cache Am I caching my fluid correctly?

[magsleyr]

Hi all, new to the forums and new to Houdini although I have a long history in Maya, I have never really gotten much into simulation until now. I've been messing around with the shelf beach tank to make a simulation of waves for a project I'm working on, but I'm a bit stuck on the caching phase.

Currently I have a file cache for my fluidcompress and then one for after my particlefluidsurface that then goes to render. Everything seemed to be okay, but I wasn't really able to scrub the timeline without long cook times seeming to come from the Import DOP I/O node. Rendering with Renderman also would get stuck on generating scene.

I saw that you can save to disk right in the DOP I/O, but I'm going with a final particle separation of about 0.07, ending up in 6 hour cache times and 100+ GB for 150 frames. I'm guessing this is normal? My machine is 32 GB RAM with a Ryzen 7 2700X CPU. Just don't want to waste time and space on something that was not the best solution.
I am left wondering if instead of the DOP I/O node I should be using a filecache after the delete_boundary_layer node, since that node seems to get rid of a lot of unneeded particles? Or maybe I am supposed to be caching in the beachtank_initial geo after the wavetank? Before the OUT null? I was under the impression that once I've cached the compression node that Houdini would have all the information it needs to render the surface that comes after and wouldn't need to "cook OP" the import_fields from the DOP I/O.

TL;DR Just caching compressed and particlefluidsurface I'm getting long cook times when going to new frames, so where else should I be caching before that to speed things up? I've attached a screen of my node tree to show what I mean.
Thanks in advance, really trying to wrap my head around the best practices and make sure I am not missing a step somewhere.

Edited August 28, 2020 by magsleyr

[Atom]

I just create the compressed cache first, which will require the full simulation time (NOTE: Whitewater will fetch its input from this node). Then I'll load that from disk and use it to generate the surface cache. If you are truly seeing a resim, perhaps another node is actually fetching from the dopnetwork, or you left the displayflag for the dopnetwork turned not (you don't need that, the import will fetch that data).

If you're not seeing the simulation message, perhaps what you are experiencing is simply slow load times of the cache data. That can't be avoided. Place cache data on SSD drive, if possible.

Try turning off your display flags and solo the import.

Edited August 28, 2020 by Atom

[magsleyr]

On 8/29/2020 at 5:18 AM, Atom said:

I just create the compressed cache first, which will require the full simulation time (NOTE: Whitewater will fetch its input from this node). Then I'll load that from disk and use it to generate the surface cache. If you are truly seeing a resim, perhaps another node is actually fetching from the dopnetwork, or you left the displayflag for the dopnetwork turned not (you don't need that, the import will fetch that data).

If you're not seeing the simulation message, perhaps what you are experiencing is simply slow load times of the cache data. That can't be avoided. Place cache data on SSD drive, if possible.

Try turning off your display flags and solo the import.

Thanks for the reply, you're right in that I only need the compressed cache for the fluid! I tested this by disconnecting the filecache from the main tree and everything is working as it should. I also went around and double checked to make sure everything was off that should be. I think when it comes to the long wait times for "generating scene" when rendering with Renderman, that probably has to do with what you said about having to load these big cache files. I find now there's almost no time to first pixel when rendering the fluid, but rendering the whitewater takes forever to get started, even cached. I'm starting to really get how everything is working now and once I got it, I never want to go back to Maya for this stuff haha. Although in one tutorial I'm following, the whitewater is generating source points from the surfaced fluid (more specifically, a separate slightly lower res fluidsurface set to Surface VDB rather than polygon soup), is there any advantages using the compressed cache to generate the source points?