Assigning processors to radial wedges

Submitted by mattkesseler on Sat, 04/01/2017 - 18:04

Hi all,

I am currently trying to simulate a granular cylinder slump with the cylinder being removed vertically after initial settlement of the particles. This will result in the particles slumping to the sides and the heap decreasing in height. I do not have the premium version of LIGGGHTS and thus cannot use fix balance to assign processors dynamically as the heap dimensions change, and I can't find a good way of mapping out the processors that doesn't lead to a lot of wasted space. Even if I use processors Px Py Pz grid custom myfile to make an efficient initial array, this would quickly become inefficient due to the changing dimensions of the mass. I had an idea to alleviate this problem in which the processors would be assigned to radial wedges, such that each wedge would always have a fairly uniform number of particles in it. Despite high communication costs due to the large surface area / volume ratio of these wedges, I think this would prove more efficient and would be fairly independent of the initial geometry of the cylinder in terms of how effectively it maintains performance as the particle heap deforms. However, I do not know whether this is possible or how to do this with the grid custom myfile functionality.

Let me know if this is a workable solution or if I need to try anything else.

Kind regards,

Matt Kesseler.

ckloss's picture

ckloss | Sun, 04/23/2017 - 11:50

Hi Matt,

unfortunately, currently there is no implementation of "wegde" BCs in LIGGGHTS

Best wishes
Christoph