[MITgcm-devel] testreport

[Jean-Michel Campin]

Hi Martin,

I don't have an example ready to use (but I tried the 2 stages,
was working for me but not a strong evidence that it will work
without any problem on sx8):
> testreport -norun (-dd) [ -of OPTFILE -MPI 2 ]
> testreport -runonly [ -of OPTFILE -MPI 2 -command MPIRUN_COMMAND ]

You probably don't need to keep the output dir "tr_sx8_201105??_0"
of the 1rst testreport (=> -dd) and no need for a "-command" option
or "-a ADDRESS" at this stage. 
And regarding the 2nd one:
a) the optfile is not used (I think), but will be recorded in the summary;
 so, if you can use the same one (could fix this later if this is 
 a problem). 
b) if you are able to send an email from the compute node, 
 can add "-a jmc at mitgcm.org"; otherwise, there is an example of a script 
 (run on the head node) than get back (from compute nodes) the output 
 of testreport jobs:
 MITgcm/tools/example_scripts/ACESgrid/aces_test_all

("-MPI 2" is doing the same thing as "-mpi", but in case you 
want to try on more procs, e.g. 6: "-MPI 6" ; but requires to 
change the -command argument)

Let me know if you have problems with this 2-stages testreport.

Cheers,
Jean-Michel

On Wed, May 11, 2011 at 10:38:22AM +0200, Martin Losch wrote:
> Hi Jean-Michel,
> 
> does your scheme of running testreport twice (once for compilation once for running the executables) work? Do you have an example script?
> I think I need this option now for my test on our SX8 that has become so crowded that I sometimes have to wait for a long time so that the test never really finished (that's what happened on this weekend).
> 
> Martin
> 
> 
> _______________________________________________
> MITgcm-devel mailing list
> MITgcm-devel at mitgcm.org
> http://mitgcm.org/mailman/listinfo/mitgcm-devel