The Connected World: From Packet Switching to the Global Web

Zusammenfassung

This article traces the architectural evolution of global networking — from the Cold War origins of packet switching, through the chaotic “Protocol Wars” of the 1970s and 80s, to the emergence of TCP/IP as the universal language of the Internet and Tim Berners-Lee’s invention of the World Wide Web. It is a story driven as much by geopolitical fear, bureaucratic stubbornness, and a handful of stubborn idealists as by engineering brilliance.

The Problem: Circuits Built for Voices

For most of the twentieth century, long-distance communication meant circuit switching — the technology underlying every telephone network on earth. When you placed a call, the telephone exchange established a dedicated physical path between caller and receiver that persisted for the entire duration of the conversation. Every switching point along that path was reserved for you alone, whether you were speaking or sitting in silence.

For human voices, this was acceptable. For computers, it was catastrophically inefficient. Computer communication is “bursty” — a file transfer sends enormous amounts of data for seconds, then nothing; a database query demands an instant response. Reserving an entire circuit for this kind of traffic was wasteful in peacetime. In war, it was existentially dangerous: a single nuclear strike on a switching hub could sever communication across an entire continent.

Paul Baran and the Nuclear Calculus

In the early 1960s, a RAND Corporation engineer named Paul Baran sat down to solve a specific problem: how could the United States military maintain command-and-control communications after a nuclear first strike? The existing telephone network, with its centralized switching hubs, offered a clear answer — it couldn’t.

Baran’s solution was radical. Instead of centralized switching, he proposed a distributed network — a mesh of nodes, each with multiple connections to its neighbors, and none indispensable. Data would be broken into small, standardized chunks — “message blocks,” later called packets — each carrying its own addressing information. Each packet would find its own route through the network, hopping from node to node, and the full message would be reassembled at the destination.

The concept meant that no single point of failure could bring down the whole network. Destroy a node, and packets simply routed around it.

Baran published his findings in a RAND report in 1964. The U.S. military’s response was skeptical and, for years, entirely unhelpful. AT&T, whose circuit-switched empire was implicitly threatened by the idea, was openly hostile. “There’s no way that can work,” AT&T engineers reportedly told Baran.

Simultaneously and independently, Donald Davies at the National Physical Laboratory in the UK arrived at essentially the same architecture. It was Davies who coined the word “packet” — the term that stuck.

ARPANET: The Network That Learned to Walk

The theoretical framework existed by the mid-1960s. Building an actual network required money, institutions, and engineers willing to spend years on something most people considered impossible.

That money came from the U.S. Department of Defense’s Advanced Research Projects Agency (ARPA). Not for survivability — the nuclear-war rationale had largely faded by then — but for a more mundane purpose: allowing expensive research computers at different universities to share resources. The project was led by J.C.R. Licklider, a psychologist turned computer visionary who had written compellingly about a future “Intergalactic Computer Network,” and later by Lawrence Roberts.

On October 29, 1969, the first message was sent over the ARPANET. It was transmitted from a computer at UCLA, operated by student programmer Charley Kline, to a machine at the Stanford Research Institute. The intended message was “LOGIN.” The system crashed after two letters. The first message ever sent over the internet was: “LO.”

The network recovered. By 1971, ARPANET connected 15 nodes. By 1973, it had reached the UK and Norway. What had started as a military resource-sharing experiment was quietly becoming the infrastructure of a global community.

Vint Cerf, Bob Kahn, and the Language of the Internet

A network of dozens of nodes raised a manageable set of problems. A network of thousands — built from incompatible machines running different software on different hardware — raised a much harder one: how could any of them talk to each other?

The answer came in 1974, when Vint Cerf and Bob Kahn published “A Protocol for Packet Network Interconnection” — one of the most consequential papers in the history of technology. They proposed a two-layer architecture:

IP (Internet Protocol): Handles routing — ensuring each packet finds its way to the correct address, across any combination of underlying networks. IP is deliberately “dumb”: it makes no guarantees about delivery or order. It simply routes.
TCP (Transmission Control Protocol): Handles reliability — numbering packets, detecting losses, requesting retransmissions, and reassembling data in the correct order at the destination.

The separation was a design decision of genius. By keeping IP simple and universal, Cerf and Kahn allowed TCP/IP to run over anything — telephone wires, radio links, satellite connections, the early Ethernet. No single hardware technology owned the internet.

Open Standards vs. Proprietary Protocols

The victory of TCP/IP was not inevitable — and it was explicitly political. Throughout the 1970s and 80s, a “Protocol War” raged between the open, flexible architecture of TCP/IP and more rigid, state-controlled standards. TCP/IP won for two reasons: it was already deployed and working on the ARPANET, and it was given away free. When universities and research institutions adopted it, they built a critical mass that no competing standard could match. The lesson — that freely available open standards can defeat technically superior proprietary systems — would repeat throughout computing history.

Jon Postel: The Man Who Ran the Internet from His Office

Behind the elegant architecture of TCP/IP lay an unglamorous but essential task: keeping track of which computer had which address. For much of the internet’s early history, this responsibility fell to a single person — Jon Postel of the University of Southern California’s Information Sciences Institute.

Postel maintained the Assigned Numbers registry — the authoritative list of who owned which protocol numbers, port assignments, and internet addresses. He also co-invented the Domain Name System (DNS) with Paul Mockapetris in 1983, which translated human-readable addresses like mit.edu into numerical IP addresses.

In January 1998, in what became known as the “Postel Incident,” Postel sent an email to eight of the twelve organizations running internet root servers, asking them to redirect to a server under his control. Within hours, he controlled the routing of a significant fraction of global internet traffic — from his office. The U.S. government, alarmed, pressured him to reverse the change. He complied. The incident exposed how much of the internet’s stability rested on one man’s reputation and goodwill. Postel died later that year, of heart surgery complications. The IANA (Internet Assigned Numbers Authority) that succeeded him required a formal international organization to replicate the trust he had embodied alone.

Dead End: The OSI Model and the War the Engineers Lost

While TCP/IP was evolving through practice and deployment, an international consortium was building what many believed would be the definitive answer: the OSI model (Open Systems Interconnection), developed by the International Organization for Standardization (ISO) through the late 1970s and 1980s.

OSI was comprehensive, rigorous, and backed by enormous resources. National governments mandated it. European telecommunications ministries funded it. The model defined seven distinct layers of network communication — physical, data link, network, transport, session, presentation, application — with clear interfaces between each.

It was also nearly impossible to implement correctly. The specification process, managed by committee across dozens of countries and thousands of stakeholders, produced documents of crushing complexity that different vendors interpreted differently. Products claiming OSI compliance often could not interoperate.

Meanwhile, TCP/IP — built by a small community of engineers working by rough consensus, documented in informal “Request for Comments” (RFC) memos, and freely available — was already running. Universities and research institutions had deployed it at scale. By the time OSI implementations began appearing, the installed base of TCP/IP was insurmountable.

In 1989, the U.S. Department of Defense — which had mandated OSI for future military systems — quietly announced it would continue using TCP/IP. OSI was never formally declared dead; it simply stopped being relevant. Its seven-layer model survived as a teaching framework, a conceptual map of networking concepts that educators still use today. The actual protocol died unmourned.

The Lesson of OSI

OSI represents the failure mode of standardization by committee: technically comprehensive, politically representative, and functionally too late. Compare the fate of Novell’s IPX/SPX and other proprietary protocols covered in Early Networking Failures — the pattern repeats. Deployment and momentum defeat elegance and comprehensiveness, almost every time.

Tim Berners-Lee and the Web

By 1989, the internet existed and worked. What it lacked was usability. Navigating it required knowing IP addresses, understanding Unix command lines, and using tools — FTP, Gopher, Telnet — that assumed technical sophistication. The internet was a highway with no signage.

Tim Berners-Lee was a British physicist working at CERN, the European particle physics laboratory in Geneva. His immediate problem was mundane: CERN employed thousands of researchers who constantly left, taking their accumulated knowledge with them. There was no good way to record and cross-reference the institutional memory of the organization.

In March 1989, Berners-Lee submitted a proposal to his manager, Mike Sendall, for “a large hypertext database with typed links.” Sendall’s handwritten comment in the margin has become famous: “Vague but exciting.” He approved it.

Berners-Lee’s insight was the combination of three existing ideas: hypertext (text with clickable links to other documents), TCP/IP (the network transport), and URLs (Uniform Resource Locators, a universal addressing system). He added HTTP (Hypertext Transfer Protocol, a simple request-response protocol for fetching documents) and HTML (HyperText Markup Language, a simple format for writing documents with links).

On December 20, 1990, the first web server went live on Berners-Lee’s NeXT workstation at CERN. Its hand-written label read: “This machine is a server. DO NOT POWER IT DOWN!!”

What distinguished Berners-Lee from many inventors was a deliberate choice about intellectual property. CERN could have patented the Web and licensed it commercially. Berners-Lee argued against it. In April 1993, CERN released the World Wide Web into the public domain, free for anyone to use and build upon. Within two years, the web had grown from a handful of academic sites to millions of pages. Within ten years, it had become the primary interface through which most of the world’s population experienced the internet.

The Web Is Not the Internet

A common conflation: the Internet and the Web are different things. The Internet is the physical and protocol infrastructure — the cables, routers, IP addresses, and TCP/IP stack — that has existed since the ARPANET era. The Web is one application running on that infrastructure: a system of documents connected by hyperlinks and accessed via HTTP. Email, file transfer, streaming video, and instant messaging are also internet applications, distinct from the Web. When Tim Berners-Lee invented the Web in 1989–1991, the Internet was already twenty years old.

Legacy: The Architecture of the Modern World

The connected world is the cumulative result of decisions made across four decades:

Baran’s distributed topology, designed to survive nuclear war
Cerf and Kahn’s insistence on a simple, universal protocol layer
Postel’s quiet stewardship of the naming system
Berners-Lee’s refusal to patent his invention

Each decision prioritized openness and resilience over control and revenue. The resulting infrastructure — open standards, free protocols, no central authority — is both the internet’s greatest strength and the source of its most persistent problems: spam, malware, misinformation, and surveillance all exploit the same openness that made the web possible.

The story of how personal computers connected to this infrastructure is told in The Personal Computing Explosion. The networking failures that preceded and ran alongside this success are detailed in Early Networking Failures.