Squid 1.1 beta Refresh model confuses me

From: Stephane Bortzmeyer <[email protected]>
Date: Mon, 18 Nov 96 13:12:27 +0100

From Squid 1.1 release notes, it seems Squid now uses a completely
different way of storing files from remote Web servers.

As I understand this sentence:

>An object is considered "fresh" if it meets these requirements:
>
> 1) AGE <= MIN_AGE, or
> 2) NOW < EXPIRES, and
> AGE <= AGE_MAX, and
> LM_FACTOR <= PERCENT

It means, that, if we have a MIN_AGE of zero in refresh_pattern (as
suggested in the sample config file), *and* the Web server does not
provide a Expires header (the vast majority, see the output of
stats/reply_headers), Squid will *always* issue a GET IMS
(If-Modified-Since) request. (I indeed observe this behaviour on the log
of Web servers.)

This is a serious problem. Many Web servers are terribly difficult to
reach from our Internet access provider Renater (that's the reason why we
have a cache, anyway). To do it just to receive a "not modified" reply
implies a huge latency on replies. The problem will not disappear soon
because very few Web servers allow you to specify properly an Expires.

Of course, I can raise MIN_AGE to something more than zero, but then, I
create trouble with "nice" Web servers which provide these Expires
headers, sometimes for legitimate reasons (I do not consider marketroid
who whises "real" hit counts as legitimate complainers).
        
Would it be possible, I don't know, to change the algorithm to:

An object is considered "fresh" if it meets these requirements:

    1) AGE <= MIN_AGE, or
    2) LM_FACTOR <= PERCENT, and // Old Squid 1.0 behaviour
       f... server didn't provide an Expires at all, or
    3) NOW < EXPIRES, and
       AGE <= AGE_MAX, and
       LM_FACTOR <= PERCENT

(BTW, our problems on the large Squid cache on an Alpha/OSF1 seem to
fade. More on that later, we say in French, "Never sell the bear's skin
before you actually kill it.")
Received on Mon Nov 18 1996 - 04:12:46 MST

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:33:34 MST