[sitefinder-tech-discuss] A small technical suggestion

[Mo McKinlay <mo.mckinlay@cirrus-multimedia.ltd.uk>]

Hi,

I'm going to leap in here with both feet, with the potential of being
shouted at, but here goes anyway.

In my mind, the single largest technical issue with SiteFinder was that of
the assumption that the WWW is the Internet - i.e., that a wildcard for Web
users wouldn't affect other services. This, obviously, is not the case - A
resource records are protocol-agnostic, as we all know.

However, there is a rough equivalent to A RRs which are protocol-specific:
SRV records.

I could be wrong (and I hope someone will correct me swiftly if I am), but if 
a wildcard SRVs existed for HTTP, other protocols would carry on as though
they weren't there. Therefore, the only problem is convincing browsers to
use SRV records (which, in the case of Mozilla-based browsers wouldn't be
that difficult).

Have I made an incorrect assumption? Please shout at me if I have. If not,
is this not what Verisign was *trying* to do when creating SiteFinder - and
would it not be the correct way to do it?

In my view, if SiteFinder must exist (and I'm going to steer well clear of
that debate for the time being), then it must not break the
well-defined behaviour of *any* protocol - and in this case, that means only
returning a positive response for HTTP.

[[ As an aside, I've just done some brief experimentation here; I note
'host' and 'dig' don't return anything useful for a an A or MX query on a
domain containing only SRV records -- but also don't return NXDOMAIN (as
this would cause cached negative responses). 

Is an 'empty' response in this sense "enough" for known protocols? FreeBSD's 
libc resolver gives "No address associated with name" [rather than "unknown 
host"] in this situation - Sendmail 8.12.8p1 on FreeBSD 4.x bounces mail
immediately with "Host not found" if mail is sent to a domain containing
only SRV records - though I can't speak for other servers. ]]

Is the use of SRV records ultimately desired, and is it workable? On the
practical side, it's unlikely that a patch for Mozilla would be nontrivial
or difficult to get accepted, but convincing Microsoft to implement the
functionality for Internet Explorer might take some doing. Not to mention
the myriad of other browsers based upon neither codebase.

Just a thought.

Mo.

-- 
E: mo.mckinlay@cmlx.co.uk
T: +44 (0) 709 200 3083
W: http://cmlx.co.uk/