On Wed, 2 Apr 2008, Arnt Gulbrandsen wrote:
If the ability to set up a mail server purely by setting up a host
(without adding an MX record) is a significant feature of the internet
mail system, then I would expect to see significant use of it. Right?
So I took 30,000 recent messages containing little spam. In those
messages I found 954 unique domains in To, From, Cc and Reply-To
fields. Among the 954 domains, practically all have MX records and most
have A records. There are 38 domains that have an A record but no MX.
All 38 are reachable, 33 answer on port 80, 21 answer on port 25, and
some of the 21 seem to be incorrectly configured.
So that leaves about 20 domains, 2% of the original.
I checked my logs of the last 7 days. My mail servers have sent emails to
29.401 different domains worldwide, 29.015 via MX records and only
286 via A records, which is less than 1 %. For a statistik about how many
emails are sent via MX records compared to A records the result would
even be lower.
The missing MX records for the 286 domains are the result of lazy or
clueless mail administrators.
The discussion here is about a standard, which means enforcing rules and
not helping people to stay lazy.
It makes no sense to transfer a feature which was needed 20 years ago but
not anymore in these days to a new protocol.
That means in the case of no MX record, I suggest the following algorithm:
- if a domain has only A records, synthesize a MX record
- if a domain has A and AAAA records, synthesize a MX record only
for the A records, the AAAA records should be dropped.
- if a domain has only AAAA records, do not synthesize a MX record
For a MTA with ipv4 stack, this would mean no change. For a MTA with dual
stack, only the ipv4 part would use synthesized MX records. A MTA with
ipv6 stack only would never synthesize a MX record.
Michael Storz