[Paraview] Paraview/pvserver running very slow

[Florian Lindner]

On Mon, 28 Feb 2011 23:06:45 +0900
=?iso-2022-jp?B?KBskQkVsNX4bKEIgKBskQkk4PWA7fhsoQikp?=, Takuya OSHIMA
<oshima at eng.niigata-u.ac.jp> wrote:

Hello,

> Is there any reason you prefer to load such a big case with
> reconstructed?  The OpenFOAM reader (and subsequent filters) runs in
> parallel only for decomposed cases. For reconstructed cases the reader
> loads everything in process 0.

Our current workflow is to compute on 128 cores and do postprocessing
on 4 to 12 cores. Therefore a decomposed case would consist of 128
processors directories. Would it still be advisable to postprocess with
just 8 cores? Since we're stil testing paraview (currently using another
postprocessing tool) our workflow is not well adapted for paraview)

However, I think this would affect only the load times, not slow
rendering when moving the view, would it?

> Also, are you sure you built ParaView with compiler optimizations
> (CMAKE_BUILD_TYPE = Release)?

CMAKE_BUILD_TYPE = Release was set on my own build which is very slow
too. I am not sure about the production build linked against SGI MPI,
but I suppose it is enabled.

Regards,

Florian


> Takuya
> 
> Takuya OSHIMA, Ph.D.
> Faculty of Engineering, Niigata University
> 8050 Ikarashi-Ninocho, Nishi-ku, Niigata, 950-2181, JAPAN
> 
> 
> From: Florian Lindner <mailinglists at xgm.de>
> Subject: [Paraview] Paraview/pvserver running very slow
> Date: Mon, 28 Feb 2011 13:15:15 +0100
> 
>>
>> Hello,
>>
>> I’m running paraview 3.8.1 under linux using 8 cpus loading a
>> reconstructed openfoam case with about 13GB per timestep (we have just
>> two time steps, start 0 and result 1). The machine on which pvserver
>> runs has 16 cpus and 70GB RAM, but pvserver uses just a small fraction
>> of the ram.
>>
>> Loading the case takes about 5 minutes (which would be fine, if
>> subsequent operations were running more smoothly).
>> Adding a slice takes another couple of minutes. Rotating (using the
>> mouse) takes about 10sec for one frame to be rendered (the slice is
>> constant colored). Computing 100 streamlines from a line source takes
>> about 10 minutes, moving the view is fluent. As soon as the slice is
>> displayed moving the view becomes extremely slow again.
>>
>> I have also tried to convert the mesh to VTK which does not really
>> change something. (Resulting VTK is of 37GB size).
>>
>> At this state interactive work is not possible. Since the dataset is
>> not extraordinary large (I think), I suspect a problem with our MPI
>> configuration.
>>
>> Offscreen rendering works (I think), we have 8 windows "Visualization
>> Toolkit – OpenGL" that opens on the client (though the windows remain
>> black).
>>
>> MPI implementation is SGI MPI. I have also tried to compile openmpi
>> 1.4.3 and pvserver from source, yet it’s still very slow. The client
>> used is the x64 build from the paraview homepage.
>>
>> MPI setup is for local use only.
>>
>> I am grateful for any help…
>>
>> Florian
>>
>> _______________________________________________
>> Powered by www.kitware.com
>>
>> Visit other Kitware open-source projects at http://www.kitware.com/opensource/opensource.html
>>
>> Please keep messages on-topic and check the ParaView Wiki at: http://paraview.org/Wiki/ParaView
>>
>> Follow this link to subscribe/unsubscribe:
>> http://www.paraview.org/mailman/listinfo/paraview