comb3 potential in mpi

LAMMPS LAMMPS Mailing List Mirror
1

Dear Sir

This week, I have got access to use the cluster computer. Then I have given below command for using several nodes

module load openmpi/2.1.6
mpirun -np 20 lmp -in input.txt

But I have found it is not faster than my desktop. it is takeing almost 45 mins for 1 step out of 500 steps. I do not have good idea about cluster computer.

Can you help me about this?

2

there is no enough information here to give a qualified answer. most likely you are doing something incorrect that is not a LAMMPS issue, but something that relates to your local machine and thus the first people to talk to would be the local machine’s help desk or user support or system manager.

at the very least, we would need to see the output of:
lmp -h
and the contents of your log file or screen capture output from the batch system to be able to make an educated guess at what the problem could be.

axel.

3

Dear Sir

Thank you very much for your quick response. I have attached the lmp -h file, log file and sbatch file table with several np. If possible, please help me.

lmp_h (21.3 KB)

cluster_info.docx (13.2 KB)

log.lammps (6.89 KB)

4

if you look at the log.lammps file, you see that each run was run across only 1 processor. that means one of two things:

  1. your LAMMPS executable was not compiled with MPI support or 2) it was compiled against a different MPI library than the one you loaded as a module.
    thus rather than running in parallel, you are just running multiple serial copies that will do the same calculation each.

this is not something that can be resolved from remote, but something that you have to do yourself or need to resolve with your local help desk.

axel.