Problem with disjoint Subdomains in parallel

Clearly defined bug reports and their fixes
Post Reply
Franz Pichler
Posts: 196
Joined: 29 Sep 2011, 12:25
Antispam: Yes

Problem with disjoint Subdomains in parallel

Post by Franz Pichler »

Hello,

i think i found a bug when dealing with disjoint subdomains in parallel. I attach a minimal example to show the problem.
In the example there is a box domain in which 10 non touching spheres are enclosed. I partitioned that with ElmerGrid and the "-metis 4 3" optiion.
In the example i use elmers heat solver, but i figured out the problem with my own solvers.
It would be nice if someone could confirm the problem that not all proccesses enter the solver. They get stuck somwehere in the MainUtils part before the Call of the singlesolver.

My Guess:
I think the problem comes down to the fact that when disjoint subdomains are partitioned so that the parenv%isneighbour fields are not forming a "chain" through all partitions then not all the partitions start the solver in the right manner. With chain i mean that the neighbourlist of partition 0 and 1 is for example FTFF and TFFF. So both of them have no connection to processor 2 and somewhow this messes up the parallel enviroment,

best regards
Franz

Sorry about hte form of the attachment but the mesh.elements was to big to be included in the second zip file. Just copy the mesh.elements file into the mesh directory.
Attachments
ELMERCASE.zip
The case for the report
(825.32 KiB) Downloaded 445 times
ELEMENTS.zip
The mesh.elements file, it was to big to include with the rest
(1.73 MiB) Downloaded 448 times
Post Reply