Re: Squid, Reload and cache hierarchies

From: Stephane Bortzmeyer <bortzmeyer@dont-contact.us>
Date: Thu, 18 Jul 96 11:39:01 +0200

On Wednesday 17 July 96, at 12 h 46, the keyboard of Duane Wessels
<wessels@nlanr.net> wrote:

> >Therefore, when a Lynx reload hits Squid, Squid sends an ICP request to
> >its parent which does *not* include the "Pragma: no-cache" :-( (Weakness
> >in ICP?) If the parent has the page in its cache, it sends it back and no
> >real reload was done.
>
> Hm, it must be that the object is returned in a HIT_OBJ reply?

Yes: the parent sees:

193.55.4.10 - - [17/Jul/1996:17:44:03 +0200] "ICP_QUERY
http://www.eu.org/" UDP_HIT_OBJ 2337

> That can (and will) be fixed.

You mean, ICP will transmit the "Pragma: No-Cache"?

> >On the other hand, when Netscape reloads, Squid bypasses the hierarchy
> >because of the "If-Modified-Since" :-( (See Release-Notes-1.0.txt.) So,
> >the desire of Netscape to save bandwidth leads to more resource
> >comsuption.
>
> This can probably be changed too, now that Squid handles IMS a little
> differently.
>
> It used to be that Squid treated all IMS requests as MISSes. It would
> then forward the request directly to the origin server. One reason
> for that is because we expect most (95%?) of IMS requests will
> end up being "304 Not Modified."

Yes but the problem is that it is useless to have a cache hierarchy if a
reload does not update the whole hierarchy. More perversion: since the
parent is not modified by a Netscape Reload, a Lynx Reload will erase the
good copy on the child with the old one retrieved from the parent (yes, I
tried it).

> I can send you a patch to try....

On the child? OK, I'll try.

More general question on ICP which seems poorly documented (or did I miss
something?). Is it possible to transmit HTTP headers in ICP packets?
Other possibility: when a client issues a request with IMS
(If-Modified-Since) or "No-Cache", why not using only HTTP to retrieve it
from the parent? Therefore, the complete semantic would be transmitted.
This would not be too costly since such requests are probably uncommon.
Received on Thu Jul 18 1996 - 02:40:21 MDT

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:32:36 MST