yade-users team mailing list archive

[Bruno Chareyre <question240608@xxxxxxxxxxxxxxxxxxxxx>]

Question #240608 on Yade changed:
https://answers.launchpad.net/yade/+question/240608

    Status: Open => Answered

Bruno Chareyre proposed the following answer:
Dear Fu,
I am not sure what you mean by "simulation scale". Is it the number of particles?
If so, there is no easy solution. If there was one we would have apply it already.

On your points:
(1)
You can't do much, unless you are prepared for big changes in the core part, I mean months of work 

(2) 
- use a timestep as large as you can, since it will decrease the total number of calculation steps
- for large number of particles, try to decrease the cost of collider by increasing verletDist and targetInterv (also use timing stats to understand how it affects computation time and find a good compromise)
- If admissible in your problem, decrease contact stiffness. It will let you increase the timestep (see 2)

(3)
run the simulation on many cores if your hardware allows that, it is usually effective up to -j4 in quasistatic systems like yours
so a quadcore cpu can help. CPU frequency and memory have only a marginal effect.

There seem to be DEM codes like ESys which are more efficient than yade
for very large problems. I don't have any experience with them but some
of the yade devs/users do. They may be of more help than me on this
point.

-- 
You received this question notification because you are a member of
yade-users, which is an answer contact for Yade.