Hi!!!

I have done a test to distribute a very simple flip sim test across 2 machines with no luck; The result always shows the first slice but the particles continue to die after living the emitter volume, like a clipping plane effect.
Then I have created an id attribute in the volume source asset and it works, but as the fluid pass to the next slice it tends to loss volume and then the particles dies (I have tested this with reseeding off, but no luck).

Please if anyone have a solution to correctly distribute a flip sim let me know.

Thanks!!!

Have the same issue, simple setup , slice 0 correct but slice1 empty, as if slice1 never gets the initial state from start so starts empty

I haven't tried in a long time (since early h11?), and it worked then, but I had to set it up manually to work on our farm. … Is the reason you want to distribute because you are being limited by ram? When I tested, the speed of 6(maybe 8?) machines was the same as a single machine, so we didn't really see the point. We ended up just putting more ram in a few machines and did sims on the single machines instead.

Ram is not the issue but the speed of iterations.

I have 64 gb ram on my machine, scene for RnD testing uses like 24 GB, but If I can slice it up I can push it a lot further.

I have one specific shot with a ship sailing in water with a single cam for 1200 frames, so i figured It might helped quiet a lot to divide the tank in slices.

Also I feel like i need to mentioned I only moved to houdini like 4 months ago from Maya, so I am no Houdini wiz.

tricecold
Ram is not the issue but the speed of iterations.

I have 64 gb ram on my machine, scene for RnD testing uses like 24 GB, but If I can slice it up I can push it a lot further.

I have one specific shot with a ship sailing in water with a single cam for 1200 frames, so i figured It might helped quiet a lot to divide the tank in slices.

Also I feel like i need to mentioned I only moved to houdini like 4 months ago from Maya, so I am no Houdini wiz.

The thing as Solitude said is that slicing it will probably not make it that much faster as there will be so much overhead having to be sent between the machines to get a correct simulation and network speed is so terribly much slower than ram speed.

Did anyone find a solution to this problem? I can definitely see the benefits of distributing. But I'm also running into issues when introducing a source and a sink.

Mike