History of the Internet

Header Stayed. The Old Host Still Dropped the Packet.

The Header Stayed. The Old Host Still Dropped the Packet.

In 1992, one address-extension proposal tried to spare the Internet a new IP header by putting extra address bits in the old header's options. The packet kept its familiar outline. A host that had not learned the new address class would still discard it.

No Voter Roll. Seven Nominators Chosen by Lot.

The IETF had a power problem in 1992. A dispute over the Internet's future addressing system turned into a harder question: who could decide which standards advanced, and who chose those decision-makers? A ballot sounded democratic until someone asked for the voter list. The IETF had no formal membership roll. RFC 1396 records a different answer: move the final standards-track decision to the IESG and give seven eligible volunteers seats on a nominating committee, drawing lots if too many offered to serve.

A Video Call Wanted 100 Milliseconds. The Map Had Other Ideas.

In 1992, a proposed Internet flow specification refused to turn a caller's preferred delay into a universal deadline. One hundred milliseconds across the Pacific and one hundred milliseconds across town had different meanings. The network needed to know what traffic a call would send and what sort of promise its path could keep.

Report Number Was Required. The Title Was Optional.

The Report Number Was Required. The Title Was Optional.

In 1992, a computer-science department could put news of a new technical report into email instead of stuffing envelopes. RFC 1357 gave the announcement a strict identity: a version, the publisher's report number, a record date and a closing identifier. The title could be left out. The report itself could still be an FTP file, a stamped-envelope request or a purchase. The little electronic record moved first; the work it advertised had its own journey.

Email Said “Withdrawn.” The Catalog Still Had to Decide.

The Email Said “Withdrawn.” The Catalog Still Had to Decide.

In 1992, a technical report could acquire a portable bibliography record by email. RFC 1357 even described how its publisher could send a later withdrawal. The message carried a complete replacement record; each receiving library still controlled what entered its own catalog.

Cable Was Shared. The Network Wasn't.

The Cable Was Shared. The Network Wasn't.

A new computer could plug into an IEEE 802.3 cable in 1985 and still fail to join the IP conversation already running there. The copper was familiar; the field that told the receiver what it was reading had changed its meaning. RFC 948 put the migration problem in plain sight: two ways to carry IP could occupy the same local wire and form two incompatible networks. Buying a compatible connector was the easy part. Getting every host to agree on the frame was the wager.

Link Was Shared. The Sales Pitch Was Bigger.

The Link Was Shared. The Sales Pitch Was Bigger.

In 1992, a network manager could sell an institution a faster connection while leaving it unable to decide who got the capacity it already bought. RFC 1346 put that incentive beside a harder engineering question: what would a guaranteed share of a shared link actually mean?

One Zero Made the Character Escape Easier to Read. It Erased the Return Trip.

In 1992, RFC 1345 offered a way to carry characters through a narrow transport alphabet by spelling them with portable mnemonics. It also offered a switch that removed the escape marker for cleaner text. The author stated the price directly: after that switch, the original character could no longer be recovered reliably.

Two Characters Saved. A Whole Link Protocol to Maintain.

On a 1200-baud telephone line, two characters per frame sound expensive. In January 1985, J. Robinson counted them, then asked the awkward question that an overhead table could not answer: who would build, debug and support a fresh reliability protocol just to save them? RFC 935 turned a small framing advantage into an argument about recoverable errors, available chips and the price of making every implementation special.

DNS Knew Who Owned the Zone. Who Could Fix the Gateway?

A gateway failed, and the domain's official contact could be the wrong person to call. In 1990, RFC 1183 gave DNS a separate record for the people expected to act. The sharp part was its admission that owning the names and repairing the machine were different jobs.

Debugger Could Stop the Gateway. Then the Gateway Could Stop Serving Everyone.

The Debugger Could Stop the Gateway. Then the Gateway Could Stop Serving Everyone.

A remote engineer can learn plenty by freezing a machine. That is a poor bargain when the machine is a working network gateway. In 1984, RFC 909 designed a loader and debugger that could reach a target across the Internet, read and write its memory, and install breakpoints. Its most revealing requirement was that inspection should not force the target to wait for every instruction from the distant host while real-time work piled up.

To Find a Route, the First Packet Had to Fail

Christian Huitema's 1992 experiment tried to move a multihomed network's gateway choice out of global routing tables and into DNS. Its prototype listened for an ICMP failure before looking up the route. The shortcut saved changes to ordinary IP routing code and put the first packet on the bill.

Switch Had Infinite Memory. The Packets Still Died.

The Switch Had Infinite Memory. The Packets Still Died.

John Nagle removed the easiest excuse for congestion: too little buffer space. His 1985 thought experiment left the switch with limitless storage and still found a way for an overloaded network to deliver nothing. The deciding rule was who got the next turn on the wire.

Seven Layers Were Not Enough. The Joke Gave Them Forty-Nine.

An Internet packet has a modest job: reach the intended recipient with its contents intact. Vint Cerf’s April 1991 reply to an ultra-low-speed research proposal offered a grander programme. Wrap the seven OSI layers inside one another until there are forty-nine, make common symbols longer to encode, and check that routing avoids the destination. A proposal that may never be transmitted was the final feature.

A Network Write Could Tell the Modem to Stop Answering

The dial-up modem was a physical box on a phone line. In 1994, its proposed management vocabulary put the answer setting, line state and reset command on the network. One writable number could make an incoming ring go unanswered.

Catalog Found the Book. Cornell’s Page Still Had a Gate

The Catalog Found the Book. Cornell’s Page Still Had a Gate

A 1994 digital-library plan made a book searchable before any patron opened its pages. The architecture kept that gap visible: the catalogue, the page files and the access decision lived in different places.

Ping Worked. NFS Timed Out. The Serial Line Had No Checksum.

Two Sun workstations could talk across a serial link in a 1991 Internet operators’ FAQ. `ping` worked. `rlogin` worked. Then an NFS mount timed out. The answer went below the file service, below IP, to a UDP checksum that one operating system had left off and the serial framing could not replace.

TMux Cut the Packet Count. The Router Lost the Port Numbers

In 1994, the Internet had a performance problem measured in packets, not bytes. TMux offered to bundle tiny terminal traffic, then found that a router enforcing TCP-port rules could no longer see what was inside.

Host Could Cancel Your Turn to Type: IBM's 5250 Telnet Interface

The Host Could Cancel Your Turn to Type: IBM's 5250 Telnet Interface

In IBM's 1991 5250 Telnet design, pressing a terminal key could start a host-directed exchange of flags, screen saves and read commands. The network carried the session. The host still decided when ordinary keyboard input could travel.

One Internet Leader Feared Big Companies. Another Expected Commercial Networks.

The Internet's 1991 leadership yearbook was supposed to introduce volunteers to a growing public. Its eighteen self-written biographies also preserved a sharper argument: what happens to a collaborative network when firms can make money running it?