[MITgcm-support] resubmit runscript for NCAR lightning

Lucas Merckelbach lmm at noc.soton.ac.uk
Mon Mar 13 05:52:40 EST 2006


On Fri, 10 Mar 2006, Dierk Polzin wrote:

> Hey does anyone have a nice run scripts to run MITgcm on NCAR Lightning..
>
> I have the mpi and netcdf working.. but can only get about 2.5 years with 12 
> processors with our 184x204 grid in 6 hours..
>
> I need to write a script to resubmit it after 6 hours or 2 years.. over and 
> over again.

Hi Dierk,

I had a similar problem running MITgcm on a linux cluster via a scheduler 
that gives the job exactly 12 hours cpu time, finished or not. The script 
prepares the runscripts, executes them and keeps them alive, i.e. it 
restarts the computation whenever the scheduler kills it (but may also 
kill the job if nan's are generated.). MPI and netCDF is supported.

I tailored the script for my needs at linux cluster of NOC, Southampton. 
It relies on LSF scheduler software to check and kill jobs from the queue:

ns0:~$ bjobs -V
Platform LSF HPC 6.0 for Linux/Qsnet, Mar 18 2005
Copyright 1992-2004 Platform Computing Corporation

and

ns0:~$ bkill -V
Platform LSF HPC 6.0 for Linux/Qsnet, Mar 18 2005
Copyright 1992-2004 Platform Computing Corporation

and

ncdump

and also

prun

the last one is to execute parallel jobs.

If you (or someone else) is interested, I can post the script (about 800 
lines).

Cheers,
Lucas








More information about the MITgcm-support mailing list