Getting Started with FLAME
69
3. Copy files from Iceberg which include the resulting xml files.
4. Configure the C++ compiler and MPI libraries on Iceberg. This was
done by creating a symbolic link to all required files like cc1plus.
5. Configure Libmboard by running ./configure. The MPI libraries are
linked here.
6. Run a model on Iceberg in parallel by using the following commands:
./xparser path_of_model.xml -p
For example,
./xparser ../model/turningKernel.xml -p
make CC=/usr/local/packages5/openmpi-gnu/bin/mpicc
LIBMBOARD_DIR=/home/ac1mk/libmboard export
LD_LIBRARY_PATH=\$LD_LIBRARY_PATH:/usr/local/packages5/
openmpi-gnu/lib
Run on parallel nodes by submitting a job.sh file which contains the
number of nodes and the time you specify to it.
#!/bin/sh
#\$ -l h_cpu=0:30:00 (Specifies how long the job is likely
to last}
#$ -cwd
#$ -pe openmpi-ib 8 (Specifies the openmpi-ib
library should be used, with 8 cores)
#\$ -q parallel.q (Specifies the queue to be used)
#\$ -v SGE_HOME=/usr/local/sge6_2 (Specifies the path)
/usr/mpi/gcc/openmpi-1.2.8/bin/mpirun
/home/ac1mk/trial2/main 100 trial2/output/p8/t100k/0.xml -r
On a Mac system this could be done directly by
mpirun -np numberOfNodes ./main numberOfIterations 0.xml -r
mpirun -np 16 ./main 100 0.xml -r
4.6 Integrating with More Libraries
FLAME uses C, XML and interacts with generated files. The models can
be enhanced by using the same principle and working with additional libraries.
For example,