| 1 | |
|---|
| 2 | ========================= |
|---|
| 3 | INSTALLING anuga_parallel |
|---|
| 4 | ========================= |
|---|
| 5 | |
|---|
| 6 | MPI |
|---|
| 7 | === |
|---|
| 8 | |
|---|
| 9 | First thing, you need to install MPI on your system. OPENMPI and MPICH2 |
|---|
| 10 | are supported by pypar (see below) so both should be ok. |
|---|
| 11 | |
|---|
| 12 | Make sure mpi works. You should be able to run a program in parallel. |
|---|
| 13 | |
|---|
| 14 | Try something as simple as |
|---|
| 15 | |
|---|
| 16 | mpirun -np 4 pwd |
|---|
| 17 | |
|---|
| 18 | should produce the output of pwd 4 times. |
|---|
| 19 | |
|---|
| 20 | PYPAR |
|---|
| 21 | ====== |
|---|
| 22 | |
|---|
| 23 | We use pypar as the interface between mpi and python. The most recent |
|---|
| 24 | version of PYPAR is available from |
|---|
| 25 | |
|---|
| 26 | http://code.google.com/p/pypar/ |
|---|
| 27 | |
|---|
| 28 | (There is an old version on sourceforge |
|---|
| 29 | http://sourceforge.net/projects/pypar/) |
|---|
| 30 | |
|---|
| 31 | Install pypar following the instructions in hte download. Should be able |
|---|
| 32 | use standard python setup.py install |
|---|
| 33 | |
|---|
| 34 | Make sure the pypar examples work |
|---|
| 35 | |
|---|
| 36 | |
|---|
| 37 | PYMETIS |
|---|
| 38 | ======= |
|---|
| 39 | |
|---|
| 40 | In the anuga_parallel directory there is a subdirectory pymetis. |
|---|
| 41 | |
|---|
| 42 | Follow the instructions in README to install. Essentially just run make. |
|---|
| 43 | |
|---|
| 44 | From the pymetis directory, test using test_all.py |
|---|
| 45 | |
|---|
| 46 | ANUGA_PARALLEL |
|---|
| 47 | ============== |
|---|
| 48 | |
|---|
| 49 | Should now be ready to run some parallel anuga code. |
|---|
| 50 | |
|---|
| 51 | Go back to the anuga_parallel directory and run test_all.py |
|---|
| 52 | |
|---|
| 53 | Hopefully that all works. |
|---|
| 54 | |
|---|
| 55 | Run run_parallel_sw_merimbula.py |
|---|
| 56 | |
|---|
| 57 | First just run it a sequential program. |
|---|
| 58 | |
|---|
| 59 | Then try a parallel run using a command like |
|---|
| 60 | |
|---|
| 61 | mpirun -np 4 python run_parallel_sw_merimbula.py |
|---|
| 62 | |
|---|
| 63 | That should run on 4 processors |
|---|
| 64 | |
|---|
| 65 | You should look at the code in run_parallel_sw_merimbula.py |
|---|
| 66 | |
|---|
| 67 | Essentially a fairly standard example, with the extra command |
|---|
| 68 | |
|---|
| 69 | domain = distribute(domain) |
|---|
| 70 | |
|---|
| 71 | which sets up all the parallel stuff. |
|---|
| 72 | |
|---|
| 73 | Also for efficiency reasons we only setup the original full sequential mesh |
|---|
| 74 | on processor 0, hence the statement |
|---|
| 75 | |
|---|
| 76 | if myid == 0: |
|---|
| 77 | domain = create_domain_from_file(mesh_filename) |
|---|
| 78 | domain.set_quantity('stage', Set_Stage(x0, x1, 2.0)) |
|---|
| 79 | else: |
|---|
| 80 | domain = None |
|---|
| 81 | |
|---|
| 82 | The output will be an sww file associated to each |
|---|
| 83 | processor. |
|---|
| 84 | |
|---|
| 85 | There is a script anuga/utilities/sww_merge.py which provides |
|---|
| 86 | a function to merge sww files into one sww file for viewing |
|---|
| 87 | with the anuga viewer. |
|---|
