Decomposition algorithms in LAMMPS?

Hi,

When I use the simulation box for my system, it has a large amount of free space where there is no atom. According to LAMMPS design, it uses spatial decomposition algorithm for parallelization. So, there will be significant idle processors and it slows down the simulation. I am thinking that the way to get around this problem is using atom-decomposition or force-decomposition. But these options are not available in LAMMPS?!

I would be very thankful if you let me know about your comments or suggestions on this problem?
Best regards

Have you tried load balancing?

http://lammps.sandia.gov/doc/balance.html

Shawn Coleman

Hi,

When I use the simulation box for my system, it has a large amount of free space where there is no atom. According to LAMMPS design, it uses spatial decomposition algorithm for parallelization. So, there will be significant idle processors and it slows down the simulation. I am thinking that the way to get around this problem is using atom-decomposition or force-decomposition. But these options are not available in LAMMPS?!

Your statement doesn’t hold as it is. You obviously didn’t study the documentation properly.

Lammps has a lot of options to improve load balance.

You can assign a suitable processor grid, you can adjust the domain sizes, you can use multi-threading (which uses atom decomposition), you can use recursive bisectioning for domain decomposition.

Please re-read the manual more carefully and don’t make claims that won’t hold without proper proof.

Thanks,
Axel

Hi Shawn,

I have not tried this. I will look into that.

Thanks for your reply

​Hi Axel,

Thanks for your explanations. I will look into these options.

Best​