Hello,
i have a little bit vague question:
I have a computational domain with clearly
discernible inflow. I use parallel::distributed::Triangulation
I would like to use CuthilMcKee renumbering in order
to get some performance boot for preconditioner+solver.
I am able to define
starting_indices In
https://www.dealii.org/8.4.0/doxygen/deal.II/namespaceDoFRenumbering.html#ab938a690bf4e2adff191fe969b0f21d3it is said, that the renumbering occurs on each processor individually.
Can I still get the gradient boost from the renumbering?
Are there other strategies how to get improvement of linear solver
convergence using Dofs renumbering in the distributed case?
Thanks
Marek C