[Wrf-users] Results variability depending on processor count

[James Correia]

Hi-
It is more matter of truncation at a certain number of processors
specifically involving domain decomposition. That is using, say 64
processors, you will always get the same answer on a particular machine. But
if you use 32 processors, the domain decomposition and subsequent
integration produces slight differences probably at the level of truncation.
Since all of our problems are highly non-linear any differences no matter
how small changes the solution.

I have seen this issue before and even in small time period integrations.
The key is the number of processors. I would imagine this is also compiler
dependent (i.e. Exactly how functions are calculated and at what
precision).  I think differences in precipitation can be rather large when
using the same data but using a different number of processors, especially
when not a power of 2.

That is my understanding. If anyone else can correct me, or add more, I
would be grateful as well.
jimmyc

On Thu, Sep 18, 2008 at 6:26 AM, Jan Ploski <Jan.Ploski at offis.de> wrote:

> Hello,
>
> WRF produces slightly different results depending on how many processors
> are used in a shared memory run (of course, the "slight" differences
> become bigger with integration time). I know that the general explanation
> for this is "rounding errors". What I would like to understand is why the
> rounding errors must appear at all. After all, data can be transmitted
> exactly from one MPI process to another, so why does it matter how many
> MPI processes are involved? I'd be grateful for a brief explanation...
>
> Regards,
> Jan Ploski
>
> --
> Dipl.-Inform. (FH) Jan Ploski
> OFFIS
> FuE Bereich Energie | R&D Division Energy
> Escherweg 2  - 26121 Oldenburg - Germany
> Phone/Fax: +49 441 9722 - 184 / 202
> E-Mail: Jan.Ploski at offis.de
> URL: http://www.offis.de
> _______________________________________________
> Wrf-users mailing list
> Wrf-users at ucar.edu
> http://mailman.ucar.edu/mailman/listinfo/wrf-users
>



-- 
James Correia Jr.
Post Doc
Climate Physics Group, PNNL
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mailman.ucar.edu/pipermail/wrf-users/attachments/20080918/f100fb29/attachment.html