It was assembled, over sixty years, out of parts built by people who mostly were not trying to build it. A memo about surviving a nuclear strike. A theory about chopping messages into blocks. A protocol written to join two networks that already existed. A naming system invented because a single hand-maintained file had stopped scaling. A hypertext program written at a physics lab to keep track of who worked on what.
This page is the order they happened in, and what each one actually replaced. Every date here is the date the thing WORKED — not the date it was proposed, funded or announced, because those are different years and the difference is usually where the story is.
The ideas, and the jumps
Strip out the companies and the money and the lawsuits and five ideas are left, and each one had to exist before the next could. Three more are here for the moments the needle actually moved — how fast the backbone got, who came to own it, and where it physically runs.
The record
28 entries, 1948 to 2011 · 3 of them disputed
1948
Shannon publishes A Mathematical Theory of Communication and gives the bit its name, along with the result that matters most here: a message can be sent over a noisy channel with arbitrarily small error, if it is coded properly. Everything after this is engineering against that theorem.
1961
Kleinrock finishes his MIT doctoral work on message flow in communication nets — the mathematics of what happens when traffic queues rather than flows, which is what a shared network does and a telephone circuit does not.
1962
Licklider writes a set of memos at ARPA describing a globally connected set of computers through which everyone could get at data and programs from any site. He then gets the budget to fund the people who would build it, which is the rarer half.
1964
Baran publishes On Distributed Communications: break the message into blocks, give the network many redundant paths, and let each block find its own way. It was written for survivability of communications and the Air Force did not build it.
Disputed. Invented twice, independently. Davies reached the same design in Britain in 1965 and it is his word — packet — that the world uses.
1965
Davies at the National Physical Laboratory arrives at the same idea from the opposite motive — not surviving an attack but making interactive computing affordable — and names the unit. He builds the NPL network to prove it.
1965
Roberts and Marill link a TX-2 at MIT to a Q-32 in Santa Monica over a dialled telephone line. It works, and it demonstrates the problem: circuit switching is a hopeless way to carry computer traffic, which is bursty and silent by turns.
196929 October
A student at UCLA types LOGIN to a machine at SRI. The system crashes after LO. An hour later the whole word goes through. Two nodes; by December there are four — UCLA, SRI, UC Santa Barbara, Utah.
Disputed. Whether this was built to survive a nuclear war is the most persistent argument about it. Its own engineers spent decades denying it; RAND’s papers spent a decade proposing exactly that. Both are true of different documents — the ARPANET was built to share scarce computers, on a design that came from survivability research.
1969
BBN delivers the Interface Message Processor, a ruggedised Honeywell that sits between a site’s own computer and the network and speaks the network’s language so the computer does not have to. Every router since is this idea.
1971
Tomlinson picks @ to separate the user from the host, because it was the one character in the printer’s repertoire that could not appear in a name. Within two years it is most of the ARPANET’s traffic, which nobody planned for.
Disputed. "First email" depends on what you count: programs had been leaving messages in files on shared machines since the early sixties. What was new was BETWEEN machines.
1973
Nodes at NORSAR in Norway and University College London join by satellite and cable. The ARPANET is now international, which turns "how do we connect our computers" into "how do we connect our networks", the problem the next decade is spent on.
1974
Cerf and Kahn publish A Protocol for Packet Network Intercommunication: put a gateway between networks, make the hosts responsible for reliability, and let the network in the middle be as dumb and as unreliable as it likes. That last decision is why the internet scaled.
1978
The reliability half and the addressing half are separated, so a program that does not want reliability need not pay for it. Voice and video are only possible because of this split, twenty years before anyone used them.
19831 January
Every host on the ARPANET has to change protocol on the same morning or fall off the network. Some do fall off. This is the date the internet, as a thing running the protocol it still runs, actually begins.
1984
Metcalfe’s local network — designed at PARC in 1973 and standardised in 1983 — becomes how a machine reaches the wire at all. The pairing of a cheap local network with a protocol for joining networks is what puts the internet in ordinary buildings.
1983
Until this point every machine on the network kept a copy of HOSTS.TXT, a single file maintained by hand at SRI and fetched by everyone. Mockapetris replaces it with a hierarchy that delegates: a distributed, cached, case-insensitive tree that has not been redesigned since.
198515 March
symbolics.com, by a Massachusetts computer maker. Six top-level domains exist. The company that registered it is gone; the name is not.
1986
The National Science Foundation builds a 56 kbit/s backbone joining five supercomputer centres, and — crucially — requires that any campus connecting to it carry traffic for its neighbours. That condition, not the cable, is what turns a research network into an internet.
19882 November
A Cornell student’s self-replicating program reinfects machines it has already taken, and brings down an estimated ten per cent of the hosts then connected. It produces the first computer crime conviction under the 1986 Act and the first CERT.
1989March
Berners-Lee circulates Information Management: A Proposal, about the ordinary problem of a laboratory where people and machines and projects keep changing and nobody can find what anyone did. His supervisor writes "vague, but exciting" on it.
1990December
On a NeXT machine at CERN: a server, a browser that was also an editor, and the first page. HTTP, HTML and URLs all exist by the end of the month. The machine has a label on it asking people not to switch it off.
19916 August
Berners-Lee posts a summary to alt.hypertext. That is the moment the web leaves CERN — not a launch, a newsgroup message with a link in it.
199330 April
CERN releases the web’s protocols and code with no fee, for ever. Gopher, its main rival, had announced licence fees that February and lost almost everything within two years. This is the single most consequential licensing decision in the history of the medium.
1993
Andreessen and Bina’s browser at NCSA shows images inline with the text rather than in a separate window, and installs on machines ordinary people had. The web goes from about 130 sites that June to over 600 by December, and then does not stop.
1995
NSFNET is decommissioned and its traffic moves to commercial providers. The acceptable-use policy forbidding commercial traffic — which had governed the network since 1986 — goes with it.
1998
The naming and numbering functions move from a US government contract to a non-profit corporation. Twenty-five years later it is still the answer to "who runs the internet", and the answer is still unsatisfying to everyone.
1999
802.11b makes wireless local networking cheap and interoperable. The internet stops being a thing you plug into and becomes a thing a building has.
2008
Mobile devices pass desktops as a way of reaching the network in several countries. The protocol designed in 1974 for joining research networks carries it without being changed.
20113 February
IANA hands out its last blocks of the address space defined in 1981. IPv6 had been ready since 1998; thirteen years later most of the network still had not moved, which is the clearest measure there is of how hard it is to change something everybody depends on.
The people
Named for one thing each, because that is how the work was actually done — and because “father of the internet” has been printed about at least eight of them. Every link goes to a page about that person, not to a search.
Overview
The record runs from Claude Shannon’s 1948 paper, which gave information its unit, to 3 February 2011, when IANA handed out the last blocks of the IPv4 address space: twenty-eight entries in six eras — the ideas, the ARPANET, the protocols, the names, the web, and what came after. Three of them carry a dispute that is printed where it applies rather than settled quietly: who invented packet switching, what counts as the first email, and whether the ARPANET was built to survive a nuclear war.
Each entry is dated to the year the thing worked and credits the person or team behind it, and the fourteen people are listed with the one piece each is known for and a link to a page about them. The timeline and the list are written into this page as text, so they read without scripts; the filter above the timeline narrows it to one era.
The protocols in the record are still running. The routing table built on them is on Routing, the root of the naming system on The Root, the cables that carry the traffic on Undersea, and any name’s live DNS answers on Resolver.
Method
The dates are the dates things worked. ARPANET is dated to the first message between two machines, not to the 1966 plan or the 1968 contract. TCP/IP is dated to the flag day the network actually switched, not to the 1974 paper that described it. The web is dated to the first page served off a machine other than the one it was written on.
Three things on this page are genuinely disputed and are marked where they appear. Packet switching was invented twice, independently, on two continents — Baran at RAND and Davies at the NPL — and the word is Davies’. The “first email” depends on what you count, because programs had been leaving messages in files on shared machines for years before Tomlinson sent one between two. And whether the ARPANET was built to survive a nuclear war is a claim its own engineers spent decades denying and RAND’s papers spent a decade proposing — both are true of different documents.
The illustrations are drawn here. There is no photograph on this page, because the estate does not ship pictures it did not take and cannot license, and a stock image of “connectivity” would be decoration pretending to be evidence. Each plate is an SVG that draws the mechanism it names — the packet, the handshake, the hierarchy — and is worth exactly what a diagram is worth.
