dns media type registration tree

[Keith Moore]

Let me play Devil's advocate here.

I'm not sure why people assume that just because they are familiar with 
how to allocate names in a particular name space, that that name space 
should then be adapted to every purpose that comes to mind.  There is 
value in having semantics associated with a name space - a value which 
is diluted by using that name space for a wide variety of purposes.  
DNS is too overloaded as it is.

Beyond that, DNS is not well-suited for media types.  DNS assignments 
are ephemeral.  They are subject to change as their assignees (e.g. the 
organizations whose names they reflect) merge, split, go bankrupt, fail 
to renew their registrations, or sell off trademarks.  They are subject 
to reassignment for arbitrary reasons. We discovered long ego that URIs 
based on DNS names are not suitable for long-term (archival) use 
precisely because those names change; that's why URNs and DOIs were 
invented.  And the utility of URNs has been nearly destroyed by misuse 
and overloading of that name space.

It must be questioned whether it is beneficial to the public to define 
new media types on a whim anyway.  The failure to pay proper attention 
to the design of media types, the failure to do security analysis of 
media types, and the failure to respect those analyses even when they 
are done is the reason why email- and web-borne viruses and worms cost 
billions of dollars to consumers.

If there really is a compelling need (meaning that it serves the 
greater good) to define new media types at a whim, a much better set of 
names already exists, one which was more-or-less designed for that 
purpose.  It's called OIDs.  They are easy to obtain.  They are 
recursively extensible just like DNS names (actually moreso).  It is 
relatively well-established that once assigned, the meanings of OIDs do 
not change.  They don't contain human-readable content that invites 
disputes over ownership.

The argument for DNS media types reminds me of countless other 
arguments for why protocol X should be used for everything (or at 
least, every instance of some large class of problem).   In the past X 
has taken on values such as SOAP, XML, HTTP, SNMP, LDAP, URLs, SSL, 
ASN.1, RPC, and even TELNET.  Most of those arguments look pretty naive 
now, but people took them seriously when they were in fashion.  Now 
it's essentially being argued that since DNS is a widely-deployed 
namespace and query protocol, that it should be used for yet one more 
thing that could be looked up.

Keith

p.s. gratuitous analogy to a famous quote:

Abraham Maslow is supposed to have said "It is tempting if the only 
tool you have is a hammer, to treat everything as if it were a nail."

I like Mike O'Dell's version better, which I'll parphrase since I 
didn't manage to write it down at the time:

"If you need to drive a nail, the fact that you have your forehead with 
you doesn't make it a good tool for the job."