In responce
OMP_NUM_THREADS was set to 1 - we already used noht - and /proc/interrupts
appears ballanced.
I installed the lam-7 beta off the beta page. And proceeded to retest
we recompiled the source to link to the new copy of lam libraries.
New results (all xeon) - total time for 122 atom run.
2 processors one machine
- tcp, sysv and usysv all - 4minutes 50seconds runtime.
- lamd - 5minutes 10 seconds runtime.
1 processor each, 2 machines - standard 100Mb ethernet connection
- tcp, sysv and usvsv all same (as expected) - 4minutes 10seconds runtime
- lamd crashed (not sure why)
(4:50 is about 19-20seconds per step in last email)
Still rather odd.
I would say memory bandwidth contention - except the athlon results seem to
be pretty good.
Gareth
_________________________________________________________________
ninemsn Extra Storage is now available. Get five times more storage - 10MB
in your Hotmail account. Go to
http://join.msn.com/?page=dept/home&pgmarket=en-au
|
