Dear Lammps-users,
I installed fftw-2.1.5, openmpi-1.4.3 and Lammps-2Apr2013 in CentOS cluster. I ran a simulation. But it is very slow (time ~ single processor). The log file shows ‘1 by 2 by 20 MPI processor grid’ after reading the data file. I checked the processor number and it shows it is working parallel.
Did you correctly run the simulation spread out over multiple cores and/or nodes, or did you see 40 processes running on a single node? MPI will just spawn 40 threads if you tell it to, even if the machine you run it on only has e.g. 4 cores.
Dear Lammps-users,
I installed fftw-2.1.5, openmpi-1.4.3 and Lammps-2Apr2013 in CentOS cluster.
I ran a simulation. But it is very slow (time ~ single processor). The log
file shows '1 by 2 by 20 MPI processor grid' after reading the data file. I
checked the processor number and it shows it is working parallel.
please have a look into the LAMMPS manual where it explains how
processors are assigned to subdomains of the simulation cell and what
kind of implications this has on the load balance. your 1x2x20
processor grid hints at having a system with a lot of vacuum and thus
you may assign processors in a way that is creating a huge load
imbalance and thus the simulation doesn't scale.
check out the processors command and the balance command.
axel.
oh, and don't use such an outdated version of LAMMPS if you want any
help with more specific problems.