[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Draft on Globally Unique IPv6 Local Unicast Addresses



    Date:        Tue, 13 May 2003 07:51:59 -0400
    From:        Margaret Wasserman <mrw@windriver.com>
    Message-ID:  <5.1.0.14.2.20030513073116.048b7c58@mail.windriver.com>

  | At this point, I think we're really having an argument about what
  | "scope" means...

This has all been most amusing...

  | I think of "scoped" addresses as unreachable, local and ambiguous,

That's an odd definition, which seems designed more such that LLs and SLs
(those kinds of addresses) are the only ones that get the label attached,
as if "scope" is a pejorative annotation, and must be avoided at all costs
just because it is evil (regardless of what it implies).

  | It seems to me, though, that they may be _scoped_ by your
  | definition and Tony's...  Could you provide a definition?

For me, a scoped address is anything which doesn't work similarly at
any two points on the connected internet (similarly to allow differences
related to RTT, packet loss rates, etc, to be ignored - but that's all that
should be ignored).   Actually all addresses are scoped - the "global"
addresses we talk of are scoped to our Internet - someone setting up a
rival internet could use all the same addresses in their own "global" scope.

But I'm not sure this matters.

  | However, Bob's proposal defines addresses that are unique throughout
  | the Internet.

That lasted all of a day...   The proposals to allow people to invent their
own random numbers rather than paying 10 Eur to some bureaucracy (along with
a delay of probably several days) took that long to appear - with the
(perfectly correct) rationalisation that allow it or not, people will do it,
so there may as well at least be a way to formalise it.

Brian contended that (some) people will want addresses that are known unique,
but without some kind of enforcement policy that's impossible, and Bob's
proposal has exactly none of that - the addresses are local and private,
which means that no-one can know what addresses I'm using (unless I tell
them) and so there's no way they can know that I just happened to pick the
same number they did.

The only way to enforce uniqueness is to make the addresses fail to work in
some respect that matters (IP6.ARPA doesn't matter, so it doesn't help).
The only way that matters for addresses really is routing, so for any address
that isn't globally routed, ambiguity will be a fact of life.   Accept it.

Christian used the birthday paradox to assume that 41 bits of random number
would not be enough - but while I am no statistician, I suspect that he
got the domain wrong - it doesn't matter in the slightest what the probability
is of a clash of identifiers with everyone else in the world, they don't
see my ID and I don't see theirs - the set of sites (organisations if "site"
is another bad word) with whom I want to have a high probability of no
duplications is the ones with whom I want to share these addresses, and
that set will usually be numbered in the hundreds, perhaps thousands, but
certainly not billions.

In any case, after all this, it is interesting to see that what we're left
with is exactly Paul Francis NUSLA proposal (from a bit over 2 years ago I
think, or was it 3?) except perhaps with the prefix changed.

Truly nothing new...  (and yes, I know the draft says that, this is not
a criticism).

kre

ps: for what it is worth, I have no problems with this proposal, except the
silly idea that anyone with half a brain would pay money to have a random
number allocated.   It retains the parts of site local that are important.
Unfortunately (for Jim), it won't allow the scoping in his implementation
to be removed, that's still needed for link locals - the only simplification
this will allow (to anything) is to avoid the need for scope-aware routing.
I'd also have no problem if the prefix were a /10 instead of /7, that's
big enough.

pps: No way could the ISOC just be handed the right to administer this space
if it were to be administered - there's enormous revenue at stake here,
and Govts all over the place are going to want to try and get it into their
domain, so they get the benefits - India and China certainly have good claims,
why should their (combined) > 2B population all be sending revenue to yet
another US based organisation?

--------------------------------------------------------------------
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 majordomo@sunroof.eng.sun.com
--------------------------------------------------------------------