On Jan 5, 2011, at 11:21 AM, David Nicol wrote:
On Wed, Jan 5, 2011 at 1:41 AM, Steve Atkins <steve(_at_)blighty(_dot_)com>
wrote:
For layering something on top of DNS the deployed implementation is
the only spec that matters - what's implemented trumps what would be nice.
it depends if we are layering over "DNS: the installed infrastructure" or
"DNS: the protocol." If the second, we are free to demand and enforce a
particular semantics.
If the second... we could do much, much better than DNS for the WAN part of any
protocol.
If we're building the infrastructure from scratch, we can demand any semantics
we like, but unless the existing software happens to implement them, we're
still out of luck unless we're planning on implementing replacements from
scratch.
The only reason DNS was originally used was because it was easy to jam into
sendmail. The only reason it's used now is that the infrastructure for it is
ubiquitous (where infrastructure includes auth servers, caches and MTA support).
I think DNS still makes sense as a protocol to use between the MTA and a local
server. It makes much less sense for WAN distribution.
Cheers,
Steve
_______________________________________________
Asrg mailing list
Asrg(_at_)irtf(_dot_)org
http://www.irtf.org/mailman/listinfo/asrg