Launching the solver
Once the grid has been preprocessed and partitioned and the NSU3D input file has been constructed, NSU3D is invoked as:
mpirun -np /nsu3d input.nsu3d
or alternatively as
mpirun -np /nsu3d input.nsu3d > nsu3d.out
mpirun -np /nsu3d input.nsu3d |tee nsu3d.out
using standard Unix utilities for storing the terminal output to a file.
Note that to run jobs using mpi (message passing interface) either mpirun or mpiexec can be used, depending on the system setup.
<nproc> specifies the number of processors/cores on which nsu3d is to be run.
This number must be consistent with the number of partitions in the grid file being read in (and specified in the input.nsu3d parameter input file).
<nproc> must be either equal to the number of partitions (fastest run time) or a factor of the number of partitions, in which case nsu3d will assign an equal number of partitions to each cpu or core.
For example, if the grid is divided into 16 partitions, and resides in the directory grid.part.16, then values <nproc>=16,8,4,2 or 1 can be used (provided enough memory is available on each compute node). On the other hand, <nproc> values such as 32, 17, or 15 are not allowed and will cause the solver to stop execution.
The next argument to mpirun or mpiexec is the executable name for nsu3d, which may require a complete pathname to be visible from the location where mpirun is being invoked.
NSU3D is typically compiled with the mpi libraries and should be run as shown above. However, it is possible to compile NSU3D without the mpi libraries (see example) in which case it should be run as a regular executable i.e.:
Running NSU3D requires a parameter input file, which is called input.nsu3d in this example. The parameter input file is specified by listing the file name directly after the nsu3d executable name. A description of the input parameter file and sample input files can be found here: Solver Input File