[lammps-users] parallel running problem

Dear Axel,

I’m using 1.8.7beta5. The visualizing process is: File–New Molecule, then I select the output file “dump.shear” and file type “LAMMPS Trajectory”. Is it right?
As Steve just mentioned, the log files for this two simulations are indeed different, I doubt the MPI environment is not configured correctly.

Best regards,
Damien
2009-07-21

Dear Axel,

I'm using 1.8.7beta5. The visualizing process is: File--New Molecule,
then I select the output file "dump.shear" and file type "LAMMPS
Trajectory". Is it right?

right. i usually just do: vmd -lammpstrj dump.shear
to load it from the command line, but that is the same thing.
what operating system are you on?

As Steve just mentioned, the log files for this two simulations are
indeed different, I doubt the MPI environment is not configured
correctly.

_how_ different is different. as you can see from the example
outputs shipped with the distribution, the trajectories are diverging
between the serial and parallel case. however, on my machine i get
an almost exact match between a local 4 MPI tasks run and the reference
run and the same for the serial. try comparing those. if the difference
is large, then you most likely have either a miscompiled lammps binary
(more likely) or a messed up MPICH installation (i use OpenMPI and it
has always worked fine for me with lammps). if the difference is small
we have to look closer. since the changes to the lammps molfile parser
were quite substantial, it is quite possible, that there are still
some bugs left.

i'll make some more thorough tests myself.

cheers,
   axel.

Dear Axel,

dear damien,

i don't think that there is anything wrong with your MPICH install.
i've now been able to reproduce your problem and it currently looks
as if there is a bug within lammps that messes up the atom index
order or the atom type flags when run in parallel. this has to be
happening at the very beginning. from then on the output is consistent.
if you load the attached .psf file first and then the trajectory
it seems ok. this will need some more systematic testing...
since all atoms use the same potential in this example, the energies
don't change and also the right atoms get pushed around. will let
you know, if i can track it down. VMD seems to be not the culprit
in this case (sigh).

cheers,
    axel.

shear.psf (148 KB)