mpiexec utility

ParaView Support
1

Hi everyone!

I am having problems when using parallel post-processing with the mpiexec utility. The issue is that I cannot make paraview plot graphs when using multiple CPUs. At the beginning it wasn’t a big problem because the simulations were quite light but now I am switching to heavier simulations and this is delaying significantly my post-processing. Here I leave an example of how I call paraview with mpiexec with a single and multiple CPUS

Single CPU
time mpiexec -np 1 pvbatch --force-offscreen-rendering myscript.py

script_singleCPU
script_singleCPU2551×1280 372 KB

Multiple CPUs
time mpiexec -np 6 pvbatch --force-offscreen-rendering myscript.py

script_multiCPU
script_multiCPU2559×1271 253 KB

I have noticed that when using multiple CPUs the graphs are not plotted (in this case the Plot Data Over Time filter) and that the axis title labels (X(m) and Z(m)) are thicker like if there were several characters overlapped.

I would like to know if there is an specific command for the mpiexec utility to control which CPU takes care of each process because it seems that the CPU are failing to communicate properly.

Thank you in advance!

2

please share your script

3

Hi Mathieu,

Thank you so much for you quickly response. I think that the problem is not in the script since it works perfectly fine when using 1 CPU. Anyways you have here my script :

post_process_video_streaklines_CL.py (40.8 KB)

4

Hi again,

I found the problem, it is linked with the name of the variables. Below you have an example of the same script opened in serial (left) and the script opened in parallel using pvserver (right). I found that when plotting data over time, the variable name is modified.

image
image1114×327 54.8 KB

Since I was addressing to “cl (stats)”, nothing was plotted in parallel because the variable is named now “cl (stats rank=0)”.

5

Which filter generate this output ?

6

The Plot Data Over Time filter. My data is named x,z,cd,cl etc… but this filter adds the (stats…) extension.

7

This filter was reworked very recently by @Yohann_Bearzi , so maybe this is fixed in master already.