Sorry for the last message, it was not finished yet - i
unintentionally pressed the send button.
Once again:
Hello,
I haven't tried the suggestion of Derek yet. Instead, I
wanted to try to split a smaller mesh with nemesis and see
if my code runs in parallel
(libmesh configure was done with
--enable-parmesh)
I split a smaller (about 1 000 000 nodes) mesh with the
following code:
---------------------------------------
//CODE 1
Mesh mesh;
TetGenIO TETGEN(mesh);
TETGEN.read("meshfile.ele");
set_subdomains(&TETGEN, mesh); // I manually set
subdomain_id
set_boundary_ids(mesh); // I manually set boundary_id
mesh.find_neighbors();
MetisPartitioner().partition(mesh,4);
mesh.print_info();
Nemesis_IO NEM ( mesh );
WriteMesh[actual_pulse]<<"main_alm.nemesis";
//END CODE 1
---------------------------------------
After this I get 4 files (running on 4 processors)
xxx.4.0, xxx.4.1, xxx.4.2, xxx.4.3 where xxx stands for the
file name.
Now, in a new program I want to carry out my calculations:
---------------------------------------
// CODE 2
Mesh new_mesh;
Nemesis_IO NEMESIS (new_mesh);
NEMESIS.read("xxx");
if (!new_mesh. is_prepared ())
new_mesh.prepare_for_use();
new_mesh.print_info();
// MeshRefinement meshrefine (new_mesh);
// meshrefine.uniformly_refine(1);
// new_mesh.print_info();
EquationSystems new_equation_systems (new_mesh);
TransientLinearImplicitSystem & T_system =
new_equation_systems.add_system<TransientLinearImplicitSystem>
("Diffusion");
T_system.add_variable ("T", FIRST);
T_system.attach_assemble_function (assemble_cd);
T_system.attach_init_function (init_geothermal_gradient);
new_equation_systems.init ();
new_equation_systems.print_info();
// ********** TIME LOOP *******
int t_write = 0;
std::cout << "t_step "<<t_step<<"
t_next_pulse[actual_pulse] " <<
t_next_pulse[actual_pulse]<<" dt "<<dt<<std::endl;
for (; t_step < t_next_pulse[actual_pulse]/dt; t_step++)
{
time += dt;
new_equation_systems.parameters.set<Real> ("time") =
time;
TransientLinearImplicitSystem& T_system =
new_equation_systems.get_system<TransientLinearImplicitSystem>("Diffusion");
*T_system.old_local_solution =
*T_system.current_local_solution;
new_equation_systems.get_system("Diffusion").solve();
}
// END CODE 2
---------------------------------------
The assemble function is similar to the one in example 9.
At every time step I get the following warning:
Warning: This MeshInput subclass only support meshes which
have been serialized!
[0] /home/ThermoPaine/libmesh_svn/include/mesh/mesh_input.h,
line 148, compiled Oct 30 2011 at 11:09:26
Warning: This MeshOutput subclass only support meshes which
have been serialized!
[0]
/home/ThermoPaine/libmesh_svn/include/mesh/mesh_output.h,
line 181, compiled Oct 30 2011 at 11:09:26
-----------
QUESTION 1:
-----------
What does this warning mean? Can I simply ignore it? I would
have said that there is still a problem with my
configuration for parallel???
-----------
QUESTION 2:
-----------
There is another thing which disturbs me:
When I partition the mesh with the first code, I get
n_subdomains()=4
n_partitions()=4
n_processors()=4
in mesh.print_info().
But, after reading the mesh with the second code, I get
n_subdomains()=4
n_partitions()=1
n_processors()=4
-----------
QUESTION 3
-----------
In several discussions I read that there are still serious
bugs in ParallelMesh. Do these bugs also concern the code I
have above? AFAIK there are bugs with adaptive coarsening
and refinement. In my code I don't use the class
ParallelMesh - so I was wondering if the data-structure,
which I get with --enable-parmesh, allows refinement and
coarsening.
So, although I read in 1 file per processor the
n_partitions() has changed.
Thank you,
Robert
------------------------------------------------------------------------------
RSA® Conference 2012
Save $700 by Nov 18
Register now!
http://p.sf.net/sfu/rsa-sfdev2dev1
_______________________________________________
Libmesh-users mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/libmesh-users
