[lammps-users] Improve parallel effeciency

I have got a problem, when running parallel simulation with boxes having large vacumm spaces, the parallel effeciency is usually very low, because loads of different processors are not balanced. So is there any method to improve the effeciency of such system without changing the box size, thanks!

hi ???,

LAMMPS doesn't have load balancing and from my understanding
it would seem somewhat difficult to implement it without a major
restructuring of the code.

there may be some options to help with improving performance,
but to make some suggestions, i would first need to know more
about your hardware setup and simulations

how "empty" is you system?
can you predict which areas are less dense than others?
what kind of interaction potentials do you use?
how large is your system and what kind of hardware
are you running on?

it would be even better if you could show us a typical breakdown
of the times from a typical run or even provide a (small)
test system that exhibits this behavior.


2010/6/7 xu <[email protected]>:

Axel's questions are all good ones. You can try to
shrink-wrap your box (boundary s) around your atoms
to avoid large empty spaces. You can also use
the processors command to allocate processors more
efficiently for some geometries, depending on where
your empty space is.