[Paraview] vtkGhostLevels

[Berk Geveci]

You are welcome and that is correct.

On Wed, Jul 30, 2008 at 12:50 PM, Dr Gerard Gorman
<g.gorman at imperial.ac.uk> wrote:
> That's really helpful - yhanks for your reply Berk. Can I just confirm
> one thing with you that's not explicitly stated - is vtkGhostLevels a
> cell vtkUnsignedCharArray of name "vtkGhostLevels"? I'm asking this for
> the purpose of writing parallel vtk files. I'm trying to debug why I'm
> seeing internal faces in my parallel files.
>
> Cheers
> Gerard
>
> Berk Geveci wrote:
>> We have been meaning to add a parallel section to the VTK book for a
>> while. I hope that this will happen in the near future. The ParaView
>> book talks about parallelism and ghost levels. Also, there are a few
>> documents here and there that talk about ghost levels. A good one is
>> the handouts from the Supercomputing '07 tutorial found here:
>> http://paraview.org/Wiki/SC07_ParaView_Tutorial
>>
>> There are going to be ParaView tutorials at both SC '08 and IEEE
>> Visualization conferences this year. I would recommend attending these
>> to those that can. Also, we cover parallelism in VTK in the courses we
>> offer. I recommend those as well.
>>
>> The basic idea is the following:
>>
>> - Algorithms that need it request ghost levels from upstream. These
>> are unstructured or general purpose (i.e. work on vtkDataSet or
>> vtkPointSet) algorithms that need extra cells to produce piece
>> independent results. Ghost level request accumulate. If there are
>> three algorithms that each request 1 ghost level, the reader will
>> receive a request for 3 ghost levels.
>> - The reader satisfied this request as best as it can. It marks ghost
>> cells as such by populating a vtkGhostLevel array (cell-centered).
>> This array contains an integer that corresponds to the ghost level of
>> that cell. 0 for internal cells, 1 and higher for ghost cells.
>> - The filter that asked for ghost levels is responsible of stripping
>> away the level it request from its output.
>> - In ParaView, the geometry filter strips away all ghost levels
>> because they are no supposed to be rendered.
>>
>> Note that algorithms that only work with structured data (i.e.
>> vtkImageData, vtkStructuredGrid or vtkRectilinearGrid) do not work
>> with pieces/ghost levels. Instead, they work with structured extents
>> that are index based.
>>
>> I hope this helps.
>>
>> -berk
>>
>> On Wed, Jul 30, 2008 at 7:20 AM, Gerard Gorman <g.gorman at imperial.ac.uk> wrote:
>>
>>> Hi
>>>
>>> I am trying to use vtkXMLPUnstructuredGridWriter but I cannot find any
>>> documentation on how ghost levels are used. I've already asked on the
>>> VTK mailing list but I've not had any response - I'm guessing because
>>> not many people use parallel.
>>>
>>> From grepping the code I'm guessing that I should add an unsigned char
>>> scalar array called "vtkGhostLevels" to indicate which elements/nodes (I
>>> think I need both might be needed depending on the algorithm) are owned
>>> and which are halo/ghost. But all this guessing has so far got me either
>>> double counting elements on the halo or a gap.
>>>
>>> Can someone please shed some light on this?
>>>
>>> Cheers
>>> Gerard
>>> _______________________________________________
>>> ParaView mailing list
>>> ParaView at paraview.org
>>> http://www.paraview.org/mailman/listinfo/paraview
>>>
>>>
>
>