DNS

[prev] [thread] [next] [lurker] [Date index for 2003/09/02]

From: David Champion
Subject: DNS
Date: 17:42 on 02 Sep 2003
* On 2003.09.02, in <20030902162044.GJ23320@xxxx.xxxxxxxx.xxx>,
*	"David Champion" <dgc@xxxxxxxx.xxx> wrote:
> * On 2003.09.02, in <Pine.LNX.4.55.0309021638100.6478@xxx.xxxxxxxxxxxxxx.xxx>,
> *	"Mark Fowler" <www@xxxxxxxxxxxxxx.xxx> wrote:
> > I hate software that doesn't map http://theirdomainname.com to
> > http://www.theirdomainname.com.
> 
> But that's asinine. We don't map uchicago.edu to smtp.uchicago.edu, or
> to finger.uchicago.edu, or to sunrpc.uchicago.edu, either. Because it's
> not. And it's not software's fault, it's people's achievement.
> 
> In a justifiably bad mood, but (unfortunately) not because of software,

Oh, all right. I'll take a shot at software, in the spirit of this post.

Mapping domain.name to www.domain.name is asinine; I stand by that.
But I'll tell you what I hate. I hate DNS. DNS is busted. DNS had a
good idea in the MX record: a short and simple way to say "whenever you
want this name for SMTP, use this other name instead." A fine concept,
a noble goal, and most useful. I can make mail for domain.name go to
smtp.domain.name with no trouble. And mail for anywhere else. What a
useful paradigm!

I need this for my http and finger services, too, as it happens. It
should be little surprise, and I imagine a lot of people do. We've
mapped finger at domain.name to finger at another box for ages -- since
before there was a web, and that coexists fine with mail. Adding a
mapping for http might be nice -- it's certainly user-friendly, and good
in principle. But it's impossible to map both finger and http jointly
with smtp. Why? Because DNS is busted.

What DNS should do is to provide a uniform, service-neutral RR mapping a
service (or port, if you'd rather) to a host. When my web browser wants
to look up http://domain.name, it looks up the service RR instead, and
perhaps finds the http mapping saying it should refer to www.domain.name
instead. Finger could easily do the same. It could even be built into
library routines for constructing sockaddr_in structures, if we wanted
to make things easier for the coder. But DNS doesn't do this; it's
short-sighted. It would rather push this burden to the application
protocol, where it doesn't belong.

DNS is busted.

-- 
 -D.    dgc@xxxxxxxx.xxx
 University of Chicago > NSIT > VDN > ENSS > ENSA > You are here
 .  .  .  .  .  .  .
 always line up dots

Generated at 14:02 on 01 Jul 2004 by mariachi 0.52