[MITgcm-support] Re: is this bug in mitgcm code?

[Matthew Mazloff]

Hi Suneet,

You are correct that the current version of the MITgcm does not allow  
useSingleCpuIo for obcs controls.

Patrick, you implemented useSingleCpuIo for the mdsio slice routines  
awhile back for me.  I have also changed them to only write slices  
for the northern boundary, not global files.  I have attached  
them....the issue with checking them in would be one should remove  
the change i made to the xz code (bracketed with CMM's) as this is  
not robust, and will crash for "useSingleCpuIo=.FALSE."

not sure what else may be wrong with the code...but worked for the  
northern boundary

-Matt


-------------- next part --------------
A non-text attachment was scrubbed...
Name: mdsio_slice_routines.tar
Type: application/x-tar
Size: 153600 bytes
Desc: not available
URL: <http://mitgcm.org/pipermail/mitgcm-support/attachments/20080509/3c95d22b/attachment.tar>
-------------- next part --------------



On May 9, 2008, at 12:12 PM, Suneet Dwivedi wrote:

> Hi Everyone,
> When I tried to run mitgcm adjoint model with obcs package on, I endup
> with the following error message:
> ---------------------------------------------------------------------- 
> ---------------------------------------------------------
> (PID.TID 0000.0001) *** ERROR *** MDSREADFIELD_XZ_GL: File does not  
> exist
> ---------------------------------------------------------------------- 
> ----------------------------------------------------------
> My model stopped at:
> ---------------------------------------------------------------------- 
> ----------------------------------------------------------
> (PID.TID 0000.0001)  MDSREADFIELD_XZ_GL: opening file: maskobcsn. 
> 001.001.data
> (PID.TID 0000.0001)  MDSREADFIELD_XZ_GL: filename: maskobcsn. 
> 002.001.data
> ---------------------------------------------------------------------- 
> ------------------------------------------------------------
> The detailed description of the problem and the solution (that worked
> for me)  is as follows:
> (i) Model works fine (stops with 'normal end') when I use single
> processor/double processor on a single node.
> (ii) Model crashes with the abovesaid error message when I start using
> 16 processors on maybe 10 nodes.
>
> When I actually started looking for the missing files
> 'maskobcsn.002.001.data', 'maskobcsn.003.001.data' and so on; I found
> that the same were written at different nodes than master node even
> though I used "useSingleCpuIo=.TRUE." I then copied all these files
> manually to the master node and after that model stopped at different
> place looking for "adxx_obcsn.0000000000.002.001.data" and then for
> "xx_obcsn.0000000000.002.001.data" and
> "pickup_obE.ckptA.002.002.data". So I copied all the files associated
> with obcs control to the master node and rerun the model and it worked
> fine this time for me (stopped with normal end). It means that
> useSingleCpuIo=.TRUE. does not work for obcs control files. I wonder
> is this a bug in the mitgcm while using obcs alongwith ecco? Am I
> missing some statement required for obcs package to write output on a
> single CPU while running the model on multiprocessors? Please help me
> sort out this problem.
> Hoping for reply,
> Cheers,
> Suneet