W3C home > Mailing lists > Public > ietf-http-wg-old@w3.org > January to April 1997

Fact-checking: do any in-service proxy caches ever ignore Expires?

From: Jeffrey Mogul <mogul@pa.dec.com>
Date: Fri, 14 Mar 97 16:15:39 PST
Message-Id: <9703150015.AA02059@acetes.pa.dec.com>
To: http-wg@cuckoo.hpl.hp.com
X-Mailing-List: <http-wg@cuckoo.hpl.hp.com> archive/latest/2667
Koen and I have been discussing offline whether it is possible
to send
	Expires: Sun, 06 Nov 1994 08:49:37 GMT
(or some similar ancient date) to ensure that *every* pre-HTTP/1.1
proxy cache will not, under *any* circumstances, cache the response.

This is NOT a question about what HTTP/1.1 specifies, or how
RFC1945 describes HTTP/1.0, or what is "right", or what is
"common."  This is a question about the worst case of current
practice; please don't start a debate over whether current practice
is good or bad.

We know that HTTP/1.1 allows caches to ignore Expires under
certain well-defined circumstances; we are NOT asking about that.

We're also not interested in proxy cache implementations that
were in use at some point in the past, if it is known for sure
that they are not in use today (nor will be used in the future).

Many of the proxy caches in use today are based on either the CERN
httpd (written by Ari Luotonen) or the Harvest/Squid cache
(written, I believe, by several people including Peter Danzig.
I checked with Ari and with Peter Danzig; both say that none of their
caches have ever ignored an explicit Expires.
Koen mentioned this:
    Shel Kaphan told me about 1.5 years ago that AOL's proxy cached 1.0
    responses for some minimum time no matter what.  I don't know if
    that is still true now.  Also, I think New Zealand's (academic)
    hierarchical proxy cache system has some kind of nontransparency
    for `certain sites with a low educational value'.

We checked with Shel, and he replied:
    I'm sorry, but once I developed enough electronic counter-measures
    for AOL and other systems out there, I stopped paying a lot of
    attention to these anomolies except in cases where particular
    systems have given us trouble.  I actually don't recall my claim
    about AOL, but if Koen says I said it, I must have said it.

    At the moment we're still not issuing Expires headers, especially
    to cause pages never to be cached, because of the much discussed
    history list interactions, so we wouldn't have been running into
    this lately anyway.

So at this point, we're not sure.  Perhaps someone with both an
AOL account and the ability to generate various Expires headers
can do an experiment on AOL's cache (or perhaps someone from AOL
can respond?).

Received on Friday, 14 March 1997 16:21:34 UTC

This archive was generated by hypermail 2.3.1 : Wednesday, 7 January 2015 14:40:19 UTC