El 07/08/2006, a las 16:20, Julien Laganier escribió:
Hi Dave, and other folks knowledgeable about IPv٦ addressing model,
[EMAIL PROTECTED] CCed, please reply only to [EMAIL PROTECTED]
While working on the issues of which addressing model to use in
NetLMM, I think I got confused with issues involved the IPv٦
addressing model (or its assumptions.)
I would therefore like to ask you if a potential NetLMM addressing
model (per-MN subnet prefix [RFC٣٣١٤]) would, in some situations,
conflict with the IP addressing model.
Background
----------
Dave's draft on issues involved with multilink subnets
[draft-thaler-intarea-multilink-subnet-issues] list some assumptions
of the IP addressing model, but there might be other that are not
specific to multilink subnets. I'd therefore like to ask you about
possible conflicts between IPv٦ and RFC٣٣١٤ addressing model.
We are considering the situation of mobile nodes (MNs) attached to a
NetLMM domain. The NetLMM domain span multiple access links, each
served by a given access router (AR). A MN attaches to one link, and
hence to one AR.
( NetLMM domain )
/ | | | \
AR AR AR AR AR
/ \ \ / \ \
MN MN MN MN MN MN
If all of the MNs in the domain uses a common subnet prefix we
obviously end-up with a multilink subnet, which is problematic as
described in Dave's draft. Now a simple way to avoid multilink subnet
issues is to use a per-MN subnet prefix, as in the IETF
recommendation to ٣GPP [RFC٣٣١٤]. That way, each of the MN moves has
a different prefix and hence none of the prefix spans more than one
link, thus avoiding multilink subnet issues.
Issue
-----
Such model has however raised a question, which is orthogonal to
multi-link subnets issues. RFC٣٣١٤ was proposed for use in a scenario
where the link between the MN and its AR is point-to-point. Now if we
consider a broadcast/multicast capable link-layer technology such as
Ethernet, then we would have a situation in which, on a given link,
the broadcast domain and hence the link-local scope are larger than
the any of the per-MN subnet prefixes scope (as illustrated below
when ٣ MNs A, B and C are connected to one such link served by one AR
R).
A subnet prefix scope: -R-------A------------------
B subnet prefix scope: -R---------------B----------
C subnet prefix scope: -R------------------------C-
link-local scope: -R-------A-------B--------C-
L٢ broadcast scope: -R-------A-------B--------C-
Do you think that this situation (i.e. link-local scope larger than
subnet prefix scope) would conflict with the IPv٦ addressing model,
or any of its assumptions?
How do the nodes configure their global addresses?
I mean are they using stateless autconf?
If they do, how do you prevent the nodes from configuring one address
per prefix the get in the RADV? wouldn't some of the nodes end up with
multiple addresses from different prefixes? breaking the goal of having
one node per prefix?
However, i guess this is not a problem for your netlmm case, since i
guess there is no problem that there are multiple mobile nodes per
prefix, you are just trying to avoid multilink subnets... right?
regards, marcelo
Many thanks in advance. Best regards,
--julien
_______________________________________________
Int-area mailing list
[email protected]
https://www١.ietf.org/mailman/listinfo/int-area
_______________________________________________
Int-area mailing list
[email protected]
https://www1.ietf.org/mailman/listinfo/int-area
