[Paraview] Compositing

[Wylie, Brian]

To make sure the data is load balanced. Run 'ProcessIdScalars' filter...
That way you can see if the 6 processors have about the same amount of
data.

:)


  Brian Wylie - Org 9227
  Sandia National Laboratories
  MS 0822 - Building 880/A1-J
  (505)844-2238 FAX(505)845-0833                       
       ____                  _    __
      / __ \____  _________ | |  / (_)__ _      __ 
     / /_/ / __ `/ ___/ __ `/ | / / / _ \ | /| / /
    / ____/ /_/ / /  / /_/ /| |/ / /  __/ |/ |/ /
   /_/    \__,_/_/   \__,_/ |___/_/\___/|__/|__/

                                    Unleash the Beast
 

> -----Original Message-----
> From: paraview-bounces+bnwylie=sandia.gov at paraview.org 
> [mailto:paraview-bounces+bnwylie=sandia.gov at paraview.org] On 
> Behalf Of Moreland, Kenneth
> Sent: Tuesday, March 21, 2006 1:47 PM
> To: Terry Jordan; paraview at paraview.org
> Subject: RE: [Paraview] Compositing
> 
> With only 6 nodes, ParaView uses a tree compositing algorithm 
> (at 8 nodes it switches to binary swap for better 
> distribution).  So the first node does indeed to some more 
> compositing, but not much.  If you are seeing a large 
> difference in load balancing, I seriously doubt the 
> compositing is a major contribution.  Are you sure your data 
> has an even distribution to begin with?  Many readers do not 
> necessarily give a good distribution of data.  You could try 
> running your visible surfaces through D3 to balance them 
> better, but if your animation changes the geometry every 
> frame you will lose more in redistribution than you gain in 
> load balancing.
> 
> -Ken
> 
> > -----Original Message-----
> > From: Terry Jordan [mailto:tejj at hotmail.com]
> > Sent: Tuesday, March 21, 2006 1:08 PM
> > To: Moreland, Kenneth; paraview at paraview.org
> > Subject: RE: [Paraview] Compositing
> > 
> > The problem we have is that the data is starting to hit the 
> theshold 
> > of what a single workstation can handle, so we are trying 
> to setup a 
> > small vis cluster to allow us to look at larger data.  Some of the 
> > data can be aninated and I am trying to allow the animation 
> to run as 
> > smoothly as possible while using parallel paraview.  It 
> just seemed to 
> > me that the first node, the one you launch the mpi pvserver 
> from seems 
> > to be doing more work than all the other nodes and I 
> thought it would 
> > balance out the workload a little if one machine did the 
> compositing 
> > alone and not have a chunk of its own it needed to render.  
> So in our 
> > case we have 6 nodes I want to run mpi pvserver on and 1 
> node to run 
> > the client.  of those 6 nodes I figured we could have 5 pcs 
> rendering 
> > chunks and one compositing the chunks and squirting it to 
> the client.
> > 
> > ----Original Message Follows----
> > From: "Moreland, Kenneth" <kmorel at sandia.gov>
> > To: "Terry Jordan" <tejj at hotmail.com>, paraview at paraview.org
> > Subject: RE: [Paraview] Compositing
> > Date: Tue, 21 Mar 2006 09:43:03 -0700
> > 
> > If there is not much difference between your server nodes 
> and client, 
> > you can very easily move all the geometry to the client and render 
> > locally.  In "3D View Properties" under "LOD Parameters" 
> there is an 
> > entry titled "Composite."  That entry has a slider bar that 
> represents 
> > the threshold at which geometry is moved to the client.  If 
> necessary, 
> > you can move that slider to the right so that it collects bigger 
> > geometries.  You can also unclick the box to turn off compositing 
> > altogether and always collect polygons.  This will backfire on you, 
> > however, if you are rendering a large quantity of data.
> > 
> > If you really want the geometry to be collected on one of 
> the server 
> > nodes, you will have to run in client/render server/data 
> server mode.
> > Make the data server span all 6 machines and the render 
> server reside 
> > on one of them.  It should be OK to have a render server and data 
> > server share a node so long as memory is not a problem.  This 
> > configuration will force all the geometry to be pushed to 
> the single 
> > render server node where it will be rendered and delivered to the 
> > client.
> > 
> > Performance-wise is this a good idea?  That depends on your system 
> > configuration.  Unless your server rendering hardware is way faster 
> > than your client and your cluster interconnect really sucks, then 
> > probably not.  I recommend experimenting with the "Composite" and 
> > "Subsample Rate" options first to try to find a good compromise 
> > amongst rendering speed, image quality, and data management.
> > 
> > -Ken
> > 
> >  > -----Original Message-----
> >  > From: paraview-bounces+kmorel=sandia.gov at paraview.org
> >  > [mailto:paraview-bounces+kmorel=sandia.gov at paraview.org]
> > On  > Behalf Of Terry Jordan  > Sent: Tuesday, March 21, 2006
> > 9:03 AM  > To: paraview at paraview.org  > Subject: [Paraview] 
> > Compositing  >  > I am running paraview (pvserver) using MPI on 6 
> > machines (a  > small rendering or vis cluster) and piping 
> it back to a  
> > > single display using pvclient.  I am wondering if I can  
> > specifiy 
> > one pc to only do the compositing.  So instead of  > one pc 
> rendering 
> > a portion of the data and compositing, one  > could just 
> composite.  
> > Is this feasible and performance-wise  > would it be a good idea?
> >  >
> >  > Thanks in advance.
> >  >
> >  >
> >  > _______________________________________________
> >  > ParaView mailing list
> >  > ParaView at paraview.org
> >  > http://www.paraview.org/mailman/listinfo/paraview
> >  >
> >  >
> > 
> > _______________________________________________
> > ParaView mailing list
> > ParaView at paraview.org
> > http://www.paraview.org/mailman/listinfo/paraview
> > 
> > 
> > 
> > 
> 
> _______________________________________________
> ParaView mailing list
> ParaView at paraview.org
> http://www.paraview.org/mailman/listinfo/paraview
> 
>