> I don't remember hearing what the "quick_abort" settings were
> on the caches that saw the spikes.
quick_abort was 10 50 50 . Our squid.conf is from 2.0 and doesn't
mention range_offset_limit, so when we ran 2.1, range_offset_limit must
have been the default.
A few people mentioned bandwidth spikes in conjuction with ICP peers
or cache digests, but our caches weren't configured to use any peers
at the time of the spikes.
Someone mentioned a Network Associates site (term-xxx.nai.com).
We had noticed problems with that site before and had disabled caching of
HTTP traffic to that network (via transparent redirection).
Several people have suggested delay_pools. I'll take a look at that.
gid
