[Top] [All Lists]

RE: Last Call: draft-klensin-rfc2821bis

2008-03-27 12:34:53
The key issue here is whether people who rely on AAAA are likely to
achieve their desired result. Today it does not matter because anyone
who relies on AAAA alone with no A fallback is going to receive almost
no mail.

That in turn is going to depend on whether software implementations are
written to fall back to A record lookup or host name lookup. Before AAAA
they were the same thing.

One software architecture would be to interpret RFC 2821 litterally and
only do A record fallback. I suspect that most people would implement
this by directing to their hostname lookup module which in an IPv6 stack
is going to be A/AAAA agile. So in that case the AAAA fallback is going
to come for free.

Coupled with the fact that DNS servers are probably going to be
configured to push AAAA records with an A record lookup by default in an
IPv6 world I suspect that there is not going to be a problem.

-----Original Message-----
From: ietf-bounces(_at_)ietf(_dot_)org 
[mailto:ietf-bounces(_at_)ietf(_dot_)org] On 
Behalf Of Tony Hain
Sent: Thursday, March 27, 2008 9:05 AM
To: 'Keith Moore'
Cc: ietf(_at_)ietf(_dot_)org
Subject: RE: Last Call: draft-klensin-rfc2821bis

Keith Moore wrote:
Tony Hain wrote:
Your arguments make no sense. Any service that has an MX creates 
absolutely no cost, and the fallback to AAAA only makes one last 
attempt to deliver the mail before giving up. Trying to force the 
recipient MTA to publish an MX to avoid delivery failure on the 
sending MTA is useless, and in no way belongs in a 
standard document.
MX records are an operational optimization, nothing more.

that's completely incorrect.

what MX records mean is that a domain name used on the 
right hand side 
of an email address need have nothing at all to do with any host or 
other service that has the same domain name.  in particular the 
servers and resources associated with that email address 
don't have to 
share the same host, network, addresses, user community, 
administration, or anything else.  (except that the 
administration of 
the DNS zone and RRs associated with that domain is the 
same for both)

in short, MX records decouple mail domain names from host 
names.  and 
this turns out to be a very useful thing to do.  e.g.

1. a domain used for mail that doesn't correspond to any actual host

2. a host that doesn't want to source or receive mail

3. when it is desirable to associate email and other 
services with the 
same domain name, and yet not have all of those services 
hosted on the 
same cpu or at the same address.

for instance, the email for and the web server 
for the same domain are hosted by entirely different companies on 
different networks -- because I couldn't find a hosting 
company that 
did an adequate job of both at a reasonable price.  and yet 
it's very 
useful to have them both associated with the same domain name.

That is all well and good, but it is completely of value to 
the receiving MTA, and under their complete control. There is 
nothing that requires a receiving MTA to follow this model, 
despite what others may see as value.
Defining the facility is what the standards need to do. 
Dictating operational practice without cause is what a 
standard needs to avoid doing. 

The function of mail delivery is between IPv4/IPv6 endpoints, and 
how those endpoints find each other is orthogonal to the actual 
service of mail delivery. Having the document state a 
2 of the possible methods is pushing the edge already

that's an incorrect way to characterize what is going on, 
because an A 
record is only a valid destination for mail to a particular 
domain if 
no MX records exist.  if even a single MX record exists, it's 
incorrect to route the mail based on an A record, even if 
an attempt 
to relay the mail via the listed MX resulted in a temporary error.

I agree that if an MX exists, the operator of the receiving 
MTA has stated its expectations, and the sending MTA needs to 
oblige. That is not the same as mandating that every 
organization has to follow the same model. If there were some 
serious technical consequence for lack of the MX record I 
would be all for specifying its use. Operational practice 
with A records shows that there is no real issue, and that 
anything that does come up is under the control of the 
impacted party with a clear mechanism to resolve it. 

Again, the text is fine as it is.


IETF mailing list

IETF mailing list