I would offer that we select the "thing" that looks the most persistent
to be the persistent identity.
Actually, you want to select the identity that's appropriate for your
purpose. DNS is not inherently better than IP for all purposes.
DNS names are often failure-prone, slow to lookup, and/or out of
sync with reality.
If the choices are: DNS name vs IP address, I think most people would
recognize that the DNS name is the persistent identity.
And if 'most people' treated this as a general rule, they'd be wrong.
There are several situations where IP addresses are more usable
than DNS - the DNS name may not even exist, lookups may not work
outside of a realm, or the name may be bound to an IP address rather
than a host. It is highly dependent on the configuration of the network
where the hosts are located and the DNS servers that serve them.
See draft-moore-nat-tolerance-recommendations-00 for a more detailed
explanation.
We should probably try to move the debate from "proof by emphatic
assertion" to analysis.
Presumably that also applies to assertions about persistence of DNS names.
Keith