[squid-users] can't handle the traffic

We are a school district that is trying to use squid/squidGuard for
filtering & caching (although the filtering is the most important in
that it is required). We have a Dell dual PIII 750 box with 1GB ram and
UW-SCSI drives running. We had a consultant come in to assist with the
initial setup. The problem is that we can't get it to handle the load.
We have 43 schools plus administration and support buildings (probably
around 5000 users typically) with traffic usually running a sustained
4.5M to 5M, occasionally a little more. When we first set up the box it
pretty promptly fell over. Since then, due to suggestions from our
consultant and from folks on the k12os list I have since disabled
caching so we could concentrate on filtering with

         acl all src 0.0.0.0/0.0.0.0
         no_cache deny all

and also limited traffic to only the elementary schools and one middle
school. We are trying to move away from websense both because of cost
and also the desire to get away from proprietary solutions for as many
things as we can. The cache box is inline between the last router in the
chain (a cisco 7206) and the PIX firewall. It handles routing all the
traffic fine but if I redirect too much to squid we startly loosing
throughput fast. The elementaries alone consume in excess of 90% cpu
resources on both processors. Adding one middle school increases that
right to 100% utilization and adding a high school makes throughput
start falling and sometimes makes squid fall over completely. The
biggest indication of a problem I see in the logs are warnings that all
redirect process are full and a suggestion to increase redirect
processes. I started the box at redirect_children 30 and have since
discovered that it apparently is capped at 32.

We've been dealing with this most of this week and are getting to the
critical point - admin folks are beginning to look wistfully at websense
again in spite of the cost.

Help!

-- 
Mike Rambo
mrambo@lsd.k12.mi.us