Geometry Dynamics and Parallel Processing

Submitted by marc on Thu, 12/08/2011 - 19:01

Hi There,

I was doing a few tests with LIGGGHTS and I saw that I can't do 2 dynamics of the same style at the same time for the same geometry. I'm trying to use 2 wiggles (move/mesh/gran) for the same geometry, but it looks like the second command overwrites the first one, so I can't have the dynamics I was expecting.

I'm trying to simulate a vibrating screen, where an eccentric is responsible for it's vibration. In this case I have to start a period translation in the X direction, and in the middle of the first oscillation I have to start a periodic translation in the Y direction. The vibrating screen needs to have a trajectory like an ellipse.

My question would be: Is there a way to use 2 dynamics of the same style at the same time for the geometry, or should I go for variable style and try to adapt for my case?

Also I was interested in doing tests in a cluster, using distributed memory parallel. Does anyone has done something similar that could share some basic instructions and a simple setup case?

Thanks a lot in advance for the help.

Best Regards,

Marc.

Forums:

ckloss's picture

ckloss | Fri, 12/09/2011 - 14:14

>> I can't do 2 dynamics of the same style at the same time for the same geometr
Correct, you will have to modify the code or use the "variable" style

>>Does anyone has done something similar
I think most of the people here work with clusters

>> that could share some basic instructions and a simple setup case?
there is nothing different about running a case on a cluster or a workstation. If there are thing that are entire specific to your cluster ( a queue stystem), then ask your admin

Christoph

moritzhoefert | Fri, 12/09/2011 - 17:24

Hi Marc,

the head of your input script might look as follows:
atom_style granular
atom_modify map array
boundary f f f
processors 20 1 1
newton off
communicate single vel yes
units si

In that case you split the domain in 20 subdomains in x direction. Check the doc for details about the processor class. In order to start a simulation you need to know details about mpi on your cluster. Ask your admin about that. For parallel simulations start liggghts with
mpirun -np 20 -machinefile hostnames lmp_1.5.0 -in settings.lmp
instead of
lmp_1.5.0 -in settings.lmp

settings.lmp is your input script and hostnames is a file containing information about which nodes are used for your simulation.

I hope that helps. Have a good one,
Moritz