.

The original specification was one combined protocol called TCP. Over the next several years, Cerf, Kahn, Jon Postel, Danny Cohen, and others split it into two layers: IP (Internet Protocol) for addressing and routing of individual packets, and TCP (Transmission Control Protocol) for reliable, ordered delivery of streams of packets on top of IP. The split allowed other protocols — UDP for lightweight unreliable delivery, ICMP for diagnostics, later streaming and voice protocols — to ride on the same IP layer without needing TCP's complexity.

PACKET SWITCHING AS A SURVIVAL PROPERTY

The defense argument for packet switching — sometimes summarized as 'designed to survive a nuclear attack' — is often exaggerated. The engineering goal was more mundane: a network that degrades gracefully when parts of it fail, and that can route around damage without requiring centralized control. In a dense Cold War threat environment, that property was highly desirable. But the same property turned out to be useful for ordinary faults: broken cables, failed routers, congested links, datacenter outages.

The same architecture that would have routed packets around a crater can route them around a severed undersea cable in the Pacific. Both have happened. The internet routes around the damage either way.

THE FLAG DAY

Throughout the late 1970s and early 1980s, TCP/IP was tested on ARPANET in parallel with NCP. By 1981, the standards were finalized in RFC 791 (IP) and RFC 793 (TCP). DARPA set a hard transition date: January 1, 1983. Every host on ARPANET had to support TCP/IP by that date or be disconnected. Stickers were printed that read 'I Survived the TCP/IP Transition.' Some hosts stopped talking. Most did not.

That day is arguably the true birthday of the internet. Before January 1, 1983, ARPANET was a research network with specialized protocols. After, it was an internetwork — the seed of the public internet that would explode into public consciousness a decade later with the World Wide Web.

WHY IT MATTERS

TCP/IP is one of the most durable pieces of infrastructure humans have ever built. The specifications are fifty-plus years old. They predate the personal computer, the mobile phone, the fiber optic backbone, streaming video, and the modern cloud. And yet, today, more than 5.5 billion people and an estimated 25 billion devices talk to each other using packet headers that Vint Cerf and Bob Kahn sketched out in a hotel lobby in San Francisco.

IPv4, the version of the Internet Protocol that was standardized in 1981, is still in use, though its 4.3 billion address space has been exhausted since 2011. IPv6, designed in the 1990s to solve that exhaustion, now carries roughly 45% of global traffic and is growing. But the fundamental architecture — dumb network, smart endpoints, packets with self-contained headers, best-effort routing — is unchanged.

Cerf and Kahn jointly received the Turing Award in 2004, the Presidential Medal of Freedom in 2005, and the QEII Prize for Engineering in 2013. They are, by nearly any reasonable accounting, the two people most responsible for the shape of the modern internet. Both are still alive. Both still work on internet research.

FUTURE — WHAT STILL COMES NEXT

Fifty years into the TCP/IP era, the architecture is showing its age in specific places. Latency-sensitive applications — video calls, cloud gaming, autonomous vehicles — push against the design assumption that the network is best-effort and endpoints absorb the cost. Content-delivery architectures have layered CDNs, caches, and edge computing on top of basic IP routing. QUIC, a newer transport protocol now carrying a significant portion of web traffic, moves much of what TCP used to do into the application layer.

But the base layer is still IP packets with headers designed in 1974. The thing you are using to read this sentence is fifty years old and still working. That is either the best or luckiest piece of engineering in the modern world — and a reminder that infrastructure tends to long-outlive the problem it was originally designed to solve.

(Sources: Vinton G. Cerf and Robert E. Kahn, 'A Protocol for Packet Network Intercommunication,' IEEE Transactions on Communications, May 1974; RFC 791 (Internet Protocol), September 1981; RFC 793 (Transmission Control Protocol), September 1981; Internet Society historical archive; Katie Hafner, 'Where Wizards Stay Up Late,' Simon & Schuster, 1996; ACM A.M. Turing Award citation for Cerf and Kahn, 2004)