Hi Margaret,

I agree it is useful to consider the problem we are trying to solve,
however, my understanding has been that we have been trying to solve the
same problem that traditional site-locals were created to solve.

I've generally understood the goals of traditional site-locals were :

1) internal / local / site / friend / entity (take your pick) address
space that was independent of the global address space assigned by the
(one or more) upstream global connectivity provider(s). 

2) no external registration required for this "site local" address
space.

3) communications sessions (eg TCP sessions) using these "site local"
addresses where impervious to global renumbering events.


Traditional site-local addressing (fec0::/48 (:: /10-/48) ) met those
goals.

However, there were some limitations :

1) the "site" name and explanation in the addressing RFC predominantly
implies a geographical scope or size for this address space. While
people on the ML have said that the word "site" in "site-local" doesn't
really have geographical connotations, it is hard to forget the root
English definition of the word. In practical terms, I might build a
network which considers the whole of Australia as a site when using
site-local addressing, which sounds a bit odd.

2) when two entities using site-local addressing attempt to interconnect
or merge their networks there is a good chance of subnet addressing
collision, necessitating some amount of renumbering of parts of one or
the other entities' networks

3) because site-local addressing is duplicated, there is referral
ambiguity once multiple sites are connected together via a single
router. There were also DNS issues (Keith is the expert on this problem
:-) ).

4) any others I've probably forgotten.

5) oops, forgot about people wanting to use site-local to site-local
IPv6 NAT when merging / interconnecting networks, rather than
renumbering site-local prefixes.


Pekka Savola suggested creating near globally unique site local
addressing, by filling the /10 through /48 bits of fec0::/48 with a
pseudo random or near unique number, derived from a hash of certain
values such as MAC addresses or time of day or other reasonable values.

This doesn't seem to be a new idea, Paul Francis proposed the same thing
in the following ietf draft (worth a read, covers a lot of what has been
coming up in emails recently about GUPIs / (near) unique site local
addresses) :

http://www.join.uni-muenster.de/drafts/draft-francis-ipngwg-unique-site-local-00.txt

I've always assumed that the original goals haven't changed, just that
near globally unique site local addressing was a solution which
addressed the traditional site-local limitations.

My discussions and thoughts on things like naming / use etc have really
only been in the context of having a better "site-local" solution.

Are we happy with the existing problem definition, that (near) globally
unique site local addressing is a better solution for than traditional
site-local addressing, or have the recent discussions on near globally
unique site local addressing / GUPIs caused us to inadvertently loose
focus on our problem definition, making it necessary for us to restate
and redefine it ?

Regards,
Mark.

--------------------------------------------------------------------
IETF IPng Working Group Mailing List
IPng Home Page:                      http://playground.sun.com/ipng
FTP archive:                      ftp://playground.sun.com/pub/ipng
Direct all administrative requests to [EMAIL PROTECTED]
--------------------------------------------------------------------

Reply via email to