[lammps-users] A question about parallel efficiency


here is a question about parallel task allocation.

I plan to simulate a system in which the physical processes we concern may occur only in a very limited region of the system, and then gradually spread. Within this small region the density may become very high and hence each atom has relatively more neighbors; thus if the simulation runs in parallel, the processor which is unluckily responsible for this region may burden much more amount of calculations and other processors will probably wait for this processor to keep the synchronization of timestep, and the parallel efficiency is lowered in such a case.

Are there any methods in Lammps to improve the efficiency for this case as much as possible?

Thanks and regards.

Xiang Gu

If you can use shrink-wrap boundaries, the processor assignment
will grow with the swelling volume. Otherwise you're out of luck.