Re: Squid Cache Design

From: Oskar Pearson <oskar@dont-contact.us>
Date: Tue, 14 Oct 1997 11:44:08 +0200

Hi

> I've been doing a lot of reading on squid, and various implementations.
> But, i'm not sure what are the god & bad points about each, if you think
> you know anything about proxy design please forward all your comments onto
> me..
I work for a large (by non-US standards at least) ISP.

We have decided on the following setup:

We have 3 caches (medium-end boxes) which handle all requests from
leased-line customers. We also have a machine at each of our 'far away'
branches that leased-line customers there can point to. We also have
a cache for our dialup section (though I still need to install it).

The only real problem with this model is that the people that aren't
close to our core caches still point to them instead of the caches at
our branches.

The branch caches are setup with the 'no-query' and 'default' options with our
central caches as parents.

I like the cluster of parents for a few reasons:
1) cheap hardware
2) reliability (you can unplug the broken cache's cable and then
        setup an IP alias on a working one

> so far all the multi-cache systems are all siblings & proxy-only, but,
> how do the caches lower down fill up at all if it's proxy-only to it's
> parent ?? or what about proxy-only/siblings amonst caches on the same
using proxy-only isn't a good idea.. it should pretty-much never be
used.. anyone know of a case where this is useful?

> level, and then parent to the next? but is it a waste to have an object
> stored twice in a parent and localy? As a side note, we are more
Not if the network between the child and parent cache is slow...

Oskar
Received on Tue Oct 14 1997 - 02:49:32 MDT

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:37:17 MST