Re: multi-host virtual sites for HTTP 1.2

> If you're not convinced that you want that large a (virtual) site, OK by
> me. But I liked your other reasons for referrals, too:
> 
> [Roy said:]
> 	There are good reasons to support a
> 	URI rewriting mechanism on the client side (e.g., to make the URI
> 	syntax truly extensible, support automated mirroring, support URNs,
> 	etc.), but I haven't considered single-site performance to be one
> 	of them.
> 
> Administrative flexibility and distributed sites were also mentioned as
> other good uses.

Yes, but the best solution for those problems do not involve changes
to the HTTP protocol -- the URI resolution table is a data format that
can be included in any entity (and thus subjected to things like
digital signatures, expiration times, and all the other things necessary
to avoid spoofing and denial of service).

>>Well, I still don't like it -- I see no reason for such a huge site
>>to be located behind a single DNS name.  The top URLs can be on the
>>main site (or just redirections from that site) with all other
>>references relative from there.  I don't see why HTTP needs to 
>>replicate the work of DNS.
>
>It doesn't. It is applying the same principles used by DNS to the part
>of the HTTP URL that DNS doesn't own, for the same reasons as DNS uses
>them in its part of the URL.

Yes, but why? Why can't it leave site-dispersal to the DNS part and
the site-specific name resolution to the remainder?  I know why it would
be useful (it was the impetus of the "path" scheme, after all) -- my
question is why is it *necessary* to support a server with a single DNS
name that actually exists on separate sites.  Furthermore, if such a thing
is necessary, why not do it as part of the DNS resolution response
(like an MX record).

>>Finding a way to do it with multiple Link header fields would be better.
> 
> So, would you suggest something like:
> 	Link: Referral <absolute-URI-1>
> 	Link: Referral <absolute-URI-2>
> 	...
er,
        Link: <absolute-URI-1>; rel="mirror"

was more along the lines I was thinking.  It probably won't fit the
exact problem you are solving, but it would be nice if it could.

> And why do you like it better? It seems weird to me, but in the way that
> sometimes says to me that it's because my intuition hasn't been trained
> properly yet.

Because implementation of Link in general is important.  It is the next
big step on the road to salvation (or something).  It is probably the most
important aspect of the HTTP protocol that hasn't been implemented
(correctly, at least).  Why, if I owned a browser that implemented Link
correctly and was at least on a par with other browsers, I'd own the
market within three months.  It would make Java applets look like
yesterday's underwear -- stiff and unwieldy.

Hmmmm, I guess I could use some sleep...

 ...Roy T. Fielding
    Department of Information & Computer Science    (fielding@ics.uci.edu)
    University of California, Irvine, CA 92697-3425    fax:+1(714)824-4056
    http://www.ics.uci.edu/~fielding/

Received on Friday, 12 July 1996 06:28:37 UTC