FEAP User Forum
FEAP => Parallel FEAP => Topic started by: blackbird on May 30, 2016, 07:11:11 AM
-
Dear all,
I tried to use parfeap for a really big structure that contains more than 5e5 elements (20 node brick element) with 13e6 nodes. The element generation implements feap standard routines (see file attached). However, after partition, there are 12 input files (as expected), but only the first 3 of them contain elements, i.e. there are 2e5 elements in the first (Iref_0001), 2e5 in the second (Iref_0002) and 1e5 in the third (Iref_0003). All other files (Iref_0004 to Iref_0012) contain only nodal coordinates but the element list is empty (also the header says 0 elements there).
Iref_00* are all larger than 150MB, therefore I only attach the first lines of two representatives...
I would be very greatful if somebody could give a hint or solution on the question of how to obtain the parallel input files properly here.
PS:
- FEAP 8.4 was used
- user material "crys" used, works well in serial computations, maybe commented for testing
-
Can you check if user material affects partitioning, use a standard feap material.
-
The graph partitioning algorithm requires a flat file. So run your file and perform an OUTMesh. They use the generated file to try an partition your problem.
(Note I did not try it myself since your Iref would not run at all for me)
-
Dear Prof. Taylor,
I replaced the Materials by
MATE,1
SOLI
ELAStic ISOTropic 3.24e9 0.35
DENSity mass 1190
MATE,2
SOLI
ELAStic ISOTropic 3.24e9 0.35
DENSity mass 1190
but unfortunately this has no effect. Partitioning into 8 parts result in the output appended, i.e. the first two partitions contain all the elements. So obviously the material is not the problem - I will head for the flat file ...
-
Dear Admin,
indeed, the flat file was the issue. The partitioning of the result of
BATCH
OUTMesh
END
works correctly.
Thanks!