History of the Internet

NTP Stopped Counting Missed Calls and Checked the Line

Network Time Protocol had a practical way to decide whether a distant clock peer was still answering: remember the replies to the last eight exchanges. In a 1990 experiment, Jon Crowcroft and Julian Onions put NTP over OSI Remote Operations and a connection-oriented service such as X.25. A live connection supplied a different answer to the first question—whether the peer was reachable. The clocks still had to exchange timing data. The old eight-slot attendance sheet simply lost one of its jobs.

Character Was There. The Screen Was Told Not to Invent It: RFC 1502

The Character Was There. The Screen Was Told Not to Invent It: RFC 1502

An international email could arrive intact and still lie to its reader. In 1993, RFC 1502 gave X.400 GeneralText a blunt rule for that gap: if a terminal could not show a character, it had to show that something was missing. Printing a plausible local character from the same byte position was the forbidden shortcut.

Every Keystroke Bought a Packet. The Form Fought Back.

Type one character, send a packet carrying TCP/IP headers, wait for the host to echo another packet back. That was the network conversation the authors of RFC 1043 wanted to quiet. Their 1988 Data Entry Terminal option for TELNET let a remote application build a full-screen form, let the user edit suitable fields at the terminal, and return a compact response. The trick was to agree first on what both ends could actually do.

A Tilde Gave Chinese Text a Seven-Bit Side Door

Some old mail paths had room for seven bits per character; Chinese text lived on machines with incompatible eight-bit habits. HZ answered with a bit of punctuation that did more work than its size suggested. In the format published as RFC 1843 in 1995, a tilde could announce a switch into two-byte Chinese text, bring the reader back to ASCII, print a literal tilde or fold a physical line. The trick lay in separating a tiny common decoding rule from the conveniences chosen by the sender.

Type a Host Name. The Manager Builds a Read-Only Identity: RFC 1503

In 1993, an SNMPv2 manager could make a user’s life easier by doing a surprising amount of administration behind one typed name. RFC 1503 proposed a resolver that could turn an unfamiliar host into temporary parties, a context and an access-control entry. The keyboard shortcut was also a decision about how the manager would speak to that host.

Mac File Had Two Forks. Email Brought One Envelope.

The Mac File Had Two Forks. Email Brought One Envelope.

In 1994, sending a Macintosh file by email could be like mailing a photograph with its frame, the label on the back, and a tiny instruction manual glued to the glass. A recipient on another computer might want the photograph alone. A Mac recipient wanted the rest of the file too. RFC 1740's answer was AppleDouble: put the Mac-specific luggage in one MIME part and the ordinary data in another. The envelope stayed intact; the contents became easier to use.

IAB Wanted a DNS Deadline. DISA Sent It to Study

The IAB Wanted a DNS Deadline. DISA Sent It to Study

The reply arrived before the requested date. The decision did not arrive with it. In 1992, the Internet Architecture Board pressed the Defense Information Systems Agency to move lingering host-table users toward DNS; DISA promised to examine the technical and economic costs. RFC 1401 preserved the exchange, including the gap between answering a letter and changing a network.

SMTP Told the Pager “OK.” Which OK?

In 1994, the Internet was asked to page people. Email already knew how to say “250 OK,” but a pager made that little green light dangerously vague. Was the message sitting in a mail queue, accepted by a radio network, or actually sent over the air? Marshall Rose’s RFC 1703 turned those into different promises, depending on the SMTP command at the start of the conversation. The difference mattered most when someone wanted the page now, rather than whenever the mail relays got around to it.

TCP Held the Second Record Hostage

The database asked for four records. One was late, another had already arrived, and the user would happily have read the second one first. TCP kept it waiting anyway. In 1994, three researchers proposed telling transport exactly which records must precede which others. The IESG declined to recommend their TCP modifications, while finding the ordering question worth keeping. A rejected mechanism left behind a sharp way to ask what “in order” actually costs.

Packet Had an IP Address. X.25 Wanted Another Number: RFC 1461

The Packet Had an IP Address. X.25 Wanted Another Number: RFC 1461

An IP packet could carry a perfectly good destination and still leave an X.25 router unable to place the call. In May 1993, RFC 1461 put that missing step on the management console: find the peer's X.121 number, or record exactly where the lookup failed.

Internet's New Media Types Took a Number and Waited Two Weeks

The Internet's New Media Types Took a Number and Waited Two Weeks

In 1994, naming a new kind of Internet content involved a public mailing list, a two-week wait and a trip to IANA. The form looked modest: type, subtype, parameters, encoding, known risks and a contact. Yet the queue solved a large coordination problem. A common name could follow a format from email into other applications while each program chose which formats it could actually open.

Five Reviews, Then the Paper Vanished: RFC 1438's Process Joke

On 1 April 1993, an Informational RFC invented a standards track for documents so empty that reviewers approved them by falling asleep. The punchline was not the nap. It was a committee process that could bestow authority on a result with no useful content.

One Korean Email Needed Two Passports

In 1993, a Korean email could arrive with a perfectly readable subject and a body full of wreckage—or the other way around. The two parts had crossed the same network but carried different encoding papers. RFC 1557 gave the body a seven-bit switching scheme and sent Korean words in the header through a separate MIME wrapper. The border ran straight through the message.

Directory Had Servers. Who Would Fill It?

The Directory Had Servers. Who Would Fill It?

An Internet-wide directory sounded like a machine problem in 1993. RFC 1430 put the expensive work elsewhere: mapping messy institutional records into one shared structure, keeping them accurate and paying for the services that connected them.

IPv4 Pool Was Free. The Future Title Office Was Up for Grabs.

The IPv4 Pool Was Free. The Future Title Office Was Up for Grabs.

In December 1994, Geoff Huston looked at an address allocation system that charged nothing for the scarce thing it handed out. His RFC imagined what would happen when the free pool could no longer meet every request: holders might sell blocks, large providers might dominate the trade, and whoever kept the transfer ledger might decide whose claim looked legitimate. The most valuable desk in this story was not necessarily the one issuing an address. It was the one a buyer would have to trust afterward.

Millions of Messages, One Copy on Busy Links. The Map Had Two Months to Live.

In 1993, BITNET's LISTSERV distribution service claimed 2–6 million messages a day and tried to cross busy links with one copy before fanning out near recipients. Its speed rested on an unglamorous condition: the operator's master-node map had to be less than two months old.

CoVis Gave Students Full Internet Accounts. It Banned Anonymity.

At two Chicago-area high schools, a 1990s science project let students hold their own Internet accounts. They could write directly to scientists and be contacted by strangers online without a teacher relaying every message. The same project told them to use their real names, keep their passwords to themselves and answer for everything traced to their accounts. RFC 1746 preserved the policy. It reads like the moment a school opened the network door, then had to decide which parts of the old classroom followed students through it.

Standards List Had an Expiry Date. The IAB Could Send a Protocol Back.

The Standards List Had an Expiry Date. The IAB Could Send a Protocol Back.

In March 1993, the official Internet protocol standards list told readers to stop using that edition after 31 July. It also showed why a proposal did not become a standard merely by surviving an IETF meeting: the IESG had to recommend advancement, the IAB had to ratify it, and working implementations had to support the claim.

Cook Gave the Public Network Fifteen Governors. Congress Gave It None.

In 1993, Gordon Cook drew a board for a national research and education network that Congress had never commissioned. Its fifteen seats would put schoolchildren's advocates, librarians, disabled users, carriers, manufacturers and federal agencies in the same room. The seating plan was also a budget argument. A faster backbone could take the money; so could a connection for an unserved school, or training for people at a school already online. Somebody would choose. Cook wanted that somebody to answer to more than the network's existing research patrons.

Directory Put “Internet” at the Top. Only Countries Had a Registrar.

The Directory Put “Internet” at the Top. Only Countries Had a Registrar.

The 1993 X.500 directory pilot gave “Internet” a place near the root of its naming tree and placed the DNS subtree beneath it. Its own guidelines called that entry an informal adoption. The only top-level entries with a recognised registration authority were countries.