yade-users team mailing list archive

[Bruno Chareyre <question700611@xxxxxxxxxxxxxxxxxxxxx>]

Question #700611 on Yade changed:
https://answers.launchpad.net/yade/+question/700611

    Status: Open => Answered

Bruno Chareyre proposed the following answer:
Hi,
This has 0% chances to work.
MPI uses a domain decomposition with one yade instance running each subdomain.
What happens in your script is that each yade instance has its own 2PFV engine trying to triangulate+solve a flow problem in a subset of particles with no walls, and with no knowledge of what happens in the other subdomains (hence inconsistent triangulations,  no connectivity of fluxes accross subdomains, etc.).
Worth than that, particles can be exchanged between subdomains, so 2PFV may look for positions of particles which are no longer in the scene.

You need to think about what really happens with domain decomposition if you want to understand what can/cannot work.
In this case it would need a specific implementation of 2PFV in order support MPI.

Bruno

-- 
You received this question notification because your team yade-users is
an answer contact for Yade.