On 5/22/2014 8:05 AM, Ned Freed wrote:
Over in DNSOP the usual arguments about CNAME and domain name variants
are being rerun. (Short summary: too many people believe that
variants can be made "the same" purely within the DNS, because they
don't understand the problem.)
One of the harder issues is application configuration.
...
So is this a new idea?
For i18n? Maybe. For email?
...
It doesn't work well at all, in part because the DNS is a lousy choice for
such
information operationally, but mostly because in practice domains require
much
richer semantics:
...
The DNS just isn't flexible enough.
...
I've never seen a writeup of the issues anywhere.
John's opening classes this as a generic problem.
Seems like what is needed is a careful casting of the kinds of
functional requirements people often make, here, and then a careful
explanation of the reasons the DNS is a poor venue for satisfying them.
I provided a list in my original response that demonstrates that a simple name
mapping is in no way sufficient. So if you're going to do this with the DNS,
you're talking about defining some new, and very complex, records. Which AFAIK
isn't anything anyone wants to do.
FWIW, we typically store domain information in LDAP along with user entries.
The records we use are shared by a bunch of different applications besides
email. I'd venture that there are >50 defined attributes in various object
classes, with ~25 of them having some relevance to email. And setups with many
(as in tens of thousands) of domains are not uncommon.
But really, Keith's answer is better. Applications should be provisioning the
DNS, not the other way around. In practice mail systems deal with large amounts
of configuration and account information. Are we going to stuff all of that in
the DNS? And if not, why complicate applications with having to get their
configuration from an additional place?
So, help the read to "understand the problem" and then to understand how
the "DNS just isn't flexible enough".
"DNS Name-Mapping Considered Harmful?" (No, I'm don't think that's the
right title; just trying to vector the direction.)
Am I missing something? Is there a serious proposal being worked on to try and
specify this nonsense? If not, why should we bother?
Ned
_______________________________________________
ietf-smtp mailing list
ietf-smtp(_at_)ietf(_dot_)org
https://www.ietf.org/mailman/listinfo/ietf-smtp