Post by Mike Barnes Post by Evan Kirshenbaum Post by Mike Barnes Post by Evan Kirshenbaum Post by Mike Barnes
Thanks for the historical perspective. Nevertheless, it seems to me
that the "www." prefix is a pragmatic convention rather than
something in "the specifications". But I'm still not entirely sure
what you meant by "it should work" above so I might have
The specifications say that you give the name of the machine. If the
machine is named "www" (in domain "foo.org"), then nothing makes it
reasonable to expect that "foo.org" (or "web-server.foo.org" or
"workstation36.foo.org" or "random.porn.com") should be an alias for
the same machine. The only reason that people make sure that it is is
that they've noticed that people expect it to be.
So do the specifications say that you can't call a machine "foo.org",
and ensure that "www.foo.org" is an alias for the same machine"?
Of course not. It's perfectly acceptable (although before this a
trifle weird) to say that a machine is "www" (in domain "foo.org") and
also "foo" (in domain "org"). That's what I meant about "going to the
trouble of setting up name servers to make sure it will work".
At the risk of flogging a dead horse - I have no problem at all with the
idea of one machine serving the domain name and simultaneously serving a
subdomain of that domain. The issue I've actually been concerned with
all along (and clearly haven't been explaining very well) is your
apparent assumption that the name "www.foo.org" is a given and that the
extra trouble is taken to honour "foo.org". To me it's quite possible
and acceptable to view it the other way round, and I still haven't
worked out whether you agree or disagree with that.
It's acceptable, it's possible, and for any organization that had more
than one machine before they decided to put up a web site, it's highly
unlikely to be the case.
Any organization that has a domain and multiple machines within it
almost certainly runs their own name server. That means that
"www.foo.org" is host "www" served by foo.org's name server, which is
owned by the Foo entity. Changing records on your own name server is
easy. "foo.org", on the other hand, is the "foo" record on "org"'s
nameserver (actually, one of the root nameservers, which handle the
first level down in the non-country-specific heirarchies). Changing
things there is a bit more hassle, which is why such domain names
tended not to be used as hostnames except for legacy purposes.
Post by Mike Barnes Post by Evan Kirshenbaum Post by Mike Barnes
I agree that in the large corporate world things probably wouldn't
be done like that, but the large corporate world isn't the only
Sure, but if it wasn't for the large corporations (and universitites
and other entities that already had established internal networks
before bringing up external-facing web sites) doing this, people
would never have internalized the notion that they should expect it.
I'm not sure what you mean by that last "it".
That specifying a hostname of "foo.org" was identical to specifying a
hostname of "www.foo.org".
Post by Mike Barnes
Speaking for myself as a browser user, I've never seen the point of
The point is to specify the machine running the web server that you
want to talk to. Originally, these were simply the preexisting names
of the computers that people brought up web servers on. When
universities and companies started allocating dedicated machines for
the purpose, they initially chose lots of different names (e.g.,
"web", "web-server", "mosaic", "external", "http"). These names
distinguished these particular machines from the hundreds or thousands
of other machines in the domain.
Because it was important to specify the correct machine (or you
wouldn't find a web server), when people started assuming that
organizations had web sites, it was important that they be able to
guess what, e.g., "ford.com"'s web server probably was called, and the
convention was pretty quickly established that it would be "www".
The result of this was that the vast majority of URLs that people saw
began "http://www." and people started thinking of the "www." as part
of the format rather than part of the name. That is, they started
reanalyzing "http://www.foo.org" as
Since the "http://" part had always been optional, they thought of the
"www." as optional, too, and assumed that "foo.org" should work. And,
since this isn't unreasonable, people who ran web sites went to the
trouble of ensuring that it would.
Post by Mike Barnes
and although I initially suspected that there was some technical
reason for its presence (as well as the history that you've
described), subsequent developments appear to indicate that it was
never really necessary. I don't think that the actions of large
corporations and universities had any influence on my thinking.
Sure they did. If "sears.com" and "mit.edu" and "times.co.uk" and the
like hadn't been kludged to work, you never would have come to the
expectation that the web site hostname would be the domain name rather
than the name of a specific machine in the domain.
Evan Kirshenbaum +------------------------------------
HP Laboratories |If I may digress momentarily from
1501 Page Mill Road, 1U, MS 1141 |the mainstream of this evening's
Palo Alto, CA 94304 |symposium, I'd like to sing a song
|which is completely pointless.
***@hpl.hp.com | Tom Lehrer