[Paraview] Very slow performance in Parallel mode

Paul McIntosh paul.mcintosh at internetscooter.com
Tue Sep 11 07:10:16 EDT 2012


This is what I use for a 12 core node – points half the data at one display and half at the other (assuming the x server is configured that way)

 

mpirun -v -hostfile $PBS_NODEFILE -bynode -np 6 /usr/local/$APPLICATION/$VERSION/bin/pvserver --server-port=11112 -display :0.1 : -np 6 /usr/local/$APPLICATION/$VERSION/bin/pvserver -display :0.0

 

If you want test data of increasing size – I guess the best approach it to generate it in ParaView itself e.g. waveform source and contour it, then just make it bigger and bigger J

 

 

 

From: Arya Mazaheri [mailto:aryanet at gmail.com] 
Sent: Tuesday, 11 September 2012 8:30 PM
To: Paul McIntosh
Cc: paraview at paraview.org
Subject: Re: [Paraview] Very slow performance in Parallel mode

 

Thanks for your reply. Is there any place which I can download sample datasets with different sizes? 

BTW, I have two GPUs installed on my node. I'm not sure whether paraview uses both of them. Is there anyway to force paraview use both of them?

 

 

On Tuesday ۲۱ June ۱۳۹۱ at ۱۴:۳۰, Paul McIntosh wrote:

I am not sure – I am on a cluster with a scheduler, so cores are selected before the visualisation work starts. Generally the size of the data is known in advance so the processors needed to process that data have already been worked out. We are normally dealing with raw image data, if I have 148M dataset it works best with one core, and I max out a 12 core node with a 10G dataset (with two NVIDIA M2070Qs).  You can have different configurations on  the server side and select them as needed.

 

From: Arya Mazaheri [mailto:aryanet at gmail.com] 
Sent: Tuesday, 11 September 2012 7:40 PM
To: Paul McIntosh
Cc: paraview at paraview.org
Subject: Re: [Paraview] Very slow performance in Parallel mode

 

So, is there any way to dynamically assign number of CPU cores to the visualization process? 

 

On Tuesday ۲۱ June ۱۳۹۱ at ۱۴:۰۶, Paul McIntosh wrote:

Arya – from my experience small data will behave worse because there is some overhead in splitting it up. If it all fits in one process then that is the best place for it. 

 

Paul

 

From: paraview-bounces at paraview.org [mailto:paraview-bounces at paraview.org] On Behalf Of Arya Mazaheri
Sent: Tuesday, 11 September 2012 7:06 PM
To: paraview at paraview.org
Subject: [Paraview] Very slow performance in Parallel mode

 

Hi there, 

I have made some experiments to see how running paraview in parallel mode can help me visualize my data more faster. 

First I ran paraview on my Macbook pro with a sphere source which has about 90MB data. It's running smoothly without considerable annoyance. Then I tried to run the same dataset on ONE cluster node with 48 CPU cores. I started the paraview with the following command:

# mpirun -np 48  /share/apps/Paraview3/bin/pvserver -display :0

 

Then I connected to the server with my Macbook to render the same dataset. But I noticed a great deal on performance degradation. It was so hard to rotate and manipulate the object in 3d view. I am surprised by this issue! Do you know the reason?

 

Thanks,

 

 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.paraview.org/pipermail/paraview/attachments/20120911/1ee78428/attachment-0001.htm>


More information about the ParaView mailing list