[Paraview] Trying to run the servers on parallel

[Kevin Hackel]

Hi Sebastien,

There is a normal ethernet cable with 1 GB/s between server and computer.
The same data was created with different number of processes, for example 4 times 330 megabyte or 8 times 165 megabyte, so that the data as a whole is 1320 megabyte big and it is an unstructered data.
Well I had not the time to test it again, but I will do.

But what I haven’t understand is the fact that when I load this data and want to change the filter from outline to wireframe or surface than my Paraview crashed after one minute or so. When I load a single part of the data (with 165MB or 330MB) I could change the filter but it needs very long to do that.
When I first use isosurfaces on the data than I have no problem to view the data or to move it around. Where could be the problem?

Kevin

From: Sebastien Jourdain 
Sent: Tuesday, December 04, 2012 2:07 PM
To: Kevin Hackel 
Cc: paraview at paraview.org 
Subject: Re: [Paraview] Trying to run the servers on parallel

Hi Kevin, 

How good is your network between the remote PC that execute pvserver and your client computer that run paraview ?
How big is your data ?
What type of data is that ? Structured or unstructured ?
Did you try to run pvserver on that very same machine without using MPI ? Does that change anything ?
Did you try to go in ParaView/Edit/Settings and compare "local rendering" vs "remote rendering" performance ?

Seb



On Mon, Dec 3, 2012 at 4:18 PM, Kevin Hackel <kevin.hackel at online.de> wrote:

  Hello community,

  This is my first e-mail to the mailing list, so I hope I do it right :-)
  And sorry for my english, I try my best to make my mail understandable. If you can I would be pleased if you answer in german. Also I want you to know that I haven’t a great knowledge about informatics.

  Well I try to use Paraview on parallel, but I think it doesn’t work right how I do it.
  I use a remote PC on which I will run the Paraview client and a extern server with up to 48 CPUs and good GPUs.
  So I start the server using “mpirun –np ‘number’ ../pvserver” or “mpirun –np ‘number’ ../pvserver –display localhost:0” (my Paraview is compiled with openmpi on a linux system). Then I connect over the remote PC in Paraview to this server (client/server mode).
  It works, but not well. I’ve got test data (about 1GB) and it takes about two minutes to change for example the filter from outline to wireframe. Also when I want to move the image with drag and drop it takes about five seconds to rerender it. I think for only 1GB it takes too long for doing this.
  Also I’m not sure if my server uses its GPUs or only its CPUs, I don’t know how to find it out. 
  Would it be better to use my remote PC only as a display?
  Could you tell me how Paraview scales with  the number of data sets and number of processes. Should I use ‘x’ processes to load ‘x’ data sets? Or is it better to use more processes than data sets (for example: Would it be better to use 8 processes for loading 6 data sets than using 6 processes?)
  Is there another way how I could start my server that it will work better? Or what can I change that it will work better?

  I hope you can help me a little bit and give me tips how I should use Paraview on parallel. Thanks.


  Greetings,

  Kevin

  _______________________________________________
  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


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.paraview.org/pipermail/paraview/attachments/20121212/47b3fcaf/attachment.htm>