Multiprocessing

[ivandatasci]

The command gmx_MMPBSA_test accepts the argument --num_processors.

Is --num_processors also available for the command gmx_MMPBSA?

[marioernestovaldes]

to run gmx_MMPBSA with multiple processors use (for example 5):

mpirun -n 5 gmx_MMPBSA -i mmpbsa.in ...

[marioernestovaldes]

Did you try with only 2 processors? Why is the "--map-by :OVERSUBSCRIBE" in the command line?

[ivandatasci]

Valid questions.

The --map-by :OVERSUBSCRIBE is my attempt to use more computer power.

My machine has 8 cores which hyper thread to 16. So, if I want to go higher than mpirun -n 8, I need to override it, for example mpirun -n 9 --map-by :OVERSUBSCRIBE.

But no problem, to keep it simple, I am re-trying now, this time sticking to the conventional mpirun -n 8.

It has been trying to index now for 10 minutes but it looks stuck, just as I reported before.

I will kill it now and try with mpirun -n 2 althought, I must say, I really could use something faster than 2X.

I'll report back in a few minutes.

[ivandatasci]

I just tried with mpirun -n 2. It seems stuck as before and I killed it after 10 minutes.

I also tried with mpirun -n 1 and that indexing step completed in about 2 seconds.

In my linux conda environment, I am using gromacs==2024.4=mpi_openmpi*.

Is there anything else that might be useful to understand this bizarre behaviour?

[marioernestovaldes]

@Valdes-Tresanco-MS could you please provide some guidance here?

[Valdes-Tresanco-MS]

If you run it using serial mode does it work or does it also get stuck? If in both cases it fails, then the problem is with gromacs, if not, the problem may be with mpi. Please attach the gmx_MMPBSA.log file for both review cases.

[ivandatasci]

As I waited for a response, tried it with a non-openmpi version of gromacs. In conda I called it simply as gromacs==2024.4.

With that, I can run mpirun -n 8 gmx_MMPBSA -i mmpbsa.in ... successfully. So, your suspicion was right.

Problem solved. Thank you both.