History of the Internet

DNS Reply Carried the Request’s MAC Forward: RFC 2845

The DNS Reply Carried the Request’s MAC Forward: RFC 2845

In 2000, DNS needed a way to authenticate a transaction without pretending that a shared secret was a public-key identity system. RFC 2845 made the response depend on the request’s MAC, then carried a running MAC across multi-message TCP replies. The result was a transcript with checkpoints—not encryption, zone-write permission, or proof that accepted data became live service.

Server Forgot the Lock. Recovery Began by Refusing New Ones: RFC 3010

The Server Forgot the Lock. Recovery Began by Refusing New Ones: RFC 3010

NFS version 4 turned a crash into an argument over remembered rights. RFC 3010 answered with a bounded order: identify the returning client, let old state be reclaimed, and keep new conflicts outside until that opportunity had passed.

Access Concentrator Waited for the Cookie to Return: RFC 2516

The Access Concentrator Waited for the Cookie to Return: RFC 2516

PPPoE had to let a host find one concentrator on shared Ethernet without asking every unsolicited broadcast to become a costly PPP session. RFC 2516 kept discovery stateless until a session existed, then required both ends to allocate a virtual interface. Its optional AC-Cookie was a narrow return-path check—not a user credential—and its other opaque tags belonged to different actors.

Signature Verified. The Resolver Still Asked Who Was Allowed to Sign: RFC 3008

The Signature Verified. The Resolver Still Asked Who Was Allowed to Sign: RFC 3008

A correct cryptographic result did not yet make a DNS signature authoritative. RFC 3008 gave resolvers a second question: was this the kind of key that could speak for the zone data it signed?

Sender Promised the Flow Would Shrink. The Router Policed the Promise: RFC 3006

The Sender Promised the Flow Would Shrink. The Router Policed the Promise: RFC 3006

A 48 kbit/s stream could be admitted as 33.6 kbit/s when a link could compress its headers. RFC 3006 made that discount local, conditional and enforceable against the flow that asked for it.

List Was Open. Posting Was Conditional: RFC 3005 and the Moderation Boundary of the IETF

The List Was Open. Posting Was Conditional: RFC 3005 and the Moderation Boundary of the IETF

The IETF’s broadest mailing list invited technical and procedural argument from an open community. Its 2000 charter also named who could stop a pattern of abuse—and who could hear the complaint when that power was used.

NAT Was a Border Function. RFC 2993 Showed the Work It Sent to Applications

The address translator could leave ordinary web traffic untouched and still change what the Internet meant to an application. In 2000, RFC 2993 mapped the hidden work: address-bearing protocols needed gateways, endpoint updates and local name administration wherever translation stopped being invisible.

Namespace Was Global. Resolution Was Not: RFC 3001, RFC 3061, and the Authority Hidden in `urn:oid`

The Namespace Was Global. Resolution Was Not: RFC 3001, RFC 3061, and the Authority Hidden in `urn:oid`

A name can travel everywhere without telling anyone where to look. When the IETF carried object identifiers into URN syntax in 2000, it preserved a delegated system of authority—and left resolution, validation and trust outside the string.

Equal Cost Did Not Make the Paths Interchangeable: RFC 2991 and the Price of Stable Multipath

Two routes can tie in the routing table and still disagree about packet size, delay and what a probe can see. In 2000, RFC 2991 made the router’s choice among those paths a question of how much disruption a network was willing to accept.

Header Said 33 Seconds. The Message Had Not Yet Been Played: RFC 2424

The Header Said 33 Seconds. The Message Had Not Yet Been Played: RFC 2424

An unopened message could already announce how long its voice clip should last. RFC 2424 made that preview portable, while stating why the number was not the same thing as a measurement of the media behind it.

Network Could Decline Without Saying No: RFC 2990 and the Missing QoS Feedback Loop

The Network Could Decline Without Saying No: RFC 2990 and the Missing QoS Feedback Loop

In November 2000, RFC 2990 described an awkward bargain at the centre of Internet quality of service. A network could remember each application's reservation and return a precise answer, but carrying that state through a high-speed core was expensive. Or it could aggregate traffic into a few classes and scale, while leaving the application to discover that its requested treatment had not arrived. The missing piece was not another packet mark. It was a return path for resource truth, failure and measurable delivery.

Checklist Graded Capabilities, Not Packets: RFC 2989

The Checklist Graded Capabilities, Not Packets: RFC 2989

In November 2000, RFC 2989 assembled a formidable checklist for network-access AAA protocols. It asked whether a candidate could scale, fail over, protect objects, carry EAP, authorize, reconcile state and deliver accounting records. Then it inserted the sentence that kept the table honest: requiring a protocol to support confidentiality was not the same as requiring every protocol message to be encrypted. The checklist graded what a design could express and require of implementations. It did not certify what any packet, proxy, server or subscriber actually experienced.

codec agreed. The nibble did not: RFC 2422

The codec agreed. The nibble did not: RFC 2422

G.726 could define the sound-coding algorithm and still leave two systems to disagree about a byte. RFC 2422 made the missing choice explicit: which four-bit codeword goes into which half of an octet, and what to do when the last pair is incomplete.

Tags Looked the Same. One Was a Limit, the Other a Preference: RFC 2987

The Tags Looked the Same. One Was a Limit, the Other a Preference: RFC 2987

In November 2000, RFC 2987 placed `charset` and `language` inside the same media-feature grammar. Both used registered tokens, equality comparison and optional quality weights. Yet the document refused the most convenient inference: identical syntax did not make the two claims equivalent. A charset was usually a machine limit. A language was usually a human preference. The history matters because systems still fail when they let the shape of a declaration decide what the declaration is allowed to prove.

MIB Moved Branches; MARS Did Not: RFC 2417

The MIB Moved Branches; MARS Did Not: RFC 2417

In 1998, the Multicast Address Resolution Service acquired a new place in the management-information tree. RFC 2417 corrected the root assigned to its MIB—but a corrected address for management data is not evidence that multicast traffic, ATM circuits, or deployed agents moved with it.

ACK That Could Not Name the Copy: How Karn Stopped TCP Learning the Wrong RTT

The ACK That Could Not Name the Copy: How Karn Stopped TCP Learning the Wrong RTT

A TCP acknowledgment can be completely true and still answer the wrong question. After a timeout, the same bytes may have crossed the network twice. The returning ACK proves that the receiver advanced, but not which copy made it advance. Karn's algorithm made that missing label operational: do not teach the timer with evidence whose cause cannot be identified.

Fourth Packet Was Certain to Drop. That Still Wasn’t the Result.

The Fourth Packet Was Certain to Drop. That Still Wasn’t the Result.

In 1998, a small TCP simulation put an awkward objection on the table: if a modem-facing router can hold only three packets, a four-packet opening flight must lose one. The experiment accepted that loss as certain, then asked a harder question—what happened to the connection when the ordinary one-packet start reached the same bottleneck?

Where the Power Sat: The Instrument That Returned EUR 11,035,874 to RIPE NCC Members

When the RIPE NCC's 2020 General Meeting approved returning that year's financial surplus to members, the operative authority was not a board discretion but a contract clause and a single recorded resolution. This examination reads the deciding instruments' own words, the recorded vote, and the two distinct paths a member could have used to challenge the result — then follows the mechanism through the 2025 billing cycle.

When the Standard Outlives Its Authors: The Orphaned Stewardship of RFC 2350

Every network operator who has ever drafted a computer security incident response plan has, knowingly or not, touched the work of a working group that stopped existing nearly twenty-five years ago. RFC 2350 — 'Expectations for Computer Security Incident Response', published as BCP 21 on 1 June 1998 — remains the Internet's current best practice for what an organisation may expect of a CSIRT. The body that produced it, the IETF working group grip ('G & R for Security Incident Processing'), started in February 1995 and is listed as concluded in December 2001. Nobody has replaced it. Nobody, in any formal sense, owns its product today. This article follows that accountability chain from authorship through abandonment to the documentary record that is now the only remaining guarantor.