[Wien] Hardware Benchmarks for lapw1c

[Florent Boucher]

Normally, even if the MPI version is compiled you can run a parallel 
lapw1 calculation without any recompilation. You just have to be careful 
in the way you prepare the machine file (one line per process with 
1:localhost), the option you use in $WIENROOT/parallel_options, the 
definition of usermote in the lapw1para script.


            if($number_per_job[$p] == 1 ) then
              if ($useremote == 1) then
                    ($remote  $machine[$p] $t $exe ${def}_$loop.def;rm 
-f .lock_$lockfile[$p]) >>.time1_$loop &
              else
                  (cd $PWD;$t $exe ${def}_$loop.def;rm -f 
.lock_$lockfile[$p]) >>.time1_$loop &
              endif
            else
              set ttt=(`echo $mpirun | sed -e 
"s^_NP_^$number_per_job[$p]^" -e "s^_EXEC_^$WIENROOT/${exe}_mpi ${def
}_$loop.def^" -e "s^_HOSTS_^.machine[$p]^"`)
              (cd $PWD;$t $ttt;rm -f .lock_$lockfile[$p]) >>.time1_$loop &

If you have 8 k points, I am expecting something like 130-140% total CPU 
time on your 8 cores by running 8 para process with OMP_BUM_THRERADS=1 
that is at least much better than using OMP_NUM_THREADS=8
Regards
Florent

 

-- 
 -------------------------------------------------------------------------
| Florent BOUCHER                    |                                    |
| Institut des Matériaux Jean Rouxel | Mailto:Florent.Boucher at cnrs-imn.fr |
| 2, rue de la Houssinière           | Phone: (33) 2 40 37 39 24          |
| BP 32229                           | Fax:   (33) 2 40 37 39 95          |
| 44322 NANTES CEDEX 3 (FRANCE)      | http://www.cnrs-imn.fr             |
 -------------------------------------------------------------------------