History of the Internet

One File Went Out. Every Workstation Had Different Missing Pieces

In 1991, three diskless workstations did not need three copies of the same file sent from a server. RFC 1235 proposed one broadcast for all of them—and a separate repair bill for the blocks each machine missed.

Atlantic Email Had a One-Hour Target. The Relays Needed a Crew

In 1990, US and European researchers wanted campus-to-campus mail to arrive properly in under an hour. Their own meeting minutes crossed the ocean by email. RFC 1210 then listed the missing machinery: interoperable gateways, operators, directories and help for the people at either end.

IMAP3 Was a Revolt Against the Guessing Game

A mail client in 1990 could be offered an optional command without a way to ask whether the server understood it. In February 1991, RFC 1203 answered that design with a counterproposal—and an unusually blunt complaint about how its predecessor had been written.

MD4 Had Three Speed Tests and One Security Conjecture

In 1990, MD4 came to the Internet with measured throughput on three computers and a proposed job in digital signatures. The speed figures had machines behind them. The claim that its 128-bit fingerprint would resist forgery had a different label in the same document: conjecture.

Backbone Was Free. The Rural Line Still Had a Bill: NSFNET’s 1990 Argument

The Backbone Was Free. The Rural Line Still Had a Bill: NSFNET’s 1990 Argument

In 1990, a research network could ride the NSFNET backbone without a backbone invoice while a distant college still faced the cost of a long leased line. RFC 1192 asked where public money should sit when a supposedly free national route met an uneven map.

Your Internet Welcome Pack Had a Phone Number

In August 1990, the Internet gave newcomers a 24-page answer sheet. Its first useful instruction was about permission. Its next few sent readers to a paper guide, a mailbox, a telephone and a neighboring site willing to help. The network was connected; the welcome desk was scattered.

Number Was Assigned. The Door Had an Asterisk

The Number Was Assigned. The Door Had an Asterisk

In July 1990, the Internet's number book contained a tiny bouncer: an asterisk. It sat beside network numbers assigned to independent IP networks. The number had been recorded; entry into the shared Internet required another conversation. The punctuation mark knew the difference.

Some Internet Clocks Were Two Years Late. Then Came the Count

In one 1988 Internet survey, a few computers answered the time question from a calendar two years away. David Mills did what any sensible person would do with a network full of confident clocks: he counted the answers, checked the denominators, and returned in 1989 with a much bigger experiment.

Port One Hired a Receptionist. TCP Services Kept Their Old Doors

In 1988, a new TCP service could seek its own well-known port number, or it could try a receptionist. Mark Lottor's TCPMUX proposal put the receptionist at port 1. A caller connected, spoke a service name, and learned whether the requested office was open. The old doors kept their numbers. The novelty was a shared lobby for the services waiting outside.

Email Knew It Was PostScript. The Recipient Had to Find an Interpreter

The Email Knew It Was PostScript. The Recipient Had to Find an Interpreter

The letter arrived perfectly. The recipient opened it and met a wall of PostScript. In 1988, Martin Sirbu proposed giving Internet mail a small, useful label: tell the reading program what sort of document is inside. The label could summon an interpreter. Somebody still had to know whether the recipient owned one.

VMTP Lost the Answer. Should It Save a Copy or Ask the Clock Again?

The server answered the question. The network ate the reply. Now the client is back, asking for the time again. In 1988, VMTP made that small annoyance a design decision: keep a copy of the old answer, or do the work again and give a fresher one. The clock was pleased. A nonrepeatable operation would need a different arrangement.

Pentagon Gave OSI a Trial Badge and a Promotion Plan

The Pentagon Gave OSI a Trial Badge and a Promotion Plan

In July 1987, the Pentagon opened the door to a rival protocol suite and handed it a temporary badge. OSI could enter government procurements beside the established DoD protocols; promotion depended on formal approval and operational homework. The old systems, meanwhile, had an unusually good retirement clause: keep talking to them for as long as they lived.

IP Number Came with an Asterisk—and No Boarding Pass

The IP Number Came with an Asterisk—and No Boarding Pass

In 1987, an IP network could have an official number and still wait outside the Internet’s gate. The DDN Network Information Center marked independent networks with an asterisk and told their administrators to apply separately for interconnection. Its number book also carried the names of people to call, old numbers on temporary duty, and an autonomous-system list. Registration was an address; entry required another conversation.

Internet Counted to 999—and Kept 66 Empty Chairs

The Internet Counted to 999—and Kept 66 Empty Chairs

In 1987, the Internet filed its first 999 RFC numbers and left 66 chairs empty. The bibliography included protocols, meeting notes, distribution lists and surveys of who actually spoke TCP. It also printed dates that refuse to march in numeric order. Joyce Reynolds and Jon Postel made the archive useful by keeping its untidy edges in view.

One Campus, Two Federal Networks—and a Cable Nobody Needed

Barry Leiner's 1987 example put two federal budgets on one university campus. NASA-funded researchers already had a network connection. An NSF-funded colleague wanted an NSF supercomputer. The absurd solution was to buy the university a second path to a federal network; the proposed Interagency Research Internet would make the networks meet each other instead.

Gateway Said Slow Down. The Host Bought a Stopwatch.

The Gateway Said Slow Down. The Host Bought a Stopwatch.

A gateway said the line was filling up. The sender needed a number. In 1987, RFC 1016 proposed turning that warning into a per-destination pause between packets: start at 75 milliseconds, add 20 when another warning arrives, and claw the pause back one millisecond at a time. The gateway complained; the host bought a stopwatch.

File Had Nine Letters. The Network Gave It Twelve Seats.

The File Had Nine Letters. The Network Gave It Twelve Seats.

The example in Sun Microsystems' 1987 XDR memo features a file named `sillyprog`. Its name uses nine bytes. On the wire, the name occupies twelve. The extra three are zeros, and they earn their seats by telling the next field exactly where to begin.

Window Understood the Numbers. The Picture Had Other Plans.

The Window Understood the Numbers. The Picture Had Other Plans.

Two computers agreed on how to read a number. Then the picture arrived and demanded a different arrangement. The 1987 X Window System protocol put this small ambush in plain sight: the client chose byte order for the conversation, while the server chose it for image data. One window, two referees.

ACK That Asked for a Receipt

The ACK That Asked for a Receipt

The receiver had room again. The sender kept waiting. Somewhere between them, a small permission slip had gone missing. In 1987, an ISO transport implementation guide proposed a deliciously circular remedy: send the acknowledgment again, then ask the other end to acknowledge the acknowledgment. Bureaucracy had found a network cable.

Scientist Sent an Email. The Equation Didn't Fit.

The Scientist Sent an Email. The Equation Didn't Fit.

In 1987, a scientist could send a manuscript by electronic mail and still lose the equation that made the manuscript worth reading. Alan Katz found the snag in an awkward place: the mail carried text perfectly well, but the people and machines drawing a fraction had no shared language. A plus sign was cheap. The fraction needed a summit.