Hi Peter,

I am responding to this via the mom4-bulletin board to solicit input
from others.  

The short answer is: I do not know. 

The long answer is that in mom4, we have what we call a "regression test
suite".  In this suite, we test the code using some basic comp-sci sorts
of metrics (amongst others).  An important metric that we insist upon is
reproducibility across restarts and reproducibility across numbers of
processors.  An example is 

1day on x-pes + 1day on y-pes = 2days on z-pes

mom4 code that is sanctioned satisfies this test on our machines
(SGIs).  We cannot guarantee that it does on others, and there are
examples where others have reproducibility problems (typically a
compiler issue).  When developing mom3, we did similar regression tests,
so at least at GFDL we have not seen behaviour as you account below. 
Perhaps the mom-community can chime-in with accounts like
yours...(ANYONE??)

-Steve 




Dr Peter Killworth wrote:
> 
> Steve, pass this on to whomever...
> I've been testing a very benign eddy parameterisation on our 24-processor
> machine here that GFDL helped get MOM3 working on.
> I find that if I start the global (and vanilla) run from scratch, and (say)
> run for 2 years, dumping every year, things are fine. I can take the restart,
> shift it into restart.dta or whatever it's called, and run another 2 years.
> And so on. Got it to 10 years with no problem.
> At this stage the restart file got lost (no backup, short of disc space).
> Started again asking for a 20-year run, dumping every 2 years.
> Even the first dump was NaNs.
> I am beginning to think it's the SGI somehow. Others have had similar
> effects. Different answers on different days.
> Query:
> have you had any reports of mom3 giving different answers depending on how
> it's started, machine it's on, etc.?
> (It may just be me, of course.)
> Peter

-- 
Dr. Stephen M. Griffies                phone: +1-609-452-6672
Geophysical Fluid Dynamics Laboratory  FAX:   +1-609-987-5063
P.O. Box 308, Forrestal Campus Site B  email: [EMAIL PROTECTED]
Princeton, NJ 08542  USA               http://www.gfdl.noaa.gov


-------------------------------------------------------
This SF.net email is sponsored by: eBay
Get office equipment for less on eBay!
http://adfarm.mediaplex.com/ad/ck/711-11697-6916-5
_______________________________________________
FMS-mom4 mailing list
[EMAIL PROTECTED]
https://lists.sourceforge.net/lists/listinfo/fms-mom4

Reply via email to