Summary

  • RFC 1897's first 6bone address plan tied a test address to a provider ASN, an existing IPv4 network, a site subnet and an interface identifier, while explicitly warning that the allocation was temporary and would be reclaimed.
  • RFC 2471 replaced the first block with 3FFE::/16 under a newer hierarchy. Registry objects, BGP4+ routes and pTLA duties made the experiment operational, but none converted test coordination into permanent production authority.
  • RFC 3701 later stopped new pTLA allocations and set 6 June 2006 as the phaseout date. Returning the blocks to IANA closed the allocation record; proving that every operational dependency had disappeared remained a separate task.

A working address could still be borrowed

The first 6bone plan did not hide its exit clause in implementation folklore. RFC 1897 called the addresses temporary, said they would be reclaimed, warned that all users would have to renumber and restricted their Internet routing to IPv6 testing. Those statements were not predictions about whether the experiment would work. They defined the authority under which it could work.

The distinction matters because a functioning resource is easy to misread. A site can configure an address, publish it in a registry, announce a route, exchange packets and run an application. Each event is evidence of use. None, by itself, changes a test allocation into property or creates an indefinite production entitlement.

RFC 1897's address layout also exposed how much contemporary structure a coordinate could carry. Alongside its experimental prefix and reserved registry identifier, it included the provider's 16-bit autonomous system number, the high 24 bits of the subscriber's routable IPv4 network, a 16-bit subnet and a 48-bit interface identifier. If an IPv4 subscriber prefix extended past 24 bits, its remaining bits entered the subnet field. The address was not merely a long random label. It compressed a provider relationship, an IPv4 location, a physical-link choice and usually a hardware-derived identity into one value.

That compression made prototype deployment convenient. It also described the future renumbering bill. Change provider, change the IPv4 context, revise the hierarchy or replace the interface assumption, and dependencies embedded in the address have to move. The warning about reclamation was therefore not bureaucratic caution placed beside the engineering. It was part of the engineering.

The architecture moved, so the test coordinate moved

The surrounding IPv6 design did not stand still. RFC 1884 supplied the early 128-bit addressing architecture, while RFC 1887 described a provider-oriented allocation model intended to carry useful topology. By 1998, RFC 2374 had introduced a different aggregatable global unicast structure, separating public and site topology and using TLA, NLA and SLA fields with larger interface identifiers.

RFC 2471 then obsoleted RFC 1897 and moved the 6bone to TLA ID 0x1FFE, producing the familiar 3FFE::/16 testing prefix. It repeated the same terms: the addresses were temporary, would be reclaimed and required future renumbering. The new NLA hierarchy was supposed to reflect 6bone transit networks and end sites, while each organization retained responsibility for its own site-level structure.

The replacement is the historical point. The first address format was not preserved merely because systems had begun to depend on it. Nor did a structurally valid old address defeat the newer architecture. Compatibility with yesterday's test plan was evidence about lineage, not authority for tomorrow's routing.

RFC 3701 later described the move from the original 5F00::/8 space to 3FFE::/16 as having occurred with little problem. That phrase is useful but bounded. It records the authors' assessment of a coordinated transition; it is not a census of every host, tunnel, DNS record, access rule or application, and it cannot prove that migration imposed no local work.

The testbed became operational without becoming permanent

The 6bone was more than an address reservation. RFC 2546 and RFC 2772 describe an operating environment using BGP4+, scoped routing policy, aggregation, filtering, DNS, a registry and delegated pTLA responsibilities. Participants maintained contact and route-policy objects, connected peers, corrected leaks and learned how IPv6 behaved outside a laboratory.

That practical seriousness did not weaken the temporary contract. It made the experiment valuable. A registry record could identify the responsible operator and intended policy. A BGP session could show that two systems exchanged reachability. An accepted announcement could prove a control-plane path. Forwarding and an application exchange required further receipts. None of those observations conferred permanent authority over the prefix.

The evidence ladder must remain intact. Definition precedes assignment. Assignment precedes configuration. Configuration precedes routing policy. A control-plane route precedes forwarding. Forwarding precedes an application result. Renumbering adds another chain: install the replacement, update DNS and tunnels, repair access rules and documentation, withdraw the old route, then observe whether stale references still matter.

Calling all of this “the address worked” erases the very thing the 6bone was built to test: whether an evolving Internet could coordinate change without treating every deployed intermediate as irrevocable.

A phaseout date converted warning into a decision

By the early 2000s, production IPv6 allocation mechanisms existed and the experimental backbone risked becoming a parallel habit. RFC 3701 set 1 January 2004 as the cutoff for new 6bone pTLA allocations and 6 June 2006 as the phaseout date. After that date, 6bone prefixes were intended not to be used on the Internet in any form, and operators could filter them. IANA was to reclaim 3FFE::/16.

The document did not promise former participants a particular production allocation. Moving out of the test space required authority from the appropriate production process, not an automatic conversion of experimental tenure. It also acknowledged a governance wrinkle: the original transition working group's charter no longer covered 6bone oversight, so the phaseout relied on the broader IETF process. That is a record of how the decision was made, not proof that every participant individually consented.

RFC 5156 later recorded both 5F00::/8 and 3FFE::/16 as returned to IANA and said they should not appear on the public Internet until reallocated. That closes the top-level allocation history. It does not prove that every obsolete literal vanished from source code, logs, DNS, ACLs or notebooks at one instant. Administrative return and operational extinction are related claims, not the same claim.

Sources