[MITgcm-support] STDOUT redirect in parallel

[Martin Losch]

Hi Neil,

this is what I do in such a case:

I copy forcing etc to the /scratch directory and run the model there. Then you'll have all I/O on that disk. This also works for distributed resources, by default the model writes to the local directory (useSingleCPUio=.false., and globalFiles=.false.)

You can also turn off most of the output to STDOUT by using the mnc package for the monitor output (monitor_mnc = .true. in data.mnc), and debugLevel = -1 (in data&PARM01), then you'll only get output in STDOUT at the beginning and the end.

Martin

On Mar 23, 2011, at 11:12 PM, Neil Swart wrote:

> Hi folks,
> 
> I run MITgcm in parallel on a server using the TORQUE resource manager and MOAB scheduler. The gpfs disks on the system really slow down the model if it writes to them - so I need to redirect everything to a local /scratch disk.
> 
> I use mnc for monitor, and I can easily redirect all the mnc output by specifying: mnc_outdir_str='/scratch/mnc' in data.mnc. 
> 
> I'm wondering if there is there is a similarly easy run time way to redirect the STDOUT & STDERR streams to /scratch - (other than by editing eeboot_minimal.F) ?
> 
> This is probably obvious and I just don't see it.
> 
> Thanks,
> Neil
> ---------------------------------------------------
> Neil Swart
> 
> Climate Modelling Group
> School of Earth and Ocean Sciences
> University of Victoria
> Victoria, BC
> Canada
> 
> http://climate.uvic.ca/
> 
> 
> _______________________________________________
> MITgcm-support mailing list
> MITgcm-support at mitgcm.org
> http://mitgcm.org/mailman/listinfo/mitgcm-support