[Paraview] Re:Re: HW acceleration in client/server mode (Alain SENOT)

Wed Jul 5 20:55:49 EDT 2006

we just fix this problem by "assign a valid display to all server nodes".
Because mpirun will first execute .bashrc on each node, by adding following lines to .bashrc file in home directory, you are able to set "DISPLAY"variable of each server node to be the local IP:0.0 respectively. and then HW acceleration in client/server mode is ok!

host=`uname -n`
ip=`grep $host /etc/hosts | awk '{ print $1; }'`
export DISPLAY=$ip:0.0

shan

>
>Message: 3
>Date: Wed, 5 Jul 2006 14:25:14 +0200
>From: Alain SENOT <alain.senot at edf.fr>
>Subject: Re: [Paraview] HW acceleration in client/server mode
>To: paraview at paraview.org
>Message-ID:
>	<OF13172559.9F3ADC0B-ONC12571A2.0040959E-C12571A2.0044395F at notes.edfgdf.fr>
>	
>Content-Type: text/plain; charset="us-ascii"
>
>I'm using Paraview 2.4.3 in client-server mode on a little linux cluster 
>with graphic cards.
>I'm using mpich 1.2.7 and compiled with the settings of the following 
>variables :
>VTK_USE_MPI  on
>MPI LIBRARY 
>MPI INCLUDE PATH
>MPI EXTRA LIBRARY
>MPI EXECUTABLE mpirun
>CMAKE C COMPILER mpicc
>CMAKE CXX COMPILER mpicxx
>
>Then I copied the files of Paraview and MPI on other nodes of the cluster. 
>
>
>When I launch the application in client-server mode, a warning message 
>appears, telling me that I won't be able to use compositing.
>"To enable compositing, compile and run the server with offscreen Mesa 
>support or assign a valid display to all server nodes."
>
>For obvious reason, I would like to choose the second solution.
>
>How can I assign a valid display to all server nodes so that the server 
>can render ?
>The DISPLAY environment value is :0.0 on all nodes for the user.
>The ~/.rhosts file lists all nodes. mode 700. The UID of the user is the 
>same on all nodes.
>MPIRUN has been added to the PATH variable, MPI libs to LD_LIBRARY_PATH.
>The machine list file contains the machines of the server.
>
>Thanks again for your help,
>
>Alain Senot
>EDF
>
>
>>Message: 3
>>Date: Mon, 3 Jul 2006 15:53:34 -0400
>>From: "Berk Geveci" <berk.geveci at gmail.com>
>>Subject: Re: [Paraview] HW acceleration in client/server mode
>>To: "Alain SENOT" <alain.senot at edf.fr>
>>Cc: paraview at paraview.org
>>Message-ID:
>> <a2ab21560607031253o23ebd97ft6c643b938ad4773c at mail.gmail.com>
>>Content-Type: text/plain; charset="iso-8859-1"
>>
>>Take a look at this post:
>>http://public.kitware.com/pipermail/paraview/2006-July/003366.html
>>
>>The users guide has a longer description of ParaView's architecture.
>>
>>
>>-Berk
>>
>>On 7/3/06, Alain SENOT <alain.senot at edf.fr> wrote:
>>>
>>>
>>> Hi everybody,
>>>
>>> I would like to know (if it is possible/how) to use the hardware
>>> acceleration on the server in client/server mode.
>>>
>>> Thanks
>>>
>>> Alain Senot
>>> EDF
>-------------- next part --------------
>An HTML attachment was scrubbed...
>URL: http://public.kitware.com/pipermail/paraview/attachments/20060705/0e70dbfa/attachment-0001.htm
>
>------------------------------
>
>Message: 4
>Date: Wed, 5 Jul 2006 09:27:36 -0500
>From: "Karla Vega" <vega at tacc.utexas.edu>
>Subject: RE: [Paraview] Streamtracer filter
>To: "Berk Geveci" <berk.geveci at gmail.com>
>Cc: paraview at paraview.org
>Message-ID:
>	<A0E12450B9ACD14C9ECD94E9886E305B0105566D at fireant.tacc.utexas.edu>
>Content-Type: text/plain; charset="iso-8859-1"
>
>Hi Berk,
>
>Thank you so much for replying. I've uploaded pictures, movies and data to: (see description below)
>
>www.tacc.utexas.edu/~vega/Blood_Flow
>
>The movies and pictures are under "Vis". I also have uploaded 25 timesteps to "Data".
>We've seeding streamlines were the velocity field was the highest. We used point-cloud and we noticed that some of the streamlines end at the walls and some just stop at some points as you can see from the pictures. We worked for a while on seeding the streamlines correctly and using different integrators and step sizes, but nothing seemed to work (although it did look better when we seeded streamlines at the highest velocity points).
>The computer we are using is a Linux (Red hat) Dell Precision 670 Mini-Tower 3.60GHz, XEON Processor w/ 2MBL2 Cache with 8 GB in memory. The paraview version we are using is 2.4.2 64 bit.
>
>I also wanted to ask you about the blending that Paraview uses. As you can see from the pictures, objects in the back are not being blended correctly. Is there a way to fix this?
>Maybe changing the rendering to be back-to-front instead of front-to-back?
>
>
>Thank you again for all your help!
>
>Karla Vega
>vega at tacc.utexas.edu
>
>
>Description:
>
>"Patient-specific abdominal aorta"
>
>"Fluid-structure interaction is calculated for a patient-specific abdominal aorta of a healthy over-55 volunteer obtained from a 64-slice CT angiography. A periodic flow waveform is applied at the inlet of the aorta, while resistance boundary conditions are applied at all outlets. The computational mesh consists of 52,420 quadratic NURBS elements, and the output of the simulation includes pressure and velocity fields. The entire time series of data is approximately 500MB."
>
>Karla Vega
>Email: vega at tacc.utexas.edu
>Work: (512) 232-4772
>Mobile: (512) 577-8723
>
>
>
>-----Original Message-----
>From: Berk Geveci [mailto:berk.geveci at gmail.com]
>Sent: Mon 7/3/2006 8:28 AM
>To: Karla Vega
>Cc: paraview at paraview.org
>Subject: Re: [Paraview] Streamtracer filter
> 
>Hi Karla,
>
>Can you be more specific? Maybe post some screenshots and data? That filter
>has been quite thoroughly tested on single processor. I also fixed a few
>bugs that were in the parallel implementation. Depending on your ParaView
>version, you might not have the fixes for the parallel version. I would be
>surprised if the integration algorithm itself has a bug. Of course, there
>are always numerical accuracy issues and you can compensate by using a
>different integrator or adjusting the step-size (the default is Runge-Kutta
>2, I believe). Known issues are:
>
>* Premature ending of streamlines in C-grids where the grid folds onto
>itself (this may be what you are seeing)
>* Premature ending of streamlines in singlular cells
>
>I also need to check a bug report that reads "Streamlines wrong for uniform
>rectilinear data... You would expect a streamline to form a circle around
>the X axis, but what you get is a streamline that seems to go in completely
>the wrong direction. The problem only seems to happen in rectilinear
>data.  If you convert to an unstructured grid (with Clean to Grid), the
>streamline appears as expected.".  I am guessing it is a bug in the voxel
>cell type. I will check this bug and fix it if necessary in next few weeks.
>
>-Berk
>
>
>On 6/30/06, Karla Vega <vega at tacc.utexas.edu> wrote:
>>
>>  Hello everyone,
>>
>> I've been having difficulties with the streamtracer filter. It seems to be
>> integrating the velocity field incorrectly and some of my streamlines will
>> not continue where they are suppose to do so. I've read through the
>> vtkStreamTracer.h file to learn about the different inputs, but it still
>> doesn't show the streamlines completely.
>> Does anyone know of any known issues with this filter?
>>
>> Thank you again,
>>
>> Karla
>>
>> _______________________________________________
>> ParaView mailing list
>> ParaView at paraview.org
>> http://www.paraview.org/mailman/listinfo/paraview
>>
>>
>>
>
>-------------- next part --------------
>An HTML attachment was scrubbed...
>URL: http://public.kitware.com/pipermail/paraview/attachments/20060705/19871a59/attachment.html
>
>------------------------------
>
>Message: 5
>Date: Wed, 05 Jul 2006 11:45:11 -0400
>From: Kent Eschenberg <eschenbe at psc.edu>
>Subject: [Paraview] Turning On Debugging
>To: ParaView <paraview at paraview.org>
>Message-ID: <5AD2AC0CC249CA12341F176B at nvlap.psc.edu>
>Content-Type: text/plain; charset=us-ascii; format=flowed
>
>Lets say I compiled from source then wished to recompile with debugging. 
>This is what I tried:
>
>   -- run ccmake
>   -- set CMAKE_BUILD_TYPE to DEBUG
>   -- configure
>   -- generate and exit
>   -- run make
>
>A great many warnings result (all, it seems, from hdf code).
>
>The above steps work fine when compiling in an empty directory. Is there 
>some other step to make this work? Or must one always wipe out the 
>non-debug version first?
>
>Kent
>Pittsburgh Supercomputing Center
>
>
>
>------------------------------
>
>_______________________________________________
>ParaView mailing list
>ParaView at paraview.org
>http://www.paraview.org/mailman/listinfo/paraview
>
>
>End of ParaView Digest, Vol 27, Issue 6
>***************************************
>