[Wrf-users] reproducing identical results on different machines/compilers

[Fotis Georgatos]

Dear Marc,

I don't know how much off-track is the following, but give it a check;
* compilers may produce code that does different initialization of the
IEEE-754 arithmetic (in fact, that are compile time parameters to control
exactly that). That can explain at least some part of the divergence.
In short, even the same compiler can exhibit effects as you describe, with
varying tuning; different compilers might have different default ieee754
tuning anyhow.

But there is more to it.

You may be hitting one more manifestation of this:
* http://en.wikipedia.org/wiki/Chaos_theory#History -> "An early pioneer of
the theory was Edward Lorenz <http://en.wikipedia.org/wiki/Edward_Lorenz>
...[...]...had discovered that small changes in initial conditions produced
large changes in the long-term outcome"
If you see the errors piling up as time advances, you can pretty much start
suspecting this possibility.

What do you think, could it be that?

best,
F.




On Fri, Mar 21, 2014 at 6:30 PM, Marcella, Marc <MMarcella at air-worldwide.com
> wrote:

>  Hi all,
>
>
> I had a question about wrf and compiler/machine differences that I was
> hoping to get some clarity on.  I am trying to run the same set of wps/wrf
> compiled executables on two different machines (the libraries for the
> machine in which the .exe's were not compiled on were placed in the exact
> same location as those on the machine in which the .exe's were
> configured/compiled on, hence they run just fine).  So essentially, the
> only difference is the actual machine hardware being used.
>
>
>
> For metgrid.exe and real.exe, the results between the two machines are
> identical, however even with identical met_ems and wrfinput/wrfbdy files,
> the wrf results are different.  By day 58 or so the average  differences in
> TA and especially precipitation, are quite large (on the order of
> 1-2mm/day).  I am trying to run this over Southeast Asia where there is a
> lot of cumulus convection.  But, I wanted to check if this is to be
> expected to observe such large differences?   I did do a quick internet and
> forum search but couldn't find specific work documenting this.  And, I
> wanted to see if there was a way in the compiling of the code (via a flag,
> etc that one can add) to get identical results across machines or even
> compilers?  I ask because what is interesting is that running an identical
> CCSM.exe on the two machines will yield identical CAM results on the two
> different machines.
>
>
> Any help or insight you could lend would be greatly appreciated.
>
>
> Thanks,
> Marc
>
>
>
> _______________________________________________
> Wrf-users mailing list
> Wrf-users at ucar.edu
> http://mailman.ucar.edu/mailman/listinfo/wrf-users
>
>


-- 
-- 
echo "sysadmin know better bash than english"|sed s/min/mins/ \
  | sed 's/better bash/bash better/' # Yelling in a CERN forum
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mailman.ucar.edu/pipermail/wrf-users/attachments/20140322/665cc149/attachment.html