[ppml] Policy Proposal 2005-8: Proposal to amend ARIN IPv6 assignment and utilisation requirement - Last Call

[Thomas Narten]

> > the /64 boundary is hardwired in a lot of places and so that is
> > contraindicated.

> it was specified NOT to be hardwired.

Right. Unfortunately, it's possible to argue both sides of this. 

Here is what I wrote on this in
http://tools.ietf.org/html?draft=draft-narten-iana-rir-ipv6-considerations

> 5.3.  On the /64 Boundary
> 
>    In theory, even more savings could be realized by reducing the size
>    of the Interface identifier, i.e., making it smaller than 64 bits.
>    While possible, this is the most difficult boundary to move in
>    practice. Considerations include:
> 
>       - Stateless address autoconfiguration [RFC 2462] assumes Interface
>         identifiers are fixed at 64 bits. Changing this would require
>         revising that specification and devising a transition plan for
>         migrating existing implementations from 64-bit identifiers to
>         something shorter.

>       - Stateless address autoconfiguration has been widely implemented,
>         with additional implementations in the pipeline. Removing those
>         implementations and replacing them with upgraded implementations
>         will take many years.
> 
>       - it is unclear how one would transition to Interface identifiers
>         of shorter length in the short term while preserving stateless
>         address autoconfiguration.
> 
>       - one transition strategy might be to disable stateless address
>         autoconfiguration completely (for generating addresses) and use
>         DHCP to assign addresses. However, client implementation of DHC
>         for address configuration is not mandatory in IPv6, and it is
>         believed that few IPv6 devices actually implement the client
>         portion of address configuration via DHC.
> 
>       - some existing IPv6 multicast standards assume that an IPv6
>         routing prefix is no more than 64-bits in length, and include
>         the 64-bit subnet prefix within an IPv6 multicast address
>         [RFC3306,RFC3956].


Actually, the text about stateless addrconf is not quite
true. Stateless addrconf can handle any prefix length, but the real
issue is that it contains the following line:


>        If the sum of the prefix length and interface identifier length
>        does not equal 128 bits, the Prefix Information option MUST be
>        ignored.  

The rules for constructing interface identifiers are specific to each
link-layer type (e.g., ethernet vs. token ring vs. ...). And in all
the individual IPv6-over-linklayer documents, the interface identifier
is specified to be 64 bits. Plus, RFC 4291 "IP Version 6 Addressing
Architecture" does say:

>    For all unicast addresses, except those that start with the binary
>    value 000, Interface IDs are required to be 64 bits long and to be
>    constructed in Modified EUI-64 format.

So, if one wants to use stateless addr conf, we've effectively wired
in /64.

Thomas