On Sep 26, 2011, at 6:21 PM, Christian Huitema wrote:
We see here a proposal to create site local IPv4 addresses for Internet
providers. The IETF previously expanded significant efforts to deprecate IPv6
site local addresses. Why exactly do we believe that IPv4 site local
addresses would be a good idea, when the consensus was that IPv6 site local
addresses caused more harm than good?
Not exactly to play devil's advocate here, but I don't think these are quite
like site-locals. It seems like they're more like "ISP locals".
One of the problems with site locals was that they were ambiguous addresses AND
that they would need to be used by applications, and we had plenty of
experience that said that RFC 1918 addresses caused harm. If 240/4 wouldn't
be used or seen by applications at all, the fact that those addresses would be
reused in other networks wouldn't be such an issue. Though I agree that these
will leak, even if they're never used as application endpoints. If one of them
ever appears as a source address on an ICMP reply, for instance, that actually
will cause problems - perhaps not problems that directly affect applications,
but problems nonetheless. (Then again, tunnel use is now quite widespread,
which means that packets travel paths that are completely invisible to the
endpoints and look like a single hop to them. And there are of course
problems with those, but we sort of deal with them.)
Another of the problems with site locals was that there was no clear boundary
that corresponded to a site, so why have a special class of addresses for a
site? If that's also true for an ISP, maybe an ISP local address isn't such a
good idea either.
What happens if two ISPs that are each using 240/4, merge? Probably the same
kind of mess you get when two enterprises that use RFC 1918 addresses merge.
Granted, ISPs might not merge as often as enterprises, but still...
It was especially important to get rid of site locals in IPv6 because IPv6 was
in very early stages of deployment, and any errors in its design would be
magnified over time. By contrast, IPv4 is a dinosaur struggling to take its
last unassisted breaths, and which is starting to be put on life support. Some
sort of extraordinary measures to keep IPv4 vlable for a short time might be in
order, even if those measures would never make sense in IPv6.
So my take is that using 240/4 is not an absolute no. But that bit of address
space is a very precious resource and there needs to be strong justification
for using it, along with reasonable assurance that it will not do significant
harm in relation to the amount of benefit it will likely provide. And merely
prolonging the life of IPv4 is probably not sufficient justification.
Keith
_______________________________________________
Ietf mailing list
Ietf(_at_)ietf(_dot_)org
https://www.ietf.org/mailman/listinfo/ietf