History of the Internet

Ethernet Handoff Was Common. An Open IGP Was Not: RFC 1009

The Ethernet Handoff Was Common. An Open IGP Was Not: RFC 1009

In 1987, the National Science Foundation required gateways from different vendors to meet at a familiar boundary: Ethernet, with serial links as a minimum companion. The same specification admitted that the vendors still had no open interior gateway protocol in common. One layer of interconnection had a shared floor; another did not.

Archie’s Search Was Only as Fresh as Its Last FTP Listing

Archie made a scattered set of anonymous FTP archives searchable by collecting their directory listings. That small distinction shaped the service: its answer could point to a filename and a place, but the file still lived on someone else’s machine—and the index could only describe what it had last seen.

When a Usenet Cancel Failed Locally, RFC 1036 Said to Stop

A cancel message could travel through the same distribution machinery as ordinary Usenet news, yet its authority remained tied to what each host could find and verify for itself. RFC 1036 drew a sharp boundary: if a system could not cancel the named article locally, it should not pass the request to its neighbors.

License Covered the Code, Not the Protocol: Gopher’s 1993 Line

The License Covered the Code, Not the Protocol: Gopher’s 1993 Line

On 11 March 1993, the Minnesota Gopher Team opened a newsgroup post by saying it had heard enough “hysteria” and rumor to set out an official position. The phrase that later stuck was simple: commercial users would need a licence. The policy that followed was not. It divided users by what their servers did, whom they served and whether the University of Minnesota’s own software was being used. The distinction at its centre was narrower still: a licence for an implementation was not a toll on the Gopher protocol.

A Port Was Not Enough: RFC 675 Put the Network in the Socket Name

A process could choose a port that made sense on its own machine. It could not assume that the same number named the same endpoint across a network of independently run systems. RFC 675 answered that scope problem by joining a port to the address of the TCP that owned it, then describing a connection as a pair of such socket names.

Plug Had Four Pins: NORDUnet and Europe’s Protocol Choice

The Plug Had Four Pins: NORDUnet and Europe’s Protocol Choice

In May 1988, a Nordic research network asked the United States for a route into NSFNET while promising to carry several rival network services at home. Its four-pin plug became a practical answer to a harder question: could researchers share a backbone before they agreed on one protocol?

A Language Tag Was One Token, Not a Navigation Map: RFC 1766

RFC 1766 split a language tag into a primary value and hyphenated subtags, then told applications to treat the complete string as one token. Its consequential boundary was administrative: the parts helped organize a shared namespace, but did not become a universal route through content.

WAIS Let a Document Become Its Own Query: RFC 1625’s Relevance Loop

A reader could type a few seed words, add a document or even one passage, and ask WAIS to find similar material. What came back was a ranked citation, not a durable result set or the document itself. RFC 1625’s small but consequential design was to make a selected text object part of the next question.

Who Chose the Branch? TAT-8's Capacity Bargain in Internet History

TAT-8 was planned for digital voice, computer and video traffic, not for an Internet protocol. Its branch to Britain and France, supplier contracts and capacity rules show how later data networks inherited physical decisions made for a broader telecommunications system.

Publishing the Flaw Was Not Publishing the Worm: RFC 1135’s Disclosure Debate

The 1988 Internet worm left a question that was easy to flatten into “publish or conceal.” Contemporary records show a narrower choice: explain the defects and the program’s methods, distribute fixes, or release a working decompilation that lowered the effort needed to make another specimen. RFC 1135 preserved several competing answers without turning them into a common rule.

Who Chose the Address? RFC 4014 Carried RADIUS Authorization into DHCP

A successful access decision does not hand a device an address. In 2005, RFC 4014 gave a network access server a narrow way to carry selected RADIUS authorization attributes into a later DHCP exchange, joining two decisions without pretending they were one.

Routine Did Not Mean Unreviewed: RFC 6709’s Test for Protocol Extensions

In protocol design, “routine” can sound like a verdict that a change is too small to matter. RFC 6709 used it more narrowly. An extension was routine only when the base protocol and deployed implementations could treat it as opaque without harm. That says something about architectural fit; it does not say that expert review is pointless.

One Internet, Many Operators: RFC 1462’s Telephone Analogy

The 1993 FYI used a familiar phone call to explain an unfamiliar topology: users experienced one Internet service, while separate networks operated and repaired their own sections.

NAT Was a Function, Not an Interface: Why RFC 4008’s MIB Had to Change

In 2005, a management station configuring NAT through RFC 4008 had to begin with an interface index. Ten years later, the successor MIB described NAT as a logical function that could span interfaces, address realms, subscribers, and multiple instances. The change was a correction in the boundary of what a common management model could safely assume.

Frame Crossed the Pseudowire. Its Service Role Did Not: RFC 7152

The Frame Crossed the Pseudowire. Its Service Role Did Not: RFC 7152

In 2014, an IETF requirements memo exposed a small but consequential gap in Layer 2 VPNs: a provider-edge router could receive an Ethernet frame across a pseudowire without knowing whether the frame had entered the remote network through a root or a leaf attachment. RFC 7152 turned that missing context into requirements for carrying an Ethernet Tree service across a multipoint network.

No AFI Meant “Any”: RFC 4012’s Four-Family RPSL Extension

RPSLng added an optional address-family clause to multiprotocol policy. Leaving it out was not silence: RFC 4012 gave the expression the scope “any,” while the older policy attributes kept their IPv4-unicast meaning.

Controller Needed a Network Before It Could Control One: RFC 7149

The Controller Needed a Network Before It Could Control One: RFC 7149

In March 2014, an IETF memo asked service providers to look past the seductive diagram of a central controller and ask a more operational question: how does that controller find, reach and safely influence the network beneath it? RFC 7149 treated startup, customer negotiation and failure continuity as part of software-defined networking—not details left for someone else.

DHCP Carried a Stable Subscriber ID. The Standard Did Not Define Its Meaning: RFC 3993

A customer could move between access paths while a provider still wanted the same configuration decision. RFC 3993 gave the DHCP relay a place to carry a stable subscriber label, but left the label’s assignment and meaning with the provider.

Level 2 Belonged to the Session, Not the Device: RFC 7147

An iSCSI target could meet different initiators through separate sessions. RFC 7147 put the negotiated protocol level where that agreement lived: in each session’s management record, not on a device-wide capability badge.

A Global Address Could Be a Loan, Not a Host's Identity: RFC 1335

In May 1992, one proposed answer to IPv4's allocation pressure began with a quiet distinction: a machine could keep its address at home while borrowing a globally unique one only when it needed to cross the Internet.