Does Collider's performance depend on constitutive law?

Asked by Fabio on 2019-05-23

Hi all,

as the title reveals we are struggling with the question of how the apllied contact law is affecting the computational time of the insertionSortCollider engine. We ran different simulations and made the observation that when using the Cundall model, the engine took about 4.09E10 (44.58%) ms to compute. Hertz-Mindlin only needed 3.32E09 (4.92%) ms. What is the reason for that? As far as I've understood, the engine creates bounding-boxes to detect mutual interactions of all body types. Shouldn't that work out regardless of the contact model? Documentation says (Page 18):

'[...] we want to compute bounds for all bodies in the simulation; suppose we want bound of type axis-aligned bounding box. Since the exact algorithm is different depending on particular shape, we need to provide functors for handling all specific cases'.

We haven't changed any of the bodies within different simualtions, so we would expect the algorithm to remain the same...
In our setup, we are currently working with ~ 45k spherical bodies and a specimen based on ~ 40k facets. If usefull, I can provide you with the code.

Thank you everyone in advance :)

Question information

Language:
English Edit question
Status:
Answered
For:
Yade Edit question
Assignee:
No assignee Edit question
Last query:
2019-05-23
Last reply:
2019-05-23
Jan Stránský (honzik) said : #1

Hello Fabio,

> If usefull, I can provide you with the code.

yes please [1] :-)
There could be many factors we now know nothing about..

thanks
Jan

[1] https://yade-dem.org/wiki/Howtoask

Hi,
No it does not unless different contact model makes the dynamics of the system completely different (faster motion means more collision detection). That's probably your case.
I'd be happy to test your scripts since 44.58% is definitely too much, in general collider should take less than 5% of the total with appropriate settings.
Bruno

Can you help with this problem?

Provide an answer of your own, or ask Fabio for more information if necessary.

To post a message you must log in.