[MITgcm-support] fortran script to join tiled-based model output into global files

Matthew Mazloff mmazloff at MIT.EDU
Tue Jan 6 13:36:47 EST 2009


Hi Constantinos,

I am using single-cpu I/O and global_files = false and getting great  
performance.  I actually have not tried single-cpu I/O = false and  
global_files = true...i actually thought there might be bugs in that  
aspect of the code...is this not true...should I try it?

so i get tiled output.  then I run a separate executable to pack the  
ctrls (i do this for memory reasons).  The reading of the tiled files  
and writing to ecco_ctrl or ecco_cost takes hours.  If the ctrls and  
gradients are passed in as global files, however, this packing takes  
less than 30 minutes.  So I thought it would be nice to post-process  
and turn the xx and adxx fields global....might also be nice for post  
processing of other things too.

And no, on ranger i do not have to use separate directories, the file  
system seems to be very good

-matt

ps> another huge advantage of tiled output is that the tapelevel  
files dont get too big so I am using only 3 level checkpointing.




On Jan 6, 2009, at 10:18 AM, Constantinos Evangelinos wrote:

> On Monday 05 January 2009 09:57:01 pm Matthew Mazloff wrote:
>
>> ps> just and fyi..on ranger it took ~2.5 hours to pack the controls
>> when written in tile format and about 20 minutes when written as a
>> global file
>
> Do you mean single-cpu I/O or global_files? Does the latter work  
> currently?
> Were you outputing the tiles in separate directories per process?
>
> Constantinos
> -- 
> Dr. Constantinos Evangelinos
> Department of Earth, Atmospheric and Planetary Sciences
> Massachusetts Institute of Technology
>
> _______________________________________________
> MITgcm-support mailing list
> MITgcm-support at mitgcm.org
> http://mitgcm.org/mailman/listinfo/mitgcm-support




More information about the MITgcm-support mailing list