[Paraview] Strange behaviour with large datasets

Wed Sep 3 03:58:46 EDT 2008

Nathanael Inkson wrote:
>
> Hi Ken,
>
> “The mesh quality filter will be run multiple times on multiblock data 
> (once per block), but I assume that is not the case for you as volume 
> rendering is not available on data with more than one block.”
>
> I think this sentence sums up my problem: the unstructured grid does 
> indeed consist of multiblock data.
>
> Ok thanks so much for your help. I will look for a different approach 
> to what I am trying to do.
>
Keep the list informed as to what you find as more users are starting to 
use very large meshes.
Stephen
>
> Thanks for your time.
>
> Nat.
>
> *Best Regards*
>
> * *
>
> *Dr. Nathanael Inkson,***
>
> * *
>
> *Computational Fluid Dynamics Developer,***
>
> *Digital Flow Solutions,***
>
> ****
>
> ******
>
> **mob: + 44 7872010167******
>
> **web: wirthresearch.com <http://wirthresearch.com>**
>
> ------------------------------------------------------------------------
>
> *From:* Moreland, Kenneth [mailto:kmorel at sandia.gov]
> *Sent:* 02 September 2008 18:41
> *To:* Nathanael Inkson
> *Cc:* Simon Hubbard; paraview at paraview.org
> *Subject:* RE: Strange behaviour with large datasets
>
> I am a bit confused. The initial question sounded like a volume 
> rendering problem; you could not see the result of the mesh quality 
> filter very well. Is the problem you are reporting that the output of 
> the mesh quality filter is incorrect? Do the ranges look wrong in the 
> information tab? Does the threshold filter not extract cells you 
> expect to be of poor quality?
>
> Does your modified mesh quality filter behave differently than the one 
> that comes with ParaView? The pipeline should not be updating your 
> filter unless it needs to refresh the calculation. The mesh quality 
> filter will be run multiple times on multiblock data (once per block), 
> but I assume that is not the case for you as volume rendering is not 
> available on data with more than one block. The cell count might be 
> off because you are running in parallel. Each process will have a 
> partition of the whole mesh, so the cell count will be smaller unless 
> you are getting the global multiprocess controller and adding up all 
> the cells.
>
> If the pipeline is inappropriately calling your filter multiple times, 
> it might be that your filter is calling its own Modified method 
> (usually indirectly through, for example, a call to one of the Set* 
> macros). You can check for that by making sure that the MTime ivar 
> does not change while invoking RequestData or any other 
> ProcessRequests method.
>
> -Ken
>
> ------------------------------------------------------------------------
>
> *From:* Nathanael Inkson 
> [mailto:Nathanael.Inkson at digitalflowsolutions.com]
> *Sent:* Tuesday, September 02, 2008 11:15 AM
> *To:* Moreland, Kenneth
> *Cc:* simon.hubbard at digitalflowsolutions.com
> *Subject:* RE: Strange behaviour with large datasets
>
> Dear Ken,
>
> Thanks for your answer but I am not sure that this is the reason as I 
> have other fields which have large values. If I count the number of 
> cells in one pass through vtkMeshQuality and output to a file, I find 
> that the answer is far less than the actual number of cells in the 
> mesh. I have a gut feeling that because the mesh is so big that the 
> MeshQuality filter is entered several times as it updates the progress 
> bar, wouldn’t that mean that the newly created quality arrays are 
> overwritten each time the filter is entered? I’m not too sure about 
> how the filters actually work.
>
> How is this prevented?
>
> Thanks
>
> Nat.
>
> *Best Regards*
>
> * *
>
> *Dr. Nathanael Inkson,***
>
> * *
>
> *Computational Fluid Dynamics Developer,***
>
> *Digital Flow Solutions,***
>
> ****
>
> ******
>
> **mob: + 44 7872010167******
>
> **web: wirthresearch.com <http://wirthresearch.com>**
>
> ------------------------------------------------------------------------
>
> *From:* Moreland, Kenneth [mailto:kmorel at sandia.gov]
> *Sent:* 02 September 2008 14:58
> *To:* Nathanael Inkson; paraview at paraview.org
> *Subject:* RE: Strange behaviour with large datasets
>
> If your mesh is typical, then probably most of the cells are of good 
> quality, and there are a handful of cells of bad quality. Thus, when 
> volume rendering most of the cells will be invisible and that handful 
> of cells will have some opacity.
>
> The probable reason that you can see them in a small mesh and not a 
> large one is that the cells are spatially much smaller (with respect 
> to the entire mesh) for the large mesh. In correct volume rendering, a 
> material of a constant density will have smaller opacity the thinner 
> it gets. Your cells are probably so small that there is not enough 
> accumulation to actually see them.
>
> You can adjust for the small scale of the cells by adjusting the 
> “Scale” parameter in the Color Scale Editor dialog box. That parameter 
> is actually a unit length specification for the opacity parameters 
> given. Setting it to about cell length divided by 10 should give you 
> decent results. However, an easier and probably more effective 
> approach would be to bag the volume rendering and use the threshold 
> filter to extract the cells of interest (in this case, the bad ones).
>
> -Ken
>
> ------------------------------------------------------------------------
>
> *From:* paraview-bounces at paraview.org 
> [mailto:paraview-bounces at paraview.org] *On Behalf Of *Nathanael Inkson
> *Sent:* Tuesday, September 02, 2008 7:31 AM
> *To:* paraview at paraview.org
> *Subject:* [Paraview] Strange behaviour with large datasets
>
> Dear Paraview people,
>
> I am working with Paraview with very large datasets (around 40Gb). I 
> have written a filter derived from MeshQuality which doesn’t seem to 
> display the returned data arrays correctly. I checked with the same 
> mesh and the MeshQuality filter. The same thing happens: I put the 
> output on volume rendering to look at the quality field and the whole 
> dataset is invisible. I have tried rescaling the range but to know 
> avail. Is this a known problem?
>
> Any advice? With smaller meshes the problem disappears and the quality 
> field is visible with volume rendering.
>
> *Best Regards*
>
> * *
>
> *Dr. Nathanael Inkson,***
>
> * *
>
> *Computational Fluid Dynamics Developer,***
>
> *Digital Flow Solutions,***
>
> ****
>
> ******
>
> **mob: + 44 7872010167******
>
> **web: wirthresearch.com**
>
> ------------------------------------------------------------------------
>
> _______________________________________________
> ParaView mailing list
> ParaView at paraview.org
> http://www.paraview.org/mailman/listinfo/paraview
>