pvserver in parallel - weird behaviour

ParaView Support
1

Hello everyone,

I am trying to post-process a Fluent dataset in Paraview, which has been exported in the Ensight Case Gold format. The dataset is large, and typically it takes around 2 minutes to load. Upon monitoring the system monitor app, I noticed that Paraview is utilizing only a single core during the loading process.

I explored the internet and read about the parallel capabilities of ‘pvserver’ and decided to try it using multiple cores to potentially reduce the loading time. I initiated pvserver on my system via mpiexec in one terminal and then opened Paraview in another terminal.

./mpiexec -np 32 ./pvserver

I successfully connected Paraview to the server, and the dataset loading worked with multiple cores being utilized, as indicated by the system monitor. However, to my surprise, the parallel version seems to take more time to complete than the single-core version. For instance, the single-core loading took around 111 seconds, while with 32 cores, it took 154 seconds. I also experimented with 16 cores, and it took 130 seconds.

I am currently using the latest version, 5.12, which I downloaded from the Paraview website. My operating system is Fedora 38, and the hardware is an HP Z6 with dual Xeon Gold 5218 processors.

Any insights or suggestions on optimizing the parallel loading performance would be greatly appreciated.

Thank you!

2

Afaik the VTK Fluent reader is not a distributed aware data reader, hence only one core is doing the work.

You may want to use another format.

BTW for I/O work, unless you are working with a distributed file system and different physical disks, you will not see much speedup even for distributed data format.

3

Thanks for the response.

I am using the ensightreader, not the VTK Fluent reader. I read in some forums that reader is parallelized.

So in your opinion, Paraview can never get an improvement in loading time with multiple cores, if I am loading the dataset from a single disk? But why the parallel reading increases the loading time compared to single core?

4

It is.

Paraview can never get an improvement in loading time with multiple cores, if I am loading the dataset from a single disk?

I would not be that categorical but I would not expect much

But why the parallel reading increases the loading time compared to single core

They are overhead when you add a pvserver.

You are using 32 pvservers, how many core do you have ?

5

I have 32 cores.