[Paraview] Paraview in Parallel-Server mode

[Randall Hand]

I've had a few issues with it (We're using Infiniband, so network bandwidth
should be negligible), some of which are just I incorrectly percieved what
was going on and some of which I think is a bug.

I think what I originally percieved as phenomenal slowless is actually a
bug.  I use 4x Subsampling for the interactive rendering, but when I let go
it never returns to full resolution, it seems to drop to something like 2x
subsampling.  I originally interpreted that as it's slowly rendering in the
background, but if I use no subsampling for the interactive displays then it
works just fine.  (I just ran this test a few minutes ago).

The other slowness I noticed was in the way lookmarks are handled.  I build
the lookmark with the MPI version so the decimated models were already in
memory when it came time for the "big interactive part".  When I repeated
the lookmark in IceT, I had to wait a good 30s or so for it to build the
decimated models and such.  Again, I just incorrectly percieved what was
going on.

In the no subsampling version, yes it beat the pants off MPI hands-down,
although you can't see what's happening on the remote displays quite as
well.  Is there some reason why I would be seeing noise on the other heads?
During interaction I see the top 1/3 of the screen rendered on all the
nodes, but during stills I just see noise.  It's minor, but I know someone
is gonna ask if they see it.




On 8/9/06, Wylie, Brian <bnwylie at sandia.gov> wrote:
>
>  Randall,
>
> Can you elaborate on the poor performance? We can get 30 frames/sec over
> 100T with IceT, I'm having a hard time believing that MPIRenderModule
> actually performs better. The only circumstance I've seen where
> MPIRenderModule outperforms IceT is when the client is like a laptop with
> really slow pixel scaling. Can you give me details of your setup... cluster
> specs... network specs... client specs...
>
> Also I'm assuming that your compositing threshold is 0MB in both tests,
> and subsampling is set to the same value, squirt compression is the same,
> etc...
>
>
>   Brian Wylie - Org 1424
>   Sandia National Laboratories
>   MS 0822 - Building 880/A1-J
>   (505)844-2238 FAX(505)845-0833
>        ____                  _    __
>       / __ \____  _________ | |  / (_)__ _      __
>      / /_/ / __ `/ ___/ __ `/ | / / / _ \ | /| / /
>     / ____/ /_/ / /  / /_/ /| |/ / /  __/ |/ |/ /
>    /_/    \__,_/_/   \__,_/
> |___/_/\___/|__/|__/
>                                     Unleash the Beast
>
>
>  ------------------------------
> *From:* paraview-bounces+bnwylie=sandia.gov at paraview.org [mailto:
> paraview-bounces+bnwylie=sandia.gov at paraview.org] *On Behalf Of *Berk
> Geveci
> *Sent:* Wednesday, August 09, 2006 9:55 AM
> *To:* Randall Hand
> *Cc:* Paraview List
> *Subject:* Re: [Paraview] Paraview in Parallel-Server mode
>
> I would stick with the default (IceTDesktopRenderModule) and tweak 3D view
> settings parameters until the performances is good. I think I have a power
> point presentation talking about the parameters and what they do. Ken
> presented it at supercomputing and it explains different options very
> clearly. I can share it.
>
> -Berk
>
> On 8/9/06, Randall Hand <randall.hand at gmail.com> wrote:
> >
> > Is there a preferred Render Module to use when running Paraview in
> > Single-Display Multiple-Server mode.  I'm running a pvserver across our
> > 12-node cluster and getting pretty poor performance with the Default.
> > Specifying the MPIRenderModule helps, but the IceTDesktopRenderModule seems
> > pretty bad.  Is there a Preferred module to use, and reasons why you'ld use
> > one over the other?
> >
> > --
> > ----------------------------------------
> > Randall Hand
> > Visualization Scientist
> > ERDC MSRC-ITL
> >
> > _______________________________________________
> > ParaView mailing list
> > ParaView at paraview.org
> > http://www.paraview.org/mailman/listinfo/paraview
> >
> >
> >
>


-- 
----------------------------------------
Randall Hand
Visualization Scientist
ERDC MSRC-ITL
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://public.kitware.com/pipermail/paraview/attachments/20060809/4d19c191/attachment.html