I have been profiling our cache servers here, and have seen something
I consider interesting:
readprofile | sort -n +2 says:
482646 do_select 997.2025
483390 tcp_select 1438.6607
111986 free_wait 1646.8529
227259 check 1721.6591
167677 sock_select 3493.2708
204091 inet_select 4638.4318
Any idea why it's spending so much time in "free_wait"? The whole
cache servers function on a "select engine" so the select's are
understandable, but the machines should have enough
memory... they aren't using any swap at all...
Should I be playing around with /proc/sys/vm/freepages ?
It's got 128M of ram, running 2.0.30 and I am not getting any "cannot
get free page" errors etc.
Ideas?
(also - what's "check" do...?)
Oskar