Mailing List Archive

content negotiation
I've been thinking about using this, but I am confused about one issue. How
will content negotiation effect proxy servers? Anyone care to enlighten me?

I see a real problem if the proxy server caches a bunch of ugly
looking data for a URL because a limited browser made a request. Or
worse, not even considering the browser making the request and
just grabbing the document in a "lesser" form?

[. Note: I'm not complaning about content negotiation, I just want to
know if we also need to write a smart proxy :-) ]

Cliff
Re: content negotiation [ In reply to ]
> I've been thinking about using this, but I am confused about one issue. How
> will content negotiation effect proxy servers? Anyone care to enlighten me?

I think Rob T pumps out a "Proxy: no-cache" - or whatever the magic
words are to stop caching. I also believe he provides enough info for
smart cachers to cache different versions of the dodcuments, but that's
probably being ignored everywhere at the moment.

When the proxy cache writers see Apache leading the way with CN. I'm
sure they'll react.

robh
Re: content negotiation [ In reply to ]
Date: Sat, 15 Apr 1995 17:43:31 -0700
From: Cliff Skolnick <cliffs@steam.com>


I've been thinking about using this, but I am confused about one issue. How
will content negotiation effect proxy servers? Anyone care to enlighten me?

I see a real problem if the proxy server caches a bunch of ugly
looking data for a URL because a limited browser made a request. Or
worse, not even considering the browser making the request and
just grabbing the document in a "lesser" form?

As Rob H. mentioned, I send out a "Pragma: no-cache" with negotiated
documents, to make sure this doesn't happen. (At Roy's insistence,
there's something you can set in httpd.conf to turn off the pragmas,
but with the current crop of caches, I'm not sure why anyone would
want to set it. WRT bandwidth, non-caching of documents with includes
has been and will continue to be a much bigger problem).

Of course, some aspects of the CN spec are still in flux. In
particular, the aspect of most interest to most of the people here,
which is negotiation of HTML levels, is in the process of being
rewritten (Dan Connolly's "toward graceful deployment of tables" page
still hasn't been updated). The change, which was agreed on at the
latest IETF, is a technical improvement --- "version" probably is
better than "level" to indicate what's going on.

However, the more time they spend perfecting this mechanism, the
longer it'll be before they can seriously try to persuade anyone to
use it, and some fairly important real-world deadlines (like the
release of Netscape 1.1 final) are coming up in the not-too-distant
future. Sic transit gloria mundi.

rst