Hello,
I'm looking for some suggestions for debugging a problem...
Under load (not much load IMO), our web-app on glassfish U1, generates a
lot of cpu mutex contention activity.
smtx=spin on mutex (lock not acquired on first try) >> a bad thing in
general.
My throughput drops dramatically when this happens. The appserver is on
an 32 cpu T1000 running solaris 10. Attached is one snapshot of mpstat
for a five second period.
For this snapshot, I calculate a total of 464,050 mutex spins across 32
cpu's in 5 seconds. That's 14,501 per cpu on average, in 5 seconds. Or
92,810 spins per sec overall.
Seems like a lot to me.
Thanks,
Russ