[Paraview] Parallel Mode Execution Algorithm

[Wylie, Brian]

Also,

There is a filter to move polydata around called 'Balance', the
limitation is the data has to be poly data.

*** Begin evangelizing ***

For general datasets of any type there is a filter called 'D3'.

D3 is a great filter for doing data distribution and computing ghost
cells in parallel.

For instance you can load in a single datafile and the 'distribute' it
to all the nodes just by invoking the D3 filter on it (you can run the
'Process Id Scalars' filter to see the distribution).

D3 should work on any data type and like a said will compute ghost cells
for you. The ghost cell computation makes a big difference (external
faces on 1 dataset went from 80M tris to 2M tris). Also obviously you
will now get the 'correct' visualization results instead of seeing file
or processor boundaries.

In addition the nice spatial distribution has the side benefit that your
parallel rendering (to desktop) and especially to tile-display will be
accelerated.

A description of this filter should have been included in the ParaView
users guide... but was not.... grumble...grumble... :) 

*** End evangelizing *** 

	Brian Wylie
	Sandia National Laboratories
	MS 0822 - Org 9227 - Building 880/A1-J
      (505) 844-2238   FAX (505) 845-0833
 

> -----Original Message-----
> From: paraview-bounces at paraview.org 
> [mailto:paraview-bounces at paraview.org] On Behalf Of Brad King
> Sent: Thursday, December 16, 2004 6:20 AM
> To: Jeremy Stout
> Cc: paraview at paraview.org
> Subject: Re: [Paraview] Parallel Mode Execution Algorithm
> 
> Jeremy Stout wrote:
> > Hello, I'm looking for any information related to the algorithms 
> > employed by Paraview when it is used in parallel mode. I 
> know how to 
> > setup Paraview to run in parallel mode.  I am interested in 
> learning 
> > how Paraview distributes the VTK pipeline (I'm assuming it 
> does) and 
> > how it handles the distribution of data files.  I would like to use 
> > this information to be able to see if I could do anything 
> to optimize 
> > my current Paraview installation.
> 
> Basically the VTK pipeline is duplicated on all the nodes.  
> Since VTK supports "streaming" data in pieces, each node 
> executes 1 of N pieces for N nodes.  Then one of several 
> approaches for rendering the results on a single screen is 
> taken.  For small data the geometry is collected onto the 
> client process and rendered locally.  For large data the 
> geometry on each node is rendered on the node itself and sent 
> to the client in the form of an image and a z-buffer.  The 
> client composits it for rendering.
> 
> For more information you can get the "ParaView Guide" book:
> 
> http://www.kitware.com/products/paraviewguide.html
> 
> -Brad
> _______________________________________________
> ParaView mailing list
> ParaView at paraview.org
> http://www.paraview.org/mailman/listinfo/paraview
> 
>