Asking the tip of using the command - processors

LAMMPS LAMMPS Mailing List Mirror
1

Hi,

I want to ask a question about tuning LAMMPS efficiency by using the command - processors.
I tried multiple combination of processors commands by using the sample “in.chute.scaled”, but I found out it made no much big difference.

  • NP: 12
  • Atoms: 38400
  • Run: 50000
  • var x=3, y=4
    Loop time
    test1 - Without processors command : 792.607
    test2 - processors * * * grid numa : 793.299
    test3 - processors * * * grid twolevel 4 * * 1: 792.687
    test4 - processors * * 1 map cart: 793.27

Did I choose a bad input sample for testing? Could you please teach me the tip of using this command?
Thanks a lot.

Best Regards,
Linda

2

Hi,

I want to ask a question about tuning LAMMPS efficiency by using the command
- processors.
I tried multiple combination of processors commands by using the sample
"in.chute.scaled", but I found out it made no much big difference.

- NP: 12
- Atoms: 38400
- Run: 50000
- var x=3, y=4

Loop time
test1 - Without processors command : 792.607
test2 - processors * * * grid numa : 793.299
test3 - processors * * * grid twolevel 4 * * 1: 792.687
test4 - processors * * 1 map cart: 793.27

Did I choose a bad input sample for testing?
Could you please teach me the
tip of using this command?

the processors keyword is primarily useful
when you have a load imbalance and when
you run across a large number of multi-core
(NUMA) nodes. with 12 MPI tasks and an
input that has a fairly homogeneous distribution
of particles in space, neither of those conditions
is met, so there is no surprise that you don't
see significant changes.

axel.