I apologize for the vagueness of this question, but was wondering if anyone else was having difficulty running two mpi jobs on the same node. Specifically some sort of memory problem perhaps? I have observed locally on my mac that mpi lammps is very greedy with memory such that it made running any other application impossible.
thanks
matthew
I apologize for the vagueness of this question, but was wondering if anyone
else was having difficulty running two mpi jobs on the same node.
LAMMPS doesn't care. how LAMMPS tasks are distributed across physical
hardware is handled by the MPI library.
Specifically some sort of memory problem perhaps? I have observed locally on
my mac that mpi lammps is very greedy with memory such that it made running
any other application impossible.
the memory consumption of LAMMPS depends largely on the kind of
calculation you are doing, the size of the system and which additional
features you are using.
the most common cause for unexpected large memory requirements as you
are alluding to are so-called "memory leaks" due to incorrect
modifications or custom styles. a large chunk of the core code of
LAMMPS is regularly tested for memory leaks and over the last ~2 years
the LAMMPS developers have systematically removed smaller memory leaks
from the code base. there is currently one known unfixed memory leak,
that is difficult to expose and even more difficult to correct.
this also has nothing to do with MPI, unless there is a memory leak in
your specific MPI library version.
axel.
There was a bad memory leak I had once on a work station that I think went away when I updated the MPI library. Try updating your MPI library.