Rand Wacker <rand(_at_)sendmail(_dot_)com> writes:
On Tue, 10 Aug 2004, David Blacka wrote:
It is more credible for me to believe that there is DNS software that won't
fall back to TCP, or that there are some that don't actually allocate 64k
(the actual maximum) of space for receiving a message.
The issue is more operational, there will be quite a bit of resistance by
site security admins to open up TCP port 53 for queries of that size, and
even more resitance by sites that receive A LOT of mail to accept the
possibility of making outbound TCP queries for large number of incoming
messages.
More importantly, it calls into question why you'd want to
use DNS in the first place. The nice aspect of DNS is that
you have exactly one round trip with UDP. At the point
that you have to truncate a response and refetch the
entire RR set using TCP (assuming that you don't tickle
the TCP bug John refers to), you might as well have used
DNS to do an A or SRV lookup and used some _other_ tcp
service that's assumedly more appropriate and/or purpose
built for this application.
So keeping the likelihood of < MTU size on the RR set
seems like it out to be a _requirement_ since this wg is
chartered to use DNS, and thus be DNS friendly. If it
comes down to a tradeoff with wildcards, we'll just have
to decide which is friendlier... but transport unfriendly
DNS seems like a pretty bad option.
--
Michael Thomas (mike(_at_)mtcc(_dot_)com http://www.mtcc.com/~mike/)