Without immediately addressing what has later been discussed in this
thread, I should probably clarify what I meant in my original post.
On Sat, Jan 3, 2015 at 10:12 PM, John C Klensin <john-ietf(_at_)jck(_dot_)com>
wrote:
If I understand the above part of your comments correctly, "one
A/AAAA record..." hasn't been necessary since HTTP 1.1
introduced the HTTP "host" verb. That was a lesson we learned
from SMTP, FTP, and elsewhere -- virtual hosts should not
require separate (either physical or virtual) interfaces and
addresses.
With the introduction of the Host header, virtual hosts indeed do not
require new interfaces or addresses, be they physical or virtual.
Regardless of this, if I was to request <http://foo.example/> using a
current HTTP/1.1 user agent, the AAAA and/or A records for foo.example.
would be queried, potentially using the Happy Eyeballs algorithm.
If the entity behind foo.example. wanted to run their primary HTTP
service on a host other than the one indicated by the AAAA or A
records, the closest solutions now would be to configure the host on
the bare domain to either route or reverse proxy to the real server.
This may occur for example if foo.example. uses a shared web hosting
provider, while wanting to serve HTTP as well as other protocols
directly on their bare domain.
The ability to use SRV records would eliminate the additional traffic
and computational overhead required for routing or reverse proxying.