I'm trying to troubleshoot a simple simulation of an opening curtain. The curtain is a FEM cloth, the rings are keyframed static objects with volume colliders. I have set up several Cloth Attach constraints (for each of the rings) which prevent the cloth from detaching from the rings and speeds up simulation a bit:
(ajz3d_curtain_fem_test.mp4)

(Don't pay attention to cloth ending up in a slightly bent state. I think it's because of wind force which didn't have the offset animated fast enough at the time of flipbook rendering).

The problem that I have is that the simulation takes an insane amount of time to calculate. Last time I checked, the beginning of the sim is calculated pretty fast, but the more it approaches the end time, the more time it takes to simulate. Up to several minutes per frame.

Up to now I worked with DOPs only occasionally, so I guess being somewhat inexperienced in this department, I must be doing something terribly non-orthodox in my DOP network, and the extreme simulation time is the result of that.

Anyway, I also tried the PBD approach, but the result after an hour of parameter tweaking is still kind of rubbery in feeling, not to mention that the curtain falls out of the rings right after they begin to move. Would converting the rings from static objects to granular solids help here? Is it possible to attach constraints to PBD objects? How to prevent the “ruberiness”?
(ajz3d_curtain_pbd_test.mp4)
(Does PBD react to Wind Force?)

I'm including a couple of HIP files in hope of someone examining them. Both require this particular asset from Orbolt to work properly (especially the FEM-based scene): http://orbolt.com/asset/mifth::mifth_make_circle

FEM: (ajz3d_curtain_fem_test.hipnc.gz)
PBD: (ajz_curtain_pbd_test.hipnc.gz)

PS. Insertion of inline attachments doesn't seem to work.

I forgot to say, but the PBD scene was much, much faster to simulate. I think it took Houdini 20 minutes to calculate 180 frames. About 5 seconds per frame on average (with slight deviations). And the per frame time was pretty stable, which I guess is something to be expected from PBD.