Thank you, Wolfgang, problem solved. But in a much simpler way. My basis 
object is not easily serializable so I have to recreate it and destroy the 
old one anyway (for now) once cells are moved to other nodes. So I do not 
send send them but rather create them and work with them on each node (if 
the corresponding cell is locally owned) and just collect some serializable 
information on a master process with mpi_gather.

I have a little technical question though: Can I attach the 
mpi_communicators that you use in the deal.ii tutorials also somehow to a 

Never tried it but according to the documentation it seems possible.


On Friday, August 23, 2019 at 6:35:18 PM UTC+2, Wolfgang Bangerth wrote:
> On 8/23/19 7:33 AM, Konrad wrote: 
> > 
> > Thanks, I see this class makes it possible to identify cells globally in 
> a 
> > distributed triangulation. I don't see though how the objects in the 
> std::map 
> > are distributed among cluster nodes as the cells are. Am I missing 
> something here? 
> Yes, the data transfer is difficult because you need to figure out who 
> owns a 
> cell for which you have an entry in your local map after 
> refinement/rebalancing. Then you need to pack the data, send it to the new 
> owner, and possibly receive information about other cells you now own 
> after 
> refinement/rebalancing. 
> You might want to look at how other classes do this that store non-trivial 
> data on each cell. An example is ContinuousQuadratureDataTransfer. 
> Best 
>   W. 
> -- 
> ------------------------------------------------------------------------ 
> Wolfgang Bangerth          email:        
> <javascript:> 
>                             www: 

The deal.II project is located at
For mailing list/forum options, see
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 view this discussion on the web visit

Reply via email to