Dears
I have a problem with multiprocessor and ReaxFF.
I am trying to run a simulation of big box on 4 nodes HPC that each contaning 8 cores. I used ReaxFF potential and used LAMMPS on windows.
the command I used is “mpiexec -n 32 -machinefile graid32.txt lmp_mpi -in input.in”.
but the memory usage increasing very rapidly a few seconds after running the simulation and the simulation stopped. the error is “process 0 exited without calling finalize”. this error showd on screen for 4 cores of each node.
after this, when I used "“mpiexec -n 64 -machinefile graid64.txt lmp_mpi -in input.in” command, the memory usage increased slowly and simulation continued running for a short time of 200 or 1000 steps and stopped again with the previous error. (run time is different at repeating simulation but is that is short)
When I’d halved the size of the simulation, there will be no problem running.
1- Why the assumption of 16 cores per node improves runs a temporary situation, but in real there is 8 cores per node.
2- please advise me to solve this problem and run the simulation.
Thanks
Mohammad
Dears
I have a problem with multiprocessor and ReaxFF.
I am trying to run a simulation of big box on 4 nodes HPC that each
contaning 8 cores. I used ReaxFF potential and used LAMMPS on windows.
the command I used is "mpiexec -n 32 -machinefile graid32.txt lmp_mpi -in
input.in".
but the memory usage increasing very rapidly a few seconds after running
the simulation and the simulation stopped. the error is "process 0 exited
without calling finalize". this error showd on screen for 4 cores of each
node.
after this, when I used ""mpiexec -n 64 -machinefile graid64.txt lmp_mpi
-in input.in" command, the memory usage increased slowly and simulation
continued running for a short time of 200 or 1000 steps and stopped again
with the previous error. (run time is different at repeating simulation but
is that is short)
When I'd halved the size of the simulation, there will be no problem
running.
1- Why the assumption of 16 cores per node improves runs a temporary
situation, but in real there is 8 cores per node.
2- please advise me to solve this problem and run the simulation.
there is not enough information here to give meaningful advice.
please always state what version of LAMMPS you are using.
you mention windows and you mention a "4 nodes HPC". but it is not clear,
whether your HPC cluster is running windows of not. if not, what OS does it
run on? if yes, are you using a pre-compiled version of LAMMPS, or did you
compile/install it yourself?
how did you measure memory consumption?
how large is your "big box"?
what does LAMMPS say, how much memory it would be using in the LAMMPS log
file?
what *other* computations do you have listed in your input deck besides
just running an MD or minimization?
axel.