Using DoFRenumbering::Cuthill_McKee and DoFRenumbering::component_wise in parallel

22 views
Skip to first unread message

Nihar Bhardwaj Darbhamulla

unread,
Sep 26, 2024, 3:47:57 PM9/26/24
to deal.II User Group
Can we deploy these sequentially after a parallel partitioning of the mesh i.e. if we have an object of type


and we subsequently ask the dof_handler object to distribute the degrees of freedom as follows

dof_handler.distribute_dofs(fe);

can we proceed to renumber the DoFs with Cuthill_McKee and component_wise? Does this create a bandwidth minimized component_wise renumbered sparsity structure for each process.

I have this question because when I'm using the serial version of triangulation, I'm explicitly calling subdomain_wise renumbering first and then Cuthill-McKee and component_wise renumbering on 4 processes, I can see 4 bandwidth minimized structures which might illustrate that the renumbering retained the individual partition identity.

Best,
Nihar

Wolfgang Bangerth

unread,
Sep 27, 2024, 10:47:31 AM9/27/24
to dea...@googlegroups.com
On 9/26/24 13:47, Nihar Bhardwaj Darbhamulla wrote:
>
> parallel::distributed::Triangulation<dim>
> <https://www.dealii.org/current/doxygen/deal.II/classparallel_1_1distributed_1_1Triangulation.html> triangulation <https://www.dealii.org/current/doxygen/deal.II/p4est__wrappers_8cc.html#ace00f2f80d9780ef9aa1007e1c22c6a4>;
> and we subsequently ask the dof_handler object to distribute the degrees of
> freedom as follows
>
> dof_handler.distribute_dofs(fe);
>
> can we proceed to renumber the DoFs with Cuthill_McKee and component_wise?
> Does this create a bandwidth minimized component_wise renumbered sparsity
> structure for each process.
>
> I have this question because when I'm using the serial version of
> triangulation, I'm explicitly calling subdomain_wise renumbering first and
> then Cuthill-McKee and component_wise renumbering on 4 processes, I can see 4
> bandwidth minimized structures which might illustrate that the renumbering
> retained the individual partition identity.

Nihar: I believe that this should work. What happens if you try?

Best
W.

--
------------------------------------------------------------------------
Wolfgang Bangerth email: bang...@colostate.edu
www: http://www.math.colostate.edu/~bangerth/


Reply all
Reply to author
Forward
0 new messages