[vtkusers] MPI Volume Rendering

[Kevin H. Hobbs]

On Fri, 2006-05-05 at 08:43 +0200, John Biddiscombe wrote:
> you are loading the data on one process, then using distributed data 
> filter to send it to others, then render.
> 

No, at least that is not my intent. The distributed data filter is
commented out right now. I'm not sure if I need it or not. The data are
prepositioned on the local hard disks of each node.

> Much better is to load the data in pieces on each processor, then render 
> and composite afterwards. (I may have misread things because each 
> processor could be receiving a different file name as argument)
> 

They all get the same file just copied to each node.

> on each processor, load a piece, render it, the composite on the client 
> or master node. If volume pieces are contiguous blocks then transparency 
> will be handled by sorting the composition order...
> 

Hmmm... it is a single piece file, I could split it up into streamed
pieces. I don't know if that helps though.

> anyway, I expect you know all this. sorry
> 
> JB
> 

At least for now the actual unstructured grid does not consume that much
RAM. It's something on the order of 100 MB. The real RAM usage _seems_
( at least according gkrellm while Paraview is running ) to happen
during the actual volume rendering. I should try putting some sort of
monitor on the cluster processes to see what's up...
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: This is a digitally signed message part
URL: <http://www.vtk.org/pipermail/vtkusers/attachments/20060505/b6cc37ad/attachment.pgp>