[Paraview] Performance of the CGNS Reader

Richard GRENON richard.grenon at onera.fr
Fri Aug 30 04:24:37 EDT 2013


Thank you for this answer, Mickael.

My 1.36 Gb CGNS dataset is built from structured meshes and Paraview 
should not 'eat' too much memory. I have checked that enabling 
multi-core in Paraview does not change anything: PV always needs about 
15 mn to load my dataset, same loading time as without multi-core.

I think that the loading time ratio 15 mn / 1 mn for PV against Tecplot 
remains too high, even if PV parses the file two times. If Tecplot takes 
advantage from multi-core (I don't know), the loading time ratio between 
PV and Tecplot should not excess 8 when using 4 CPUs. A ratio of 15 
leading to 15 mn of loading time or more for larger datasets is 
unacceptable for interactivity. So PV is unusable for large CGNS 
datasets, unless using batch mode. I think that an effort in redesigning 
the CGNS reader would be welcome.

Best regards.

Richard

Le 29/08/2013 20:55, Mickael Philit a écrit :
>
> Hello,
>
> First, the CGNS reader coming through the VisItBridge is not working 
> in parallel, it's a plain serial reader.
> Second, there are limitations to the current cgns reader way of doing 
> thing, since :
>   - At the beginning, it parses the whole file (this takes a lot of 
> time) to get variable names, blocks and so on, before actually reading 
> the data.  [ I think that tecplot is cleaner because it seems to read 
> the whole CGNS file in one pass ]
>  - meshes are read in a temporary array and converted to a VTK vector 
> of coordinates (thus memory manipulation)
>  - for unstructured meshes, convertion from 'integer' to 'long' of 
> cells connectivity eats memory.
> The CGNS reader can improve but at the cost of redesining some parts 
> to fit better in paraview and go for parallel.
>
> Mickael
>
>
> On 29/08/2013 16:50, Angelini, Richard C (Rick) CIV USARMY ARL (US) 
> wrote:
>> As a followup to this that may be related - does the CGNS reader 
>> through the VisItBridge work in parallel?    I've loaded up a couple 
>> of different CGNS datasets and then applied the ProcessIDScalars 
>> filter and it doesn't appear to be distributing the data - even 
>> multi-block CGNS files.
>>
>>
>> ________________________________
>> Rick Angelini
>>
>> USArmy Research Laboratory
>> CISD/HPC Architectures Team
>> Building 120 Cube 315
>> Phone:  410-278-6266
>>
>> ________________________________________
>> From: paraview-bounces at paraview.org [paraview-bounces at paraview.org] 
>> on behalf of Richard GRENON [richard.grenon at onera.fr]
>> Sent: Thursday, August 29, 2013 10:38 AM
>> To: paraview at paraview.org
>> Subject: [Paraview]  Performance of the CGNS Reader
>>
>> Hello.
>>
>> I am testing the CGNS reader of Paraview 4.0.1 64 bits running on a
>> Linux Workstation having 4 CPUs and 5.8 Gbytes of memory. Paraview was
>> installed from the binaries available on the download page.
>>
>> I am trying to load a 1.36 Gbytes CGNS file that is available through
>> the network.
>>
>> While loading this file, the Paraview Windows is frozen and cannot be
>> refreshed, and I must check with the "ps" command on a terminal window
>> or with a system monitor if PV  is still running or if it is really
>> frozen. A progress bar for all readers would be welcome in a next 
>> release.
>>
>> Finally, the file can be loaded, but it always takes about 15 mn (+ or -
>> 1 mn depending of the load of the network), while Tecplot always loads
>> the same file within less that 1 mn !
>>
>> How do you explain this poor performance of the CGNS reader ? Can it be
>> improved, or am I missing something ? Is there some Paraview option that
>> could reduce loading time of large files ?
>>
>> Best regards
>>
>> -- 
>>    Richard GRENON
>>    ONERA
>>    Departement d'Aerodynamique Appliquee - DAAP/ACI
>>    8 rue des Vertugadins
>>    92190 MEUDON - FRANCE
>>    phone : +33 1 46 73 42 17
>>    fax   : +33 1 46 73 41 46
>>    mailto:Richard.Grenon at onera.fr
>>    http://www.onera.fr
>>
>> _______________________________________________
>> 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
>> _______________________________________________
>> 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
>
>


-- 
  Richard GRENON
  ONERA
  Departement d'Aerodynamique Appliquee - DAAP/ACI
  8 rue des Vertugadins
  92190 MEUDON - FRANCE
  phone : +33 1 46 73 42 17
  fax   : +33 1 46 73 41 46
  mailto:Richard.Grenon at onera.fr
  http://www.onera.fr



More information about the ParaView mailing list