History of the Internet

Route the Sender Tried to Choose: How IPv4 Source Routing Lost Its Default Trust

The Route the Sender Tried to Choose: How IPv4 Source Routing Lost Its Default Trust

IPv4 once carried a small, mutable itinerary inside the packet itself. The history of source routing is the history of how that sender-supplied path changed from a protocol feature into an explicitly configured exception.

Peer Returned a DNS Address. It Had Not Proved Names Would Resolve: RFC 1877

The Peer Returned a DNS Address. It Had Not Proved Names Would Resolve: RFC 1877

The useful answer in a 1995 dial-up exchange could begin with a deliberately useless address. A PPP client placed zero in an IPCP name-server option precisely so the remote peer would reject the value and suggest another. That suggestion could become negotiated configuration only after a new request and an exact acknowledgement. Even then, the evidence ended at the link's control plane. RFC 1877 could arrange where a resolver would try to send a question; it could not prove that the server was reachable, responsive, authoritative, trustworthy or used by the application.

Address That Meant Everyone: How IPv4 Broadcasts Chose Ones Over Zeroes

The Address That Meant Everyone: How IPv4 Broadcasts Chose Ones Over Zeroes

IPv4 did not gain broadcasting through a new flag. It gained it by teaching hosts and gateways that some destination values did not name one machine at all. A field of ones came to mean everyone within a scope, while receivers carried an awkward obligation to understand the earlier zero convention too.

Bit That Made Routers Report Their Limits: IPv4 DF and Path MTU Discovery

The Bit That Made Routers Report Their Limits: IPv4 DF and Path MTU Discovery

One IPv4 flag began as a prohibition: do not split this datagram. Path MTU Discovery turned that refusal into a measurement loop. An oversized packet would be discarded, a router would report the narrow link, and the source would learn what the path could carry—provided the error made it home.

Readable Without a Parser Did Not Mean Safe to Parse: RFC 1874

In 1995, one MIME label tried to preserve two truths about the same structured document. A person without SGML software might still read its general meaning; a machine with SGML software might interpret far more than the visible lines. RFC 1874 called the first case `text/sgml`, but it did not call the content inert. Decoding, finding a separate boot map, parsing instructions and permitting system-level commands were later decisions. The label offered a fallback. It did not issue a safety receipt.

Departure Packet Asked Everyone to Forget. It Could Not Prove They Did: RFC 1868

The Departure Packet Asked Everyone to Forget. It Could Not Prove They Did: RFC 1868

In 1995, a dial-in user could leave one communications server and return through another while the local network kept sending packets to the first. RFC 1868 proposed a sixteen-byte correction: broadcast an unusual ARP Reply that told every peer to delete the departed address from its cache. The economy was elegant. The evidence was not. The broadcast recorded an instruction from one node; it did not record which peers heard it, which implementations understood a zero-length hardware address, which caches changed, or whether a fresh path began to carry traffic. UNARP shortened stale belief without turning a shared Ethernet into a synchronized ledger.

Message Reached the Partner’s System. The Transaction Had Not Been Accepted: RFC 1865

The Message Reached the Partner’s System. The Transaction Had Not Been Accepted: RFC 1865

A purchase order can cross the Internet without crossing the final boundary that matters. In 1996, RFC 1865 described a dedicated SMTP connection delivering EDI directly to a trading partner’s system and called that delivery assured. The phrase was useful at the transport layer and dangerous everywhere else. MIME could identify the business object; SMTP could place it at the partner; a later signed receipt could bind message identity, content integrity and recipient response. None of those records alone said that the partner’s business application accepted the order, that a human understood it, or that a contract had taken effect. Internet EDI did not abolish the trading-partner agreement. It made the agreement’s technical and institutional seams easier to see.

Shortest Timer Usually Won. The Client Still Had to Absorb the Race: RFC 1863

The Shortest Timer Usually Won. The Client Still Had to Absorb the Race: RFC 1863

Redundancy sounds decisive only from a distance. Put two route servers beside the same clients, and it appears that either can replace the other. RFC 1863 described the harder machinery: each server kept its own picture of who was already being informed, waited for a load-ranked timer, checked again and only then began sending. The shortest delay made one server the probable winner. It did not eliminate simultaneous winners. The client still had to tolerate duplicates, and the session timer had to leave enough time for failure detection and reassignment. A decade later, the IETF recorded that this particular design had never become running code.

Address Was Supposed to Declare the Charging Rule Before Contact: RFC 1681

The Address Was Supposed to Declare the Charging Rule Before Contact: RFC 1681

In 1994, one IPng paper imagined a Gopher server redirecting a caller to a paid destination before any useful warning could appear. Its answer was to make the destination address speak first: some bits would identify who paid, or point to a charging algorithm. RFC 1681 exposed a problem that remains sharper than its proposed mechanism. A price signal can arrive before contact without proving who authorized the cost, what was delivered or whether the resulting bill is valid.

Index Could Refuse the Question Before Retrieval: RFC 1862

The Index Could Refuse the Question Before Retrieval: RFC 1862

A search box appears to sit outside the document it finds. The object keeps its permissions; the index merely points. In 1994, an IAB workshop saw why that division was incomplete. A directory could expose the existence of a record before the object server ever had a chance to deny access. RFC 1862 therefore kept enforcement at the object while asking indices to carry enough access information to refuse a question earlier. Its lesson is precise: discovery is a separate disclosure surface, and a result proves only what one index chose and was permitted to reveal.

Test Measured the Reserve by Spending It: RFC 1628

The Test Measured the Reserve by Spending It: RFC 1628

A UPS exists to preserve power when ordinary supply disappears. RFC 1628 defined a diagnostic that learned more about that reserve by deliberately drawing it down. The result could improve confidence in battery replacement and run time, but the test left the protected load with less protection until recharge. Better evidence was not free: it changed the resource being measured.

Page Was Delivered. The Person Had Not Read It: RFC 1861

The Page Was Delivered. The Person Had Not Read It: RFC 1861

In 1995, a pager protocol confronted a problem that still defeats much newer systems: the word “delivered” can end an inquiry too early. RFC 1861 made a gateway report not one success but a sequence of bounded states. A request could be accepted, held in a queue, transmitted to a field unit, viewed by its subscriber, answered and finally closed. Each receipt belonged to a different observer. By refusing to let machine delivery impersonate human attention, the protocol produced a small but durable lesson in how network records earn trust.

Switch Named the Service. The Endpoint Had Not Opened It: RFC 1618

The Switch Named the Service. The Endpoint Had Not Opened It: RFC 1618

An ISDN call could arrive at the number reserved for PPP and still meet a closed service. RFC 1618 made the order explicit: the switch could identify which doorway had been called, but only a human or local program could authorize PPP to use it. Without that administrative `Open`, the receiver had to reject the call before accepting the circuit.

Etiquette RFC Assigned Responsibility. It Did Not Create a Global Referee

The Etiquette RFC Assigned Responsibility. It Did Not Create a Global Referee

The Internet wrote down its manners in 1995 and immediately limited what the document could mean. RFC 1855 was Informational, not an Internet Standard. It offered a minimum that organizations could adapt, not a universal code that the IETF would enforce. That restraint was more than a disclaimer. The guide’s architecture divided conduct among the person sending a message, the administrator running a site, the moderator tending a group, the operator paying for a service and the local authority able to investigate a complaint. Netiquette became useful precisely because publication did not pretend to create a sovereign.

Channel Number Was Not the Circuit: RFC 1613 and X.25 over TCP

The Channel Number Was Not the Circuit: RFC 1613 and X.25 over TCP

Two calls could arrive with the same twelve-bit channel number and still be different calls. RFC 1613 solved that 1994 X.25-over-TCP problem by refusing to make the enclosed number carry more identity than its local interface allowed. The TCP stream, the logical XOT interface and the outgoing mapping completed the circuit. The field alone did not.

Permission That Survived Deletion: How IMAP Separated ACL Entries from Effective Rights

The Permission That Survived Deletion: How IMAP Separated ACL Entries from Effective Rights

The administrator removed Fred from the mailbox list. Fred could still write to it. Nothing supernatural had happened, and the server did not need to be broken: Fred might also belong to a group, or receive the same authority through `anyone`. In the access-control language that IMAP began standardising in 1997, deleting one visible rule was not the same operation as revoking the result. The protocol had to make that distinction observable without dictating every server’s local identity system.

Address That Had to Be Empty: How SMTP Stopped Errors from Mailing Errors

The Address That Had to Be Empty: How SMTP Stopped Errors from Mailing Errors

The most important address in SMTP is sometimes no address at all. A delivery failure has to travel back to the system responsible for the original message, yet a failure to deliver that failure must not produce another failure notice, and another after that. `MAIL FROM:<>` closes the recursive edge. The empty reverse-path does not erase accountability; it moves the last problem from an unbounded remote conversation into a bounded local duty for logs, queues and the postmaster.

Quarter-Hour Had Ended. Its Counters Had Not: RFC 1595

The Quarter-Hour Had Ended. Its Counters Had Not: RFC 1595

A performance window can be over without being final. RFC 1595 made that awkward truth visible in 1994: if ten severely errored seconds straddled a SONET/SDH quarter-hour boundary, the agent could later move seconds in the completed interval from error counters into unavailable time. Its successors offered a choice—revise the early answer or report ten seconds late. Either way, a counter needed a layer, an end, a threshold, a window and a read time before it could be treated as evidence.

Number That Stopped Naming Every Packet: How IPv4 Narrowed the Identification Field

The Number That Stopped Naming Every Packet: How IPv4 Narrowed the Identification Field

IPv4 put a 16-bit Identification value in every header so a receiver could reunite fragments of one datagram. Over time, the value was treated as a more general packet identity—even when fragmentation was impossible. Link speeds then outran the number space, implementations exposed activity through counters, and middleboxes learned to depend on meaning that reassembly did not need. RFC 6864 repaired the field by drawing a boundary: identity remains strict where fragments may exist; an atomic datagram's number means nothing.

Address That Recruited a Subnet: Why Routers Stopped Forwarding Directed Broadcasts

The Address That Recruited a Subnet: Why Routers Stopped Forwarding Directed Broadcasts

One IPv4 datagram could cross the Internet as an ordinary routed packet, reach the router responsible for a remote network and become a local broadcast there. With a victim's address forged as the source, that last-hop conversion could turn willing hosts on someone else's subnet into a reply engine. The durable repair was not to abolish broadcast. It was to stop treating remote fan-out as the router's automatic duty.