speed up the simulation with more "cores" in anuga

[mikilterribile]

I have tried to speed up the simulation with Anuga using "domain = distribute(domain, verbose=True)" at the beginning and " domain.sww_merge(delete_old=True)" at the end but it looks like the speed is the same and that anuga doesn't use multiple cores.
I use a normal private laptop, with 8 cores available.
Thanks in advance,
Best regards,

[stoiver]

@mikilterribile Check out the examples in anuga_core/examples/parallel. First you create an ordinary sequential domain on process 0 and then run the distribute function: domain = domain.distribute() then as you mentioned you can merge all the parallel sww files at the end with a domain.sww_merge() command.

But to enable a parallel run you need to use the mpiexec command. Suppose you script is named run_model.py and you want to use 8 MPI processes, then you need to run the command:

mpiexec -np 8 python -u run_model.py

I suggest experimenting with the optimal number of processes for you machine and your model. I suggest you chose the number of processes so that the size of the partitioned submeshes are no smaller that 1000-2000. Some experimentation is sensible.

We have recently (version 3.2.0) updated some of the evolution code to use openmp. You can try to use openmp instead of MPI.

In this case you would run your script as follows:

OMP_NUM_THREADS=8 ; python -u run_model.py

You could hard code the setting of OMP_NUM_THREADS in your .bashrc file by adding:

export OMP_NUM_THREADS=8

and then every anuga script will use threads (unless you explicitly change that within you script with a domain.set_omp_num_threads() call.).

Let us know if either of these methods help.