Message-Id: <m0tw4sk-0000kiC@mamba.ast.cam.ac.uk>
Date: Mon, 11 Mar 96 10:31 GMT
From: drtr@ast.cam.ac.uk (David Robinson)
To: new-httpd@hyperreal.com
Subject: Re: cache verification
Sender: owner-new-httpd@apache.org
Precedence: bulk
Reply-To: new-httpd@hyperreal.com
Status: O

> On Sun, 10 Mar 1996, David Robinson wrote:
> 
> > 1: I stole the Last-modified factor from the CERN server, so your last
> >    statement is false.
> 
> No, it's not. As I understand it, CERN uses that directive to determine 
> when to get rid of the cached file. i.e. it caches the file until the 
> time figured out from the Last-modified header times the factor runs out, 
> but it still does a conditional GET each time.

Hmm, you may be rght.

> > 2: You can achieve the behaviour you want by setting the last-modified
> >   factor to zero; then all documents without an Expiry header will 'expire'
> >   immediately.
> 
> But this is not very desirable, as it then entails retrieving the whole 
> document each time.

No it doesn't. It still caches the document.

> > 3: The behaviour of not checking the remote server is very useful;
> >   for a lot of people a cache that always checks the origin server can
> >    be worse than useless for small documents. (small = few hundred bytes).
> > 
> Average size of a web document is about 5k. Average size of a 304 
> response sent to a conditional GET: 90 bytes.

I said small documents! For these, the limiting factor is the RTT; the saving
of one network packet for a conditional GET compared to a document fetch is
often not significant.

> > There is no overriding reason to treat a missing Expire as 'expire now'.
> > In a lot of cases that is the wrong assumption to make; most documents
> > don't expire immediately. Of course, that's not say that this isn't the
> > right assumption to make sometimes.
> 
> It's a question of perferring to get the right document over the wrong 
> one. The following scenario:

> 1. Someone creates a document in June, 1996.
> 2. Your proxy accesses in February, 1998.
> 3. The person changes it in March, 1998.

> Anyone using your proxy will not see the changes for a whole two months.

Again, this is simply not true. The proxy will not cache a document without
and Expires header for longer than MaxExpire without checking the origin
server. MaxExpire is 12 hours, by default (I think).

> This doesn't quite sound like desired behavior to me. And, yes, this isn't
> completly hypothetical - I actually do have some web documents that
> haven't been updated since 1994, and I'm sure other people do, too. What
> if I decide to change them tomorrow?
> 
> While I'm admit that for certain applications, an
> always-return-cached-version mode may be desirable, I think that, at
> least, there needs to be an option to act as I've described.

It does. RTFM.

 David.