migration seg. fault after mutiple part distribution
|Assignee:||E. Seegyoung Seol||% Done:|
migration procedure doesn't run after multiple part distribution.
distributed mesh? NO # parts per proc: 2 # migration steps: 10
do ghosting? NO
write mesh files? NO
MPIHOME/bin/mpirun -np 8 singlePart_parallel ../meshes/cube_16K.sms out.sms 0 10 2 0 0
#2 Updated by E. Seegyoung Seol over 5 years ago
The quick fix for this error is put in r3423.
However, this issue is still open since we've noticed seg. fault in some cases (test/parallel/main.cc which does aggressive testing by doing migration with randomly picked partition object while incresing #parts from 1 to 19 gradually.
Please be noted that the API for FMDB_Mesh_Migr changed now to take both "global" source part id and "global" dest. part id for each partition object (entity and p-set).
#3 Updated by Cameron Smith over 5 years ago
- File 16_512_slacTDR_RGN added
With ParMA r133 ran on a 512 part mesh, 32 parts per process on 16 processes, a seg fault occurs at line 788 of mPart.cc in the call to FMDB_Mesh_DspNumEnt(...) on line 179 after migration. Attached is a log file from the run. FMDB r3428 was compiled in debug mode.
The input mesh is here:
#4 Updated by E. Seegyoung Seol over 5 years ago
- Status changed from In Progress to Resolved
Multi-part migration is now in service. To use multi-part, call
int FMDB_Mesh_SetNumPart (pMeshMdl mesh, int numPart);
before migration or global partitioning. For more information, please refer to the User's Guide or the sample code located in $FMDB_SRC/test/parallel.
Also, please let us know any further errors and issues you encounter. Thanks for your patience.