Re: Accelerator Cache as a Parent Cache?

From: Henny Bekker <>
Date: Tue, 22 Apr 1997 18:43:08 -0100

Dear Thomas,

>My objective is to use a Squid cache as a front end for two web servers that
>have identical content. Ideally Squid would act kind of like a load balancer
>between the two systems. At this point though one of the systems though is
>more powerful, so I wanted it to be favored.
That sounds a lot like "server side caching" or in a simple form known as a
HTML accellerator. Normally Web caching-servers doesn't cache volatile data
e.g. information taken from a database..

>After reading all the Squid docs I have seen that there are hooks for a
>redirector program. I imagine if I were really good at Pearl programming and
>knew how to extract ICMP data from the "Network Probe Database" I could write
>a redirector that might rewrite URLs to the fastest responding system, yet
>unfortunately I'm not a Pearl guy, and the info I have on the "Network Probe
>Database" is pretty limited. If anyone can supply some help here, it would be
You might write a simple redirector program, running on the Web caching-server
acting as a HTML accellerator, which is using a static load balancing (e.g.
redirect relative more requests to the powerfull server than to the less
powerfull server.)
In this you don't need to adjust the "Network Probe Database" server..

>The Squid.conf file has the "Neighbor Selection Algorithm" which allows you to
>"weight" parents. My first question, is that if you have 2 parents defined,
>one with a heavier weight, if the heavy weight parent does not end up
>responding does Squid then forward the request to the parent with the lower
With the above solution you don't need neighbour caches.. All requests are
cached on the central HTML accellerator.. Ofcource you could install HTML
accellerators also on the database servers but that seems to be redundant..

>Next question, can the parent caches be HTTP Accelerator caches? So what I'm
>getting at is this, a request comes into the Squid cache and misses. Squid
>thenforwards the request to the heavy weight parent. The heavy weight parent
>doesnot have the object cached, so it then forwards the request to the HTTP
>server it is configured to "accelerate" for.
Yes. That's generally the way it goes.. However (as mentioned above) queries
to databases are normally not cached but I don't know this specific database

>Hopefully if I get the desired answers to these questions, the end result will
>be that I have one Squid cache that acts as a front end serving all the
>cacheable objects, and then forwards "misses" to the parent Accelerator caches
>based on the parents weights.
>Hopefully there is someone out there on these email lists that can shed some
>light on this stuff. In the end anything I learn, develop, or discover, I will
>write it up and forward it on to the Squid community.
I'm curious at other people's thoughts about server side Web caching...

Cheers, Henny
E-Mail: ! Disclaimer: The main obstacle to progress
http ! progress is not ignorance,but
PTT: Voice: +31 30 2536971 Fax: +31 30 2531633 ! the illusion of knowledge
X400: /G=Henny/S=Bekker/OU=cc/O=ruu/PRMD=surf/ADMD=400net/C=nl o
Paper: H.J.Bekker, Utrecht University, Computer Centre _ /- _
        Po Box 80011, 3508 TA Utrecht Nederland (_) > (_)
Received on Tue Apr 22 1997 - 10:09:00 MDT

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:35:01 MST