[MITgcm-devel] problems with profile package on BlueFire

[Matthew Mazloff]

Hi Gael,

Yes, I suspect its an issue with the (excessive) number of files to be  
simultaneously initialized.  Several files are written OK before  
crashing.  Thanks to ARGO there are a lot of files to initialize!  I  
will try to confirm this now by running with just a small number of  
profiles.

Assuming this is the problem, however, do you have any ideas on a  
remedy?  I have already tried with each processor writing to its own  
local directory and this did not help.  Perhaps we can use MPI to let  
the write commands go in processor order...

Although this is all a bit surprising cause model initialization has  
so much I/O and everything seems to go OK...

Thanks,
Matt







On Mar 25, 2011, at 2:47 PM, Gael Forget wrote:

> Hi Matt,
>
> I am not sure what the problem may be. I am using the profiles package
> with the binary option on bluefire without problem. So I suspect it  
> is not
> a problem with the code. When I get crashes from pkg/profiles (on
> bluefire or elsewhere) it usually results from missing subdirectories,
> or an excessive number of binary files. You may want to try run
> verification/global_with_exf to narrow down the issue you have.
>
> As for the netcdf output I have not used it in ages since it appeared
> very slow. At the time it worked fine though, if I recall correctly.
>
> Cheers,
> Gael
>
> On Mar 25, 2011, at 1:55 PM, Matthew Mazloff wrote:
>
>> Hello,
>>
>> The model is crashing in profiles_init_fixed.F (actually  
>> profiles_init_ncfile.F) on BlueFire.  It does not like initializing  
>> all the profile files and gives the error:
>>
>> "profiles_init_ncfile.f", line 3557: 1525-014 The I/O operation on  
>> unit 10 cannot be completed because an errno value of 2 (A file or  
>> directory in the path name does not exist.) was received while  
>> opening the file.  The program will stop.
>>
>>
>> This is for the default profilesfile_equi_type.EQ.2 -- binary  
>> writing.  If I set profilesfile_equi_type.EQ.1 -- netcdf writing --  
>> all files are initialized but I have a suspicion that  the model  
>> values are being written incorrectly -- can anyone verify that   
>> profilesfile_equi_type.EQ.1 works?
>>
>> And, perhaps, more importantly, does anyone know why I am getting  
>> this error on BlueFire when initializing the profile package using  
>> binary I/O?
>>
>> Thanks
>> Matt
>>
>>
>> _______________________________________________
>> MITgcm-devel mailing list
>> MITgcm-devel at mitgcm.org
>> http://mitgcm.org/mailman/listinfo/mitgcm-devel
>
>
> _______________________________________________
> MITgcm-devel mailing list
> MITgcm-devel at mitgcm.org
> http://mitgcm.org/mailman/listinfo/mitgcm-devel