Page 1 of 1
Batch option of dinver command line
Posted: Tue Jun 28, 2011 10:11 am
by akoehler
Hi Marc,
Just upgraded dinver to the latest version. I have seen that the "-batch" option of command line dinver is no longer available. Instead one can use "-optimization". Is it still possible to start several processes at the same time and create report files like *_0?.report ?
Regards
Andreas
Posted: Wed Jul 06, 2011 2:50 pm
by admin
Hi Andreas,
No the philosophy has changed. Each dinver instance runs a single inversion at a time spread over several processing cores instead of several inversions. The current 'testing' release does a not so bad job when parallelizing the inversion. The 'devel' release is now using all cores nearly at their full speed. It is currently available only through git repository.
It you want to generate several runs with different seeds, you must write a small bash script to start dinver sequentially. If you are not using option '-j' to set the maximum number of cores, the default spreads over all. So do not start several dinver instances at the same time. Wait for the last to complete before starting the next one.
Code: Select all
N_RUNS=5
for (( i=0; $i<N_RUNS ; i=$i+1 )); do
echo "------- run $i --------"
dinver <YOUR OPTIONS> -o run_$i.report | tee run_$i.log
done
The seed is loosing significance when dealing with parallel computing. In fact, even with the same seed you cannot have exactly the same run except if you specify '-j 1'. With the 'devel' release seed is even hidden from user interface (either command line or GUI).
Cheers,
Marc