[MITgcm-devel] lab_sea salt_plume

[Jean-Michel Campin]

OK, in lab_sea/code, SIZE.h has been changed 
but not SIZE.h_mpi. This might explain those large
differences.

By the way, why do we need Olx=Oly=5 (instead of just 4 
for T & S advection scheme) ?

Jean-Michel

On Wed, Dec 26, 2007 at 11:10:36AM -0500, Jean-Michel Campin wrote:
> Hi Dimitris,
> 
> It's about this new test within lab_sea dir. 
> that you added:
> 
> I noticed that with mpi, the results are quiet different from
> the standard output (or with g77 on faulks):
> For instance, on aces with g77, we only get 1 digit of matching
> for cg2d, which is much worse than any other experiment:
> http://mitgcm.org/testing/results/2007_12/tr_aces-gnu_20071226_0/summary.txt
> 
> And cg2d_ini_res (but uvel, theta behave similarly) is already
> different at the 2nd iteration:
> 1 cg2d : g77+mpi on aces    output.salt_plume.txt
>    1  1.62385282835673E+00  1.62385282835673E+00 :  16
>    2  4.85548954979253E-01  4.68227543998213E-01 :   1
>    3  2.45622971174368E-01  2.25845413764413E-01 :   1
>    4  1.63522154679596E-01  1.43321071390522E-01 :   1
>    5  1.20614641817605E-01  1.10517741406029E-01 :   1
>    6  9.70754606193202E-02  8.54199023969107E-02 :   1
>    7  8.69449945190776E-02  7.15457138959820E-02 :   1
>    8  7.75254707036916E-02  6.36283224074781E-02 :   1
>    9  6.97929202192410E-02  5.94210606948534E-02 :   1
>   10  7.07050338339757E-02  5.84864259736977E-02 :   1
> 
> It could be worth to check more carefully the tiles computation
> (in salt_plume pkg ?): the mpi run has only 1 tile per processor
> is more likely to be right.
> 
> Jean-Michel
> _______________________________________________
> MITgcm-devel mailing list
> MITgcm-devel at mitgcm.org
> http://mitgcm.org/mailman/listinfo/mitgcm-devel