Using ICP queries will likely work fine in your current situation, but if you can put CARP in use (I'm not sure if LVS supports it), it might give better results (less overlap between caches). A bit of poking for LVS and CARP turns up http://www.austintek.com/LVS/LVS-HOWTO/HOWTO/LVS-HOWTO.L7_switch.html#id2949641, which suggests using multicast ICP on separate NICs. *shrug*
CARP is indeed the way to go, but preferable i would like a more
flexible way, and I am considering KTCPVS,.
although its beta and will probably need some work. because i can then
implement some kind of high availabilty by simply using mon or
ultramonkey to remove the server from the loadbalancer if it is down.
What i understood from CARP is that i need to edit config files and
restart squid if one goes down. This is far from ideal since its hard to
automate.
http://kb.linuxvirtualserver.org/wiki/KTCPVS
about ext3/reiser.
I will see what i do, ill prob stick with ext3
