11. Parallel CommandsΒΆ
The parallel commands are currently only working in the Linux version. The parallel OpenSeesPy is similar to OpenSeesMP, which requires users to divide the model to distributed processors.
You can still run the single-processor version as before. To run the parallel version, you have to install a MPI implementation, such as mpich. Then call your python scripts in the command line
mpiexec -np np python filename.py
where np
is the number of processors to be used,
python
is the python interpreter, and
filename.py
is the script name.
Inside the script, OpenSeesPy is still imported as
import openseespy.opensees as ops
Common problems:
- Unmatch send/recv will cause deadlock.
- Writing to the same files at the same from different processors will cause race conditions.
- Poor model decomposition will cause load imbalance problem.
Following are commands related to parallel computing:
- getPID command
- getNP command
- barrier command
- send command
- recv command
- Bcast command
- setStartNodeTag command
- domainChange command
- Parallel Plain Numberer
- Parallel RCM Numberer
- MUMPS Solver
- Parallel DisplacementControl
- partition command