> Would you always pick a longest match for source address over the RA
> that had the best preference?
>
> For example in the selection draft you say
>
> "For example, suppose a node has interfaces on two different links,
> with both links having a working default router. Both of the
> interfaces have preferred global addresses. When sending to a global
> destination address, if there's no routing reason to prefer one
> interface over the other, then an implementation MAY preferentially
> choose the outgoing interface that will allow it to use the source
> address that shares a longer common prefix with the destination."
>
> The RAs now convey "routing reasons", which affect the
> selection (in that
> without them, another choice might be made).
I'm still not understanding your scenario. The paragraph you qoute above is
a suggestion of how source address selection concerns *might* be used to
influence the choice of router. (I don't think it's a great idea but I
wanted to discuss it in the draft.) Whereas I think you're asking about the
other direction - how will the routing preferences influence source address
selection. The answer is simple: the routing preferences influence the
choice of sending interface, and that strongly influences source address
selection.
> Of course, this may be exactly want you want, but I suspect the common
> "multihomed" scenario at present is sites with a 2001: link
> and a 3ffe:
> link, one of which is probably much better than the other, even though
> both may route happily anywhere.
OK. Suppose a host has two interfaces. On Interface 1, it receives RAs
advertising a 2001 prefix. Interface 1 is better for reaching 2001
destinations. On Interface 2, it receives RAs advertising a 3ffe prefix.
Interface 2 is better for reaching 3ffe destinations. The host has a default
route on interface 1 and a default route on interface 2. Now the host is
sending to a 2001 destination.
This is a situation where the paragraph you qouted applies. An
implementation might choose the default router on interface 1 (the "right"
answer), by looking at its autoconfigured addressses on the two interfaces
and realizing that it has a "better" source address on interface 1.
Now, suppose the router on interface 1 advertises a more specific route for
2001::/16 and the router on interface 2 advertises a more specific route for
3ffe::/16. Now the routing table will select the "right" interface - it
doesn't need any help from source address selection.
I think this is a better approach. Which is why I've implemented it and not
implemented the quoted paragraph.
> > > I'm also interested in just how the routers will be
> configured to make
> > > the advertisements, as it would seem this would impact the router
> > > renumbering process?
> >
> > My first thought is that this should not impact router
> renumbering. Router
> > renumbering is about making configuration changes uniformly
> across a set of
> > routers, while the advertisement of default router preferences &
> > more-specific routes is very topology dependent. I think
> it's something that
> > an administrator would need to selectively configure on
> relatively few
> > routers, not configure uniformly across a site.
>
> The question is how deeply the "topologically dependent"
> changes are made
> by the administrator through the network. Having manually configured
> preferences in the config file only adds to the list of
> things that (might)
> need changing if the network is renumbered, or a network
> you're giving a
> preference to is renumbered? I appreciate there are other
> issues such as
> IOS access lists and of course IPs embedded in host configs.
> How realistic
> a goal is (relatively) transparent renumbering?
Consider an enterprise with 100 routers. I certainly do not imagine an
administrator will be going into each of those routers' config and
individually tweaking advertised router preferences and routes.
Maybe an administrator might have a policy like here's a specific /16 or /48
route that I want to advertise in RAs, and it would be nice to update this
prefix across a bunch of routers. I imagine an extension of router
renumbering could handle this.
Rich
--------------------------------------------------------------------
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]
--------------------------------------------------------------------
