[Paraview] no improvement while using in parallel

[Kevin Hackel]

Hello all,

I’m trying to use paraview in parallel. For that I’ve got a self-compiled paraview version 3.14 and I use client/server mode.
In paraview I use these settings: use multicore, remote render threshold (100MByte), client outline threshold (0MByte) and interactive subsample rate (2pixels).

I’ve got unstructered test data. The data was created in four processes as *.vtu file (4times 352MByte) or as a whole *.pvtu file.

So I start with ssh the pvservers on the server (there are 48cores on the server) with 
mpirun –np ‘number’ ../pvserver –display localhost:0
and then I connect via my PC to this server (there’s a ethernet cable with 1GB/s between server and PC).

I tried different numbers of pvservers (2,4,8 and 16).
I loaded the data and used different filters like wireframe, contour, slice and clip and I moved the data around (with drag and drop).

But I could only see a little improvement in the time of representation (e.g. the contour filter took 5.3s on 4pvservers and 5s on 16pvservers).
I thought I could see a greater improvement.

So why do I see no improvement or when can I see an improvement? Or am I doing something wrong?

Also I’m surprised that it took lesser than 4 times as long when I use the whole data in compare with only a fourth of the data.
e.g. on 16 pvservers it took 6s to change the representation of a fourth of the data to wireframe and it took 20s for the whole data
Why is this happening?

Thank you for your help.


Greetings,
Kevin
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.paraview.org/pipermail/paraview/attachments/20130122/5b7b20c4/attachment.htm>