NUMBER OF SPECTRAL POINTS LESS THAN NUMBER OF PROC

[Biao-Zhao]

Hi all,

I’m trying to run WaveWatch III with 1024 processors, but the model stops immediately with the following error message:

**> *** WAVEWATCH III ERROR IN W3INIT :
>     NUMBER OF SPECTRAL POINTS LESS THAN NUMBER OF PROC.
>     NSPEC, NAPROC =     900    1024**

Since WW3 does not actually perform domain decomposition in spectral space, I’m wondering why this limitation exists.

Does this mean I must reduce the MPI processes or adopt unstructured grid?

For you information, following is my namelist file for ww3_grid

$ WAVEWATCH III Grid preprocessor input file
$ ------------------------------------------
  'C3200 Grid '
$
    1.1 0.04118 25 36 0.
$
$ Set model flags ---------------------------------------------------- $
$  - FLDRY         Dry run (input/output only, no calculation).
$  - FLCX, FLCY    Activate X and Y component of propagation.
$  - FLCTH, FLCK   Activate direction and wavenumber shifts.
$  - FLSOU         Activate source terms.
$
  F T T T T T
$ geogr. CFL time step is the 2nd one....
$ note that PR1 shows instability before PR3
$ min spacing for active portion of grid is 10 km
$ from  T=1/0.0368, Cg=25.447, I get dt=393 s
  180.  90.  90.  10.
$
 &PRO2 DTIME = 0. /
$ &PRO2 DTIME = 2777. /
 &PRO3 WDTHCG = 0., WDTHTH = 0. /
$ &PRO4 RNFAC = 0., RSFAC = 0. /
  &MISC CICE0 = 0.25, CICEN = 0.75, FLAGTR = 4 /
 &OUTS USSP = 1, IUSSP = 3,
  STK_WN = 0.04 , 0.1102, 0.3305 /
END OF NAMELISTS
$
  'CURV' T 'NONE'
  2123 1090
  20 1.0 0.0 1 1 '(....)' 'NAME' 'C3200.lon'
  21 1.0 0.0 1 1 '(....)' 'NAME' 'C3200.lat'
  -0.1 2.50  22  1.0 1 1 '(....)' 'NAME' 'C3200.depth'
  24  1.0  1 1 '(....)' 'NAME' 'C3200.obstr'
  25  1  1 '(....)' 'NAME' 'C3200.mask'
$
   0. 0. 0. 0.  0

Any clarification or background on this design choice would be greatly appreciated!

Thanks,

[JessicaMeixner-NOAA]

Hi @Biao-Zhao - To remove the limitation on MPI tasks, you must move to an unstructured grid as the structured grid does do domain decomposition for particular parts of the code for spectral points, or use less MPI processors than number of spectral points.

[Biao-Zhao]

Hi@JessicaMeixner-NOAA , Thank you very much for the helpful information! I’m working on improving the wave-coupled processes in the SHiELD–MOM6–WW3 and conducting real case simulations, and at this stage it seems that my only practical option is to use fewer MPI processors. As far as I know, there hasn’t yet been an application case of unstructured grids within FMS coupler. Thanks again for the clarification!