On 2010-12-06, Amos Jeffries <squid3@xxxxxxxxxxxxx> wrote: > > If you have max_filedesc available you don't need to touch --with-maxfd. > The build option just sets the default for the squid.conf option. Hmm.. I tried pretty hard to increase it beyond 16K without succeeding, before I noticed the comment in the default squid.conf saying: # The maximum value for max_filedesc is set by --with-maxfd during # compilation. and the RHEL RPM was build with --with-maxfd=16K. Once I rebuilt it I was able to go to 32K and 64K. So are you completely sure --with-maxfd shouldn't be necessary ? > > Others seems to have no problems with FD as high as 400K. The OS can > handle a few million apparently. Great, thanks! > > If the number of connected clients gets extremely high you could tune > the client persistent connection timeout to reduce resource usage a bit. I am a bit scared to touch the client side. My clients are stupid STBs, and one never know how they react to changes.. But do you think tuning down persistent_request_timeout to f.ex. 30s should be a safe setting ? > An upgrade to 2.7 will also make the requests complete and be ready for > release faster if you want to go that way. We try to stick with RHEL-released packages, so the next version we will use will most likely be squid-3.1.4 which is in RHEL6. -jf
