volume rendering performance is bad on huge data
Volume rendering performance is really bad, when run on huge data, in parallel. This needs to be looked at and if possible, adressed.
Here is how to replicate, using a source.
- ParaView 5.6.0, Linux, remote server 4 nodes (64 ranks). Total memory is about 256 GBytes total.
- Sources/ Unstructured Cell Types. Tet. I was able to set the size as 200x200x200. Apply.
- Color by Polynomial. Volume Render.
- Clear the Timer log.
- Increase the size of the mesh by 1, for instance to 200x200x201. Apply.
- Rotate the image. Possible bug. I kept having the screen go blank. Rotating the image again fixed this. Don't know who's bug this is (RGS or ParaView).
- Check the timer log.
Issue 1: We are spending LOTS of time in volume rendering, and I believe that more nodes won't scale. This serial behavior is killing us. Please fix it.
Issue 2: Even though the client says 60 seconds has elapsed, the server side says 20'ish seconds, and only 5 or 10 are accounted for. Timer log needs to be updated to account for this time.