[lammps-users] Performance

Hi,
I am wondering how to get better parallel performance on my
simulation. I am doing langevin on a single worm like chain, 1000
beads long. It is very very slow, and running on multiple CPUs does
not seem to help. Is there a way to run this in parallel? Right now I
have no neighbor list, however nsq did not help, and bin will not run
on a simbox my size. Since there is only one polymer, I doubt it would
anyway.

–Josh

That's not a model/geometry that LAMMPS can parallelize
very well, as it doesn't fill a box.

If all you are doing is 1000 atoms with no pair interactions
and no neighbor list, it should run exceedingly fast on
such a small problem. How many timesteps per second
is it running at on one proc? All you are doing is NVE and bonds?

Steve

Hi,
I am wondering how to get better parallel performance on my
simulation. I am doing langevin on a single worm like chain, 1000
beads long. It is very very slow, and running on multiple CPUs does

how can a simulation with only 1000 beads be slow?

please provide some example files that allow to assess
the situation. what can be done, depends largely on
the details of your simulation setup.

not seem to help. Is there a way to run this in parallel? Right now I
have no neighbor list, however nsq did not help, and bin will not run

i don't understand what you mean by this.

on a simbox my size. Since there is only one polymer, I doubt it would

this doesn't make sense.

there may be a chance to speed things up with threading instead
of domain decomposition, but as i wrote before, it is very system
dependent how much of a speedup can be achieved.

have you checked, in what part of the MD loop lammps is
actually spending all the time?

cheers,
   axel.