Robert V. Coward wrote:
I agree. But we have infrastructure problems that really push hard to make it a single ip. We'll be doing WCCP and standard proxy. But a large number of the clients have hardcoded proxy ips and make it prohibitive to change it to a new address. I want to have a cluster of boxes spread horizontally across the network, but I have what I have.
l4 swich or software based load balancer (we used ultramonkey/lvs for this)
topology looks like this:
/- cache0
incoming requests --> lvs -- cache1
backup lvs \- cacheN
you're limited to the scalability of a single load balancer but if you
have to you can put something huge in that role like an f5. fault
tollerance is achieved through active failover and of course you can
afford to lose cache boxes with only minor or localized consequences as
well...
it's is generally in my experience much easier to achive scalavbility
and fault tollereance though multiple boxes rather than spending in
excess for the the best possbile basket to put all your eggs in.
R
--- On Thu, 7/17/08, jason bronson <jasonbronson@xxxxxxxxx> wrote:
From: jason bronson <jasonbronson@xxxxxxxxx>
Subject: Re: Squid in the Enterpise
To: "Adam Carter" <Adam.Carter@xxxxxxxxxxxx>
Cc: "squid-users@xxxxxxxxxxxxxxx" <squid-users@xxxxxxxxxxxxxxx>
Date: Thursday, July 17, 2008, 8:33 PM
On Thu, Jul 17, 2008 at 8:15 PM, Adam Carter
<Adam.Carter@xxxxxxxxxxxx> wrote:
Our planned deployment box is a 8-way, 16GB
ram, 1TB (6 disks
I think) server which will be running RedHat
Enterprise Linux.
There's been some recent list discussions about
how squid uses CPU - you'd be much better off with 4
load balanced dual core boxes than one 8 core box. RAM is
cheap so put 16gig in all four :-) Just make sure you
install the 64 bit kernel.
I would have to second that, its why google is so fast they
have many
small servers not one big one. but for the price of dual
core machines
i think you can afford more then 4 why not go with 8 dual
core
machines the trouble is be careful with how you handle the
1 Terabyte
drive...if you mount it from across the lan it could be a
bottleneck.
