particle INSERTION depends on PARALLELIZATION?!

Submitted by rob1 on Fri, 01/29/2016 - 18:47

Hello everybody,
I am a new user of LIGGGHTS and I am running simulations by inserting a particle packing in a fixed volume:

insert/pack ... particles_in_region 100000

Of course, as I expect, I have a "less inserted than required" since less than 100000 particles are needed to fill the volume.

My problem is that when I parallelize I have a different number of particle inserted, depending on how many processors I am using:
simulation 1 - 4 cores - 27368 particles
simulation 2 - 8 cores - 27288 particles
simulation 3 - 16 cores - 27149 particles

Please, may you help me in solving this problem?
In addition, when I specify e.g. particles_in_region 27250, the inserted particles are never the same number!

Thanks a lot for your help,
rob1

Forums:

Daniel Queteschiner | Mon, 02/01/2016 - 09:34

Particle insertion is indeed dependent on the processor grid. To avoid costly communication overhead, LIGGGHTS does not insert particles near the processor boundaries. The more processors you use the more boundary regions are present which slightly reduce the total space for particles to be inserted. As you observed this becomes critical for very dense insertion packs (or streams). The simple sequential inhibition (SSI) algorithm that LIGGGHTS uses is simply not able to create a dense packing. Hence, if you need a dense packing you will have to compress the particles after insertion, e.g. by using the fix mesh/surface/stress/servo command.

Especially when using a poly-disperse particle distribution you should also be aware that as soon as LIGGGHTS inserts less particles than required, the size distribution will be different from the one specified. This is because LIGGGHTS starts inserting particles with the largest radius and then continues with the next smaller particle size. When there's no more space left for more particles of the current size, LIGGGHTS will simply stop insertion - no matter which particle size of the distribution currently comes in (and even though smaller particles of the distribution would still fit into the insertion region).

NTT1508 | Wed, 04/27/2016 - 09:15

Hi Daniel,

Can you suggest some tips to optimize the insertion step which is usually time consuming. For example it took me nearly 18 hours to insert 20,000 particles with radius from 1mm to 0.001mm.

Thank you,
Regards,