Sigma layers fail in parallel when using flredecomp
Affects | Status | Importance | Assigned to | Milestone | |
---|---|---|---|---|---|
Fluidity |
In Progress
|
Medium
|
Jon Hill |
Bug Description
Email from Gustavo Mastrorocco Marques:
Hi Jon,
Thanks for your help!
I followed your suggestions, but the extruded mesh (using sigma-layers) still looks weird when running in parallel.
This is what I've done: I generated a 2D mesh with gmsh that has 3207 elements. I then used Fluidity to extrude this mesh in the vertical using 12 sigma-layers (total of 38,484 elements) and run the simulation using 4 processors (~ 10,000 elements per processor). The attached figure (mesh-4nodes.png) shows how the mesh looks like.
However, when running in serial, the same configuration gives a reasonable mesh (see attached figure, mesh-1node.png). If I use a z-coordinate the final mesh also looks fine, when running in both serial and parallel.
I could not find any examples that use sigma-layers in Fluidity. Has anyone used this option before? Perhaps the problem is the combination of sigma-layers and flredecomp (or fldecomp)?
I've attached all the files needed to reproduce the problem (files.zip).
Related branches
Changed in fluidity: | |
status: | Fix Committed → In Progress |
Same error as with https:/ /code.launchpad .net/bugs/ 1099943. No idea why sigma layers should be chained. Removing this seems to fix the issue though.