Therefore, that leaves one under-discussed question: should a prefix be
used to avoid collision with other records as one measure of preventing
the requirement of DNS over TCP?
No, for two reasons.
The first is that it is not a very effective way to keep multiple TXT
records out of the same DNS node. Based on what I've read, there
aren't a whole lot of other TXT records at the same nodes where MARID
records need to live. The bloat problem occurs when you use
wildcards. Since no DNS server I know supports wildcards of the form
_marid.*.phoo.com, you have to put all of your wildcards at *.phoo.com
and that's where you get overlap and bloat. If specific nodes do get
overloaded, SPF and Sender-ID both let you use a short record pointing
to a longer record somewhere else as a band-aid.
The other is to keep in mind that using TXT records is a botch. DNS
was designed to have a separate record type for each separate
application. Type numbers are 16 bits so number assignments are not
particularly precious, and queries by type number are the right way to
get the records you want and not the records you don't want in a small
DNS response. It's a historical accident, and arguably a historical
mistake, that there haven't been hundreds of types assigned in the
past two decades.
I realize that short-sighted implementation decisions by some DNS
software authors and vendors make it difficult to deploy new record
types in the near term. The right response to that is to permit TXT
records as a transition measure (no doubt for a very long transition),
but to keep in mind that it's a transition and in the longer term, we
should deploy new record types as needed to support new services and
applications.
Regards,
John Levine, johnl(_at_)iecc(_dot_)com, Primary Perpetrator of "The Internet
for Dummies",
Information Superhighwayman wanna-be, http://www.johnlevine.com, Mayor
"I dropped the toothpaste", said Tom, crestfallenly.