Re: Problem with old pages

From: Andreas Strotmann <Strotmann@dont-contact.us>
Date: Mon, 8 Jul 1996 10:18:39 -0700

On Jul 6, 0:57, Duane Wessels wrote:
> Subject: Re: Problem with old pages
> guenther.fischer@hrz.tu-chemnitz.de writes:
>
> >Second: If I get the page from my neighbor, does my squid give a new
> >TTL from scratch or does he see the remaining TTL from the neigbor?
> >If not and I think it is the reality - 2 neighbors could produce a ever
> >living page.
>
> Hello Guenther,
>
> You are absolutely right, this is a serious bug. I'm embarassed that
> I didn't think of it before.
>
> Well, there's a few ways we can fix this.

> opinions?
>
Yes:

What I have never understood is why the ICP protocol doesn't send both the
time-last-checked and the time-last-modified or expiry time for a hit (the
latter two need only one field with a single bit telling which of these is
encoded).

That way, any cache in a hierarchy or in a neighborhood can enforce its
own lifetime policies without forcing those policies on its neighbors.

I believe that you could also solve the problem mentioned above: a
neighbor would never retrieve an item that, according to its metric, has
expired.

This presupposes that the time-last-checked is copied from the neighbor
that furnishes the page by retrieving that time stamp from the ICP 'hit'
packet due to which the page was retrieved from the neighbor (the other
time stamps are of course available from the HTTP header).

There's a problem that the page might change in the short time between ICP
and HTTP requests, but that's erring on the safe side.

An added benefit of doing this is that it could be much safer to add in
Netscape caches that do not delete files once they are "stale" (they save
bandwidth by issuing "if-modified-since" requests for stale files). The
time-stamp data is readily available there in such a way that a Perl
script can produce correct ICP replies independent of the Netscape proxy.

Andreas

-- 
Andreas Strotmann       / ~~~~~~ \________________A.Strotmann@Uni-Koeln.DE
Universitaet zu Koeln  /| University of Cologne   \
Regionales Rechenzentrum| Regional Computer Center \
Robert-Koch-Str. 10    /|    Tel: +49-221-478-5524 |\   Home: -221-4200663
D-50931  Koeln        __|__  FAX: +49-221-478-5590 |__________~~~~~~~~~~~~   
Received on Mon Jul 08 1996 - 01:21:51 MDT

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:32:35 MST