|Reported by||Andreas Dedner (A.S.Dedner@warwick.ac.uk)|
|Reported at||Dec 5, 2009 22:17|
|Operating System||Unspecified / All|
|Last edited by||Martin Nolte (email@example.com)|
|Last edited at||Feb 13, 2011 21:18|
It was decided to add a
boundarySegmentIndex to the interface.
There are a few questions especially for parallel grids still open:
- is the boundary index local, i.e., zero starting consecutive on each processor?
- how does load balancing work? If boundary ids are user data they have to be communicated during load balancing, i.e., some suitable method on the datahandle is required. Also if the boundarySegment index are non persistent there must be a way to attach it persistently to the grid during load balancing.
- how do I obtain the size of the new boundary segment index set? Also for serial grids.
I tried to find answers to these questions in Oliver's additions to the grid howto - but I might have missed it.
- I guess it should be ''local''
- before load balancing one could attach the boundaryId to the corresponding
codim 1 face using
idSet.id(*inter.inside(), inter.numberOfInside(), 1)Then gather/scatter on codim 1 entites (of course then called on all codim 1 entities). And then move them from the map to a vector based on the new boundarySegmentIndex
- add a method boundarySegmentIndexSize to the grid
Perhaps we should postpone the introduction of the boundarySegmentIndex method to after 2.0 and until we have decide on a general way to attach data to intersections. A special index set providing indices for intersections in general and a second index for intersections at the boundary. I would not like to add a new method to the intersection and then have to deprecate it right away because we decide to introduce a new index set.