[squid-users] Squid web accelerator, how to minimize load on backend servers?

Hi,

I am running Squid in web accelleration mode on several servers. They
are set up to share cache data using ICP:

cache_peer cache1.xxx.no sibling 80 3130
cache_peer cache2.xxx.no sibling 80 3130
cache_peer cache3.xxx.no sibling 80 3130
cache_peer cache4.xxx.no sibling 80 3130
cache_peer cache5.xxx.no sibling 80 3130
cache_peer cache6.xxx.no sibling 80 3130
cache_peer cache7.xxx.no sibling 80 3130
cache_peer cache8.xxx.no sibling 80 3130

Then I have a couple of origin servers:

cache_peer foo.finn.no parent 80 0 no-query originserver
cache_peer bar.finn.no parent 80 0 no-query originserver

If I take one cache server out of service, then when I put it in service
later on, it seems it will prefer to get missing data from the origin
servers. How do I change that to preferring neighbour caches (siblings)
over origin servers (parent(s))? Also during normal operation, if there
is one new object, I prefer if as few as possible cache servers has to
get it from the origin server(s).

Can I use "nonhierarchical_direct off" to achieve this? What about
hierarchy_stoplist? I wish to treat all URLs equally.

PS: I am getting UDP_HIT in my access logs, so I assume the cache
servers can get data from each other using ICP.

Cheers,

-- 
Anders.