Before There Was a Phone Book (1965-1983)

Zoe · 2026-08-06 · 9 min read

The Phone Book That Runs the Internet: A History of DNS · Part 1 · 1965–1983

Why Is It Always DNS?

Why is it always DNS? There, I said it and got that out of the way. How else are you supposed to start writing about DNS? It’s an initialism that has become synonymous with an outage.

When was the last time there was a major outage and people didn’t flood Reddit or other forums blaming DNS? It’s the easy target, but very few people understand how DNS truly functions. It was a solution born out of necessity that, if handled properly, will continue working.

In this series I will go into detail about DNS, how the problem it solves started in 1965, and how it’s naturally evolved over the 60 years of its existence through its iterations from ARPANET to modern day DNS.

Three Terminals, One Idea

You read that right, the original idea for DNS came around in 1965 by a man named Bob Taylor while he was working at the Advanced Research Projects Agency, which surprise.. when read as an acronym is “ARPA”. He worked as the Director of the Information Processing Techniques Office at the time and it’s famously been explained that he had three computer terminals in his office. Mind you, in 1965 a computer terminal isn’t like a small computer you have sitting on your desk now.

Yet none of those systems could talk to each other. If he wanted to reach a colleague on the Berkeley system, he had to physically get up, walk to that terminal, and remember an entirely different set of commands and login procedures than the one two feet away. Three computers, three logins, three sets of habits, and zero ability for any of them to share so much as a single message. This was the initial spark that inspired Bob.

Bob went on to co-author a paper in 1968 with J.C.R Licklider titled “The Computer as a Communication Device”. This paper made a case for the general use of networked computers.

In that paper, it is laid out what the ambition of what would become ARPANET was.

“Today the on-line communities are separated from one another functionally as well as geographically. Each member can look only to the processing, storage and software capability of the facility upon which his community is centered. But now the move is on to interconnect the separate communities and thereby transform them into, let us call it, a supercommunity. The hope is that interconnection will make available to all the members of all the communities the programs and data resources of the entire supercommunity.” - The Computer as a Communication Device - J.C.R. Licklider and Robert W. Taylor

Meanwhile, in 1966 Larry Roberts and Thomas Merrill connected a Q-32 computer in Santa Monica to a TX-2 computer in Massachusetts over a leased phone line, thus giving early proof that time-sharing computers could talk to each other over distance, and that the phone network’s circuit-switching model wasn’t a good fit for how computers communicate. This, in short, was another validation of others seeing and experiencing the same problem that Bob laid out in his terminal example.

June 3rd 1968, Larry Roberts submitted the formal ARPANET program plan to ARPA. Miraculously, the program was officially approved within the month on June 21st. July 29th of 1968, ARPA sent out a formal Request for Quotation to 140 bidders to build the specialized network switches known as Interface Message Processors (IMP).

The first four sites chosen for this were UCLA, Stanford Research Institute, UC Santa Barbara, and the University of Utah.

It wouldn’t be until the following year in September 1969 when Bolt, Beranek and Newman (BBN), a Cambridge MA engineering firm, delivered the first IMP to UCLA. Not but a month later on October 29th, Charley Kline attempted to send the word “login” to the Stanford Research Institute (SRI) node. Incredibly, the system crashed after just the letters “l” and “o”.

A second attempt was done an hour later when the full connection was restored, and this was sent successfully, thus being the first full message sent through ARPANET.

By December 1969, all four of the original nodes were connected and included early protocols such as Network Control Program (NCP) for host-to-host communication, with Telnet and an early version of file transfer capability layered on top.

50 kbit/s leased line (×4, same spec)first message · Oct 29, 1969Host (SDS Sigma 7)IMP #1UCLAHost (SDS 940)IMP #2Stanford Research InstituteHost (PDP-10)IMP #4University of UtahHost (IBM 360/75)IMP #3UC Santa Barbara

While ARPANET was explicitly restricted to government and approved research use, this is the beginning of the predecessor to what we know to be DNS.

The Phone Book Problem

But, as you can probably guess, it wasn’t all sunshine on the ARPANET. There is a remnant of it that some IT technicians will have heard of and even messed with in modern day: HOSTS.TXT.

Every host on the ARPANET needed a human-readable name mapped to a numeric address. The solution to this was HOSTS.TXT, a single master file maintained by Elizabeth “Jake” Feinler’s team at the Stanford Research Institute’s Network Information Center.

Each machine on the network would periodically download a full copy of HOSTS.TXT to continue resolving names locally. It functioned exactly like a phone book where everyone in the country would have to update it by hand every time someone moved.

While this was manageable for the original 4 hosts, by September 1973 the ARPANET had expanded to 43 sites, and even more drastically had increased to 213 by 1981, with a new device joining on average every 20 days. This method of hand-managing a single resolution file, developed back in 1971, was no longer viable.

The downfall was structural, a crack in the foundation of how the ARPANET functioned. Every new host meant an edit to one file, and every edit meant redistributing the whole file to every other machine on a rapidly growing network. There was no delegation, no hierarchy, and one team at SRI was the sole authority and sole bottleneck. This is the exact problem DNS was invented to solve.

The Network’s Other Problem: Its Own Language

While the naming problem was quietly getting worse through the 1970s, ARPANET had a second, unrelated ceiling it was about to run into. NCP had gotten the network talking to itself, but it only worked within ARPANET. It had no way to bridge to any other network, and by the early 1970s, other networks were exactly what the Defense Department wanted to connect.

In the spring of 1973, Bob Kahn and a young Stanford professor named Vint Cerf sat down together and started sketching out something bigger: a protocol that could stitch together packet radio, packet satellite, and ARPANET into one common system, regardless of what any of them looked like underneath. They published the design in May 1974 as “A Protocol for Packet Network Intercommunication,” and called it TCP. It was later split into two pieces, TCP and IP, and together they became what we now just call TCP/IP.

Getting the whole network to actually switch over took years. In 1981, Jon Postel, by then running the show at USC’s Information Sciences Institute, published the formal transition plan for the roughly 400 hosts still running on NCP, with a hard deadline: January 1, 1983. Engineers still call it “flag day.” Every host had to be ready by that date, and the cutover was permanent — there was no falling back to NCP once the switch was thrown.

By 1983, ARPANET had a brand new language and a naming system that had been quietly breaking under its own weight for a decade. And the same small circle of ARPA-connected researchers, Jon Postel among them, was about to hand someone the job of fixing the second problem, once and for all.

That someone was a researcher named Paul Mockapetris. In November 1983, the same year flag day rewired ARPANET’s entire language, he published two documents that would quietly become the foundation for how you find anything on the internet today.

Sources