I put Roy's l_c perf patch on HotWired's servers for a few hours... but
the FIN_WAIT_2s flew through the roof, so I had to go back to
-DNO_LINGCLOSE. (IRIX 5.3). There was some speculation that this might
help the FIN_WAIT_2 situation, but it doesn't look like it. As far as
performance goes I didn't really get a chance to compare, I'll try again.
One idle thought I had was that we might try playing with the order of
things -- do the shutdown(sd,1) after a select() timeout or a successful
read().
The other thing I found, just like Chuck said a while back, I played with
the Min/MaxSpareServers settings and removed all the huge spikes from my
load averages. The change in load *could* be related to locality of
reference in the L2 cache, but I'm not convinced that is the only thing.
We're talking a difference of 50% vs 20% "cpu usage" as reported by the
status module. (I'd really like to revamp status to do 5 minute rolling
averages.)
Dean
|