History of the Internet

CompuServe Offered GIF24. PNG Was Already Moving.

Unisys’s licensing clarification targeted commercial software using LZW, not people viewing images. As CompuServe proposed a free, open GIF successor, PNG was already on its own timetable—and the two projects did not end with a simple winner and loser.

Some .CA Names Could Expire Unless Their Owners Reapplied

The 2000 .ca handover did not give every old registration the same path. For one cohort, the existing registration would expire unless its holder applied through a certified registrar for that exact name again. The new registry also had to build its records from registrant submissions.

90-Day Certificate Was a Bet on Automatic Renewal

The 90-Day Certificate Was a Bet on Automatic Renewal

In November 2015, Let’s Encrypt capped its certificates at 90 days and recommended renewal at day 60. The shorter clock was meant to limit damage from a compromised key or mistaken issuance. It also put recurring renewal on software’s schedule, while leaving administrators time to recover by hand.

NetBSD 1.0 Announced Five Binary Releases. HP Needed Two Patches.

The NetBSD Project dates version 1.0 to 26 October 1994. In its 8 November announcement, it named five binary releases then available, marked the HP 9000/300 build as not ready, and separated three more ports that had source support but no complete binary distribution. The HP release page records the two patches that closed that gap.

First IMP Arrived on August 30. The 1989 Poem Moved It to Tuesday.

The First IMP Arrived on August 30. The 1989 Poem Moved It to Tuesday.

RFC 1000 places UCLA’s first ARPANET Interface Message Processor on the loading dock on Saturday, August 30, 1969. Leonard Kleinrock’s 1989 poem puts the machine at the gate one day after Labor Day. His prose history assigns that Tuesday to the host connection.

They Asked Internet Organizers for Three Paragraphs. They Sent Back 33 Pages.

In February 1991, Gary Malkin asked current members of the IAB, IESG and IRSG for three things: a biography, their education and interests, and their view of Internet growth. The 1992 successor to his first collection ran to 33 pages and carried 24 portraits, with the submitted words left intact apart from formatting.

Pretty Copy Had to Match the Plain One: RFC 2223

The Pretty Copy Had to Match the Plain One: RFC 2223

In 1997, the RFC Editor allowed PostScript to carry diagrams and a polished page, but kept ASCII as the primary version. The second file could join the RFC archive only after the editor agreed that it matched the first.

First ARPANET Login Crashed at G. The Log Kept Six Words.

The First ARPANET Login Crashed at G. The Log Kept Six Words.

At 10:30 p.m. on 29 October 1969, UCLA’s handwritten record marked a host-to-host success with six words about a conversation between two hosts. Leonard Kleinrock later recalled the first remote-login attempt reaching the letter G before the SRI host crashed. A small software repair made the next attempt work. The famous beginning of networked computing fits inside one failed command, one retry and a log line that says almost nothing about either.

CA*net Reached the North on 56 Kbps. NTnet Was Still in Beta.

By late 1994, YukonNet and NTnet had joined CA*net over leased 56 Kbps lines. The national backbone had received a T-1 upgrade; NTnet’s first dedicated-line users were still testing service.

Empty Kerberos Field That Still Crossed a Border

The Empty Kerberos Field That Still Crossed a Border

In September 1993, RFC 1510 gave Kerberos V5 a way to carry authentication across administrative realms. Its `transited` field records intermediate realms; separate ticket fields identify both ends of the exchange. A direct hop between two realms can leave that field empty. The blank is part of the design; the service receiving the ticket still has to decide whether the trust path is acceptable.

Ping Failed. The Error Was the First Good News: RFC 1574

The Ping Failed. The Error Was the First Good News: RFC 1574

An OSI network manager in 1994 could send a new echo request to a machine that did not understand it. If the machine returned an error, the answer still settled one urgent question: something at that hop had heard the probe. RFC 1574 turned that awkward reply into part of a minimum diagnostic kit, alongside a route trace and a way to inspect the device's own routing table.

Email Read Receipt Had to Ask Nicely

The Email Read Receipt Had to Ask Nicely

A sender adds one header to an email and waits for the little thrill of knowing what happened at the other end. In March 1998, RFC 2298 gave that wish a machine-readable form. It also gave the recipient’s mail program the last word. The header could ask for a receipt. The program could send one, refuse, or say nothing. Even a returned “displayed” report described what software did with a message, leaving the reader’s understanding inside the reader’s head.

Internet’s Gigabit Dream Had a Waiting Room

The Internet’s Gigabit Dream Had a Waiting Room

In April 1990, a report from an Internet Research Steering Group workshop put the gigabit network on the table. The room had plenty of ideas for moving bits faster. Then came the awkward question: how long would a program still wait for another machine to answer? A brighter fiber and a faster processor could make that wait feel longer, measured in useful work the processor might otherwise have done. The workshop's best story was about the time outside the headline speed.

Cable Was Real. The IP and OSI Maps Could Disagree: RFC 1609

The Cable Was Real. The IP and OSI Maps Could Disagree: RFC 1609

A router can sit in one rack and play different parts in different networks. In 1994, RFC 1609 tried to give those parts separate entries in an X.500 directory, all pointing back to the same equipment. The proposal's useful question still bites: when an operator asks for a network map, which network has actually been mapped?

MD5 Came with the Answers in the Back

In April 1992, Ronald Rivest published an algorithm for turning a message of any length into a 128-bit digest. He also packed the Internet memo with C source code, a test driver and the answers that driver should print. A programmer could read the recipe, compile the example and see whether a local machine produced the same fingerprints. That was a powerful way to make agreement travel. The later history of MD5 shows why matching answers and making a safe security claim are different jobs.

Please Find a Person Attached

Email had learned to carry pictures, sound and files. On 1 April 1993, two engineers offered it a new attachment: a human being. Their RFC specified the size of the box, the resolution of the scan and the unhappy fate of a laboratory rat at a full disk. The joke worked because the paperwork sounded distressingly familiar. Thirty-two years later, an official RFC still had to explain why the proposed media type stayed out of the registry.

Mail Connection Broke. The Server Asked for Byte 6,135: RFC 1845

The Mail Connection Broke. The Server Asked for Byte 6,135: RFC 1845

The broken connection in RFC 1845's 1995 example did not force its client to send the whole message again. The receiving SMTP server answered the retry with a number: `355 6135`. It had kept the first 6,135 octets of message data. The client could resume with the rest, provided both sides knew exactly which transaction and which bytes that number described.

FTP Had Six Slots. The Future Internet Brought More Luggage.

A file could be ready to move while its address could not fit into the command that told FTP where to send it. The old `PORT` command carried four decimal octets for an IPv4 host and two for a TCP port. In 1993, with several next-generation Internet designs on the table, David Piscitello proposed a larger suitcase: FOOBAR, or *FTP Operation Over Big Address Records*. Its joke of a name hid a practical question. How does an application keep working when the network address it baked into its own conversation changes shape?

One Envelope Saved the Mailing List. The Route Slips Went in the Bin.

Before an inbox could be called a fire hose, busy mailing lists had a simpler problem: sending every submission to every subscriber cost work and headers. Moderators began packing many messages into one digest. RFC 1153 wrote down the house rules in 1990, including a brisk instruction to discard the `Received` lines from each enclosed message. The collective envelope travelled efficiently. The route of each original letter became harder to reconstruct from the digest alone.

Email Was Still in Flight. X.25 Reached for the Hang-Up Switch.

The Email Was Still in Flight. X.25 Reached for the Hang-Up Switch.

A telephone call has a tidy ending: say goodbye, put down the receiver. A computer circuit can be less polite. X.25 could discard data still travelling when its virtual circuit was cleared. In 1989, RFC 1090 put SMTP directly on that circuit and discovered that the final act of sending mail was also a choreography of waiting. The sender had to hear the service say goodbye before touching the switch.