On 02/01/2018 04:18 PM, Wolfgang Bangerth wrote:
I am trying to compute a quantify in my finite element code that
requires integrating along a line, starting from a reference cell.
With one processor, this is fine. However, with multiple processors,
there is the chance that due to element partitioning, the line extends
into the region of cells owned by other processors. One option for me
is to increase the size of the ghost cells so that the line will
always be in a region of cells owned by the same processor. However,
dealii ghost cell size is always one.
Yes. But even if there was a way to extend the ghost region, you may
have lines that extend across the entire domain and in that case you may
end up with a situation where one processor would have to know *every*
cell. This is not scalable. So this does not work.
By the way, to make that point more succinctly: What you were trying to
do is *bring the data to the program*. In parallel programming, this is
almost always the wrong approach. What you should be doing is to *bring
the program to the data*. In other words, if you need to compute
something from data that is already there, do the computation on the
processor that has the data.
Best
W.
--
------------------------------------------------------------------------
Wolfgang Bangerth email: [email protected]
www: http://www.math.colostate.edu/~bangerth/
--
The deal.II project is located at http://www.dealii.org/
For mailing list/forum options, see
https://groups.google.com/d/forum/dealii?hl=en
---
You received this message because you are subscribed to the Google Groups "deal.II User Group" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to [email protected].
For more options, visit https://groups.google.com/d/optout.
