ietf-mxcomp
[Top] [All Lists]

RE: suggested new RRtype experiment

2004-05-22 15:00:30

Eric Hall says:

It's a rotten idea [storing XML in DNS], exhibiting a complete lack
of understanding about the nature of DNS.

If folks aren't going to use small fast datagrams that will work
within
the cache-intensive infrastructure, they should use another protocol.

I completely agree with the second point. We don't want to prescribe
anything that forces people to use DNS over TCP.  We don't want to
prescribe
anything that can't be cached.

However, these maxims don't preclude XML.  For almost everyone, their
MARID
info will trivially fit into a single DNS record with no work.  For
really
big complicated domains, they may need to work harder, but can still
avoid
the need to fall back to TCP. HotMail, for example, has published Caller
ID
records (which are of a size similar to what we're contemplating). No
TCP
is required -- in fact, their DNS server doesn't even accept TCP
connections.
If the biggest email handler on the planet can do it without TCP, then
everyone probably can, too. (Yes, they use indirection through 3 other
records.  The biggest handful of ISPs might be expected to need to do
the
same.)

-- Jim Lyon
   mailto:jimlyon(_at_)microsoft(_dot_)com