[vtkusers] PStructuredGrid questions and ghostlevels

Kozdon, Jeremy (CIV) jekozdon at nps.edu
Thu Jun 27 22:58:14 EDT 2013


Thanks for the info George. This what I figured and I guess there is no free lunch. I had been trying to avoid the communication on my side since some of the fields I output are never communicated otherwise and had also been trying to avoid having all the processors write to the same file since I have had MPI I/O issues on some systems. I guess I will implement a few different things and can always postprocess the files if necessary.

From: George Zagaris <george.zagaris at kitware.com<mailto:george.zagaris at kitware.com>>
Date: Thursday, June 27, 2013 6:34 PM
To: Jeremy Kozdon <jekozdon at nps.edu<mailto:jekozdon at nps.edu>>
Cc: Vtkusers <vtkusers at vtk.org<mailto:vtkusers at vtk.org>>
Subject: Re: [vtkusers] PStructuredGrid questions and ghostlevels

Hi Jeremy,

To the best of my knowledge, partition interfaces in VTK need to abut.

Although I do understand the reasoning of what you are trying to achieve, from a finite difference code perspective, leaving a gap is problematic b/c it creates an implicit cell that could span two or more partitions/zones. For most post-processing/visualization operations, the algorithms need to loop over cells, e.g., consider contouring or interpolating the data over a line. Not having an explicit definition of all the cells in a partition makes this more complicated -- it would require building that information when reading in the partitioned grid, e.g., by growing the grid to the right and communicating the data (assuming we run in parallel).

More importantly, VTK structured grids can also store values at cell-centers. The output scheme that you describe would only work for node-centered data.

So, to best support node-centered and cell-centered data, partition interfaces need to abut.

Hope this helps.

Best,
George



On Thu, Jun 27, 2013 at 4:04 PM, Kozdon, Jeremy (CIV) <jekozdon at nps.edu<mailto:jekozdon at nps.edu>> wrote:
Hello,

I am trying to get vtk to properly work with a parallel finite difference code. What I would like to do is have each processor write the data they "own" to a separate file. For instance
   processor 0 would write local extends "0 9 0 10 0 0" to file file0.vts
   processor 1 would write local extends "10 19 0 10 0 0" to file file1.vts
   processor 2 would write local extends "20 29 0 10 0 0" to file file2.vts
   ...
I thought that when I wrote the pvts file to combine the data this would be ghostlevel="0" since there is no overlap of the data, but I get errors on a lack of extents. For instance in the above it would be for lack of extents "9 10 0 10 0 0" and "19 20 0 10 0 0", i.e., where the parallel boundary occurs.

I am wondering if there is a better way to do what I am trying to accomplish / whether it is possible to do such formatting with vtk. I would prefer not having processors write duplicated data if possible.

Thanks!
Jeremy
_______________________________________________
Powered by www.kitware.com<http://www.kitware.com>

Visit other Kitware open-source projects at http://www.kitware.com/opensource/opensource.html

Please keep messages on-topic and check the VTK FAQ at: http://www.vtk.org/Wiki/VTK_FAQ

Follow this link to subscribe/unsubscribe:
http://www.vtk.org/mailman/listinfo/vtkusers

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.vtk.org/pipermail/vtkusers/attachments/20130628/6d8288a3/attachment.htm>


More information about the vtkusers mailing list