Date: Thu, 7 Mar 2002 19:53:30 +0100
From: "Hesham Soliman (ERA)" <[EMAIL PROTECTED]>
Message-ID:
<[EMAIL PROTECTED]>
| => It means something like this:
|
| A B
| GGSN ------- Host----<optional>--- whatever you want tp put
|
| GGSNs MUST NOT use the /64 to configure addresses
| on their own interfaces.
This is exactly why there needs to be an IPv6 over whatever draft
(in the IETF domain). Must of the (DAD related) part of this
discussion has all been noise, as it has all been missing the point.
In particular:
| But no DAD is needed on link A.
|
| Makes sense?
No, this is what is causing the confusion. It isn't that DAD isn't
needed on A, it is that (as you are proposing) doing it there would
simply be wrong, as no address is to be configured there. One never
does DAD on a link other than the one where the address is to be
configured - and your proposal is to have the A link above be unnumbered,
with the prefix assigned to the host to use on its other link(s) (perhaps
purely on its internal loopback interface, perhaps B if it exists).
It is important to get the model right - if you even allow yourself to
think that what the host is configuring the prefix on the 3GPP link,
then it makes no sense at all to restrict the GGSN from also configuring
an address on the link - the /64 has plenty of address in it, and if it
is configured as a /64 (not two /65's or similar) then it really applies only
to one link (joining it would be via a bridge).
So, what's needed is an I-D that sets out exactly how the 3GPP link is to
be used, and justifies that. Note that justifying the "MUST NOT" I quoted
above from your message might not be easy - telling operators how they're to
be permitted to configure their links is something the IETF does quite
rarely. In particular, unnumbered links are liked by some, and hated by
others, so you're going to need to have some very strong reasons to require
that mode of operation as the only acceptable mode.
If you do manage that, then you don't need to say that DAD isn't required,
as doing DAD in the circumstances above would simply be crazy - it isn't
a case of some kind of exception to the general DAD rule, it is a case where
doing it would be completely contrary to the rest of the architecture.
So this doesn't even get to reaching the case that has been discussed of
some link types where it might be possible to assign addresses without DAD.
On the other hand, there are still link local addresses to be assigned on
link A, and there there's no possible justification for the GGSN not to
assign itself one on the link, nor for the host (or whatever it turns out to
be, "node" is a better word, many people believe that "host" excludes
"router") either. Then the question of whether DAD is required for the
link local addresses becomes relevant, and needs to be specified.
kre
--------------------------------------------------------------------
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]
--------------------------------------------------------------------
