Skip to content. | Skip to navigation

Personal tools
Log in
Sections
You are here: Home Developers Tools LMDZ testbench on Vargas

LMDZ testbench on Vargas

The script used to ru a "testbench" of a given revision of LMDZ on Vargas

The multibench_vargas.bash script is on my account on Vargas  (Power6 at IDRIS):

/homegpfs/rech/dzt/rdzt206/bench_vargas/multibench_vargas.bash

It basically retreives a given revision of LMDZ, compiles it in 'seq', 'omp', 'mpi' and 'mpi_omp' modes and for each case runs a test 32x24x11 benchmark simulation (same test as in the install.sh script)

In practice, all you need to do is run the script (you might have to modify the LMDZ5BRANCH variable in the script to select the branch you want to test; the current value is trunk) :

./multibench_vargas.bash > multibench_vargas.bash.out 2>&1

Output messages report about the various steps, including compilation of the model, and include the last lines sent to standard output from the run (which should end with a neat "Everything is cool" messge if all went well); for instance:

.....

OK, compiled mpi version of GCM
x BENCH32x24x11
x BENCH32x24x11/callphys.def, 90 bytes, 1 media blocks.
x BENCH32x24x11/gcm.def, 2057 bytes, 5 media blocks.
x BENCH32x24x11/init.x, 5282 bytes, 11 media blocks.
x BENCH32x24x11/limit.nc, 8504052 bytes, 16610 media blocks.
x BENCH32x24x11/physiq.def, 2407 bytes, 5 media blocks.
x BENCH32x24x11/planete.def, 595 bytes, 2 media blocks.
x BENCH32x24x11/run.def, 597 bytes, 2 media blocks.
x BENCH32x24x11/start.nc, 308628 bytes, 603 media blocks.
x BENCH32x24x11/startphy.nc, 384584 bytes, 752 media blocks.
x BENCH32x24x11/traceur.def, 42 bytes, 1 media blocks.
x BENCH32x24x11/bench.sh, 227 bytes, 1 media blocks.
x BENCH32x24x11/supcrit.data, 10 bytes, 1 media blocks.
x BENCH32x24x11/parameter_mix.data, 45 bytes, 1 media blocks.
x BENCH32x24x11/wake.data, 26 bytes, 1 media blocks.
x BENCH32x24x11/bench.sh~, 227 bytes, 1 media blocks.
running mpi bench ...
last line of gcm output is:
Everything is cool

....

At the very end of the script, files 'restart.nc' et 'restartphy.nc' of the simulations are compared (binarily) to test the coherence between runs in the different modes; one then obtains for instance :

Comparison between outputs from the runs:
-----------------------------------------
seq and mpi restart.nc files are identical
seq and mpi restartphy.nc files are identical
!! seq and omp restart.nc files are different !!
!! seq and omp restartphy.nc files are different !!
!! seq and mpi_omp restart.nc files are different !!
!! seq and mpi_omp restartphy.nc files are different !!
!! mpi and omp restart.nc files are different !!
!! mpi and omp restartphy.nc files are different !!
!! mpi and mpi_omp restart.nc files are different !!
!! mpi and mpi_omp restartphy.nc files are different !!
omp and mpi_omp restart.nc files are identical
omp and mpi_omp restartphy.nc files are identical

 

The script can be run with an optional argument, the LMDZ revision number to test. Thus, with :

./multibench_vargas.bash 1546 > multibench_vargas.bash.out 2>&1

it is revision 1546 that is tested.

29 June 2011


 

Recapitulations of the results obtained with this script for various branches can be found here.

 

13 December 2010

Add comment

You can add a comment by filling out the form below. Plain text formatting. Comments are moderated.

Enter the word