[Paraview] [XDMF] testing collections in parallel

Jerry Clarke clarke at arl.army.mil
Thu Sep 4 12:19:34 EDT 2008


Renato,

Answering from bottom to top .....

Jean's problem seems to be twofold. The first one was when running in 
parallel, internally XDMF was allocating memory for the entire dataset
way down in XdmfArray.cxx before re-sizing it to the proper size for the 
  piece. This has been fixed and checked in.

The second seems to be something to do with how the vtkXdmfReader is
keeping track of the vtkInformations from the vtk pipeline. This is
partially fixed. Dave Demarle just checked in a fix.

cvs up, and try again ... I'm going to try it here on some data.

Jerry

Renato N. Elias wrote:
> 
> Jerry,
> 
> I've been doing some tests with XDMF and it seems that, now, the 
> "spatial collection of temporal collections" (I will refer this 
> combination as type 1 for simplicity) is working better than the 
> "temporal collection of spatial collections" (type 2). I've implemented 
> both and when I load in the type 1 the ParaView filters work just fine. 
> I can use the calculator, isocontour, volume rendering, D3, etc... but, 
> when I try the type 2, although the model is correctly loaded and 
> rendered, some filters present a problematic behavior. D3 reports some 
> kd-tree errors, the calculators does not compute anything, volume render 
> closes the connection with the server. Maybe it's also related to the 
> memory allocation and model extent problems reported by Jean sometime 
> ago. Have you found anything? Can I do anything to help in finding the 
> problem? I'm collecting some files for test and could send you later.
> 
> regards
> 
> Renato.
> 
> p.s.: I can't say nothing about parallel efficiency since the models I'm 
> using are not so large and the time series are small as well.



More information about the ParaView mailing list