[ODE] trimesh performance

[Geoff Carlton]

Wow, are you saying the colliding a single body against this 60-poly 
trimesh took 100ms just for the contact generation?  Or was it in the 
solver?  Even with some worst case Opcode problem it sounds extreme.

Geoff


Ian McMeans wrote:

> Oh, I hadn't been doing that. However, adding the cache clearing 
> doesn't seem to fix things (I'm doing it once a frame for now).
>
> Okay, I just tried using two different meshes in place of the one I 
> was previously using - one is a simple box (about 10 verts), and it 
> performed very well. The second mesh is decently complex (it's a 68kb 
> 3ds file), and it also performs quite well.
>
> Any idea what would be wrong with my "problem mesh"?  It didn't seem 
> to have a specific region of the mesh that was problematic - when it 
> was resting upside down, or on its side, or right-side-up, things were 
> equally slow.
>
> Thanks for your suggestion, though - ODE is back down to ~6 
> milliseconds, which is nice. I'll upload the problematic mesh 
> somewhere if people want it for testing (to see if they get the same 
> problems I do - it's only 29kb). I could also share (parts of) my 
> program.
>
> Ian
>
> Megan Fox wrote:
>
>> If I understand correctly, you're using a trimesh with a body (vs your
>> previous experience with terrain, which would not have a body). 
>> Wouldn't this be a case where the trimesh's temporal coherence cache
>> would be active?  Have you tried clearing it periodically?
>>
>> -Megan Fox
>>
>>  
>>
>>> I'm having some troubles with trimesh performance. I have a network 
>>> game
>>> where players control trimesh objects, and the performance is pretty
>>> terrible when trimeshes have resting contacts (ODE takes 100ms to do 
>>> the
>>> physics every frame). When I use a box placeholder, it runs quite fast,
>>> even with resting contacts (the ODE simulation takes about 0.6
>>> milliseconds of each frame).
>>>
>>> Now, I'm not totally sure what's causing this terrible performance. In
>>> my test situation, all that happens is a trimesh falls onto a box and
>>> rests there, it's not even trimesh-trimesh collision. However, when I
>>> make the trimesh fly (so it's no longer in resting contact), then ODE
>>> takes barely any time at all. So it seems that this has got 
>>> something to
>>> do with trimesh collisions. I'm pretty sure the problem isn't 
>>> related to
>>> networking, because the networking code almost never teleports objects,
>>> and even when I make the game 100% locally simulated, it doesn't 
>>> improve
>>> performance.
>>>
>>> Does anyone have any idea what would be causing such poor performance?
>>> It doesn't seem to be fixed by changing the number of max contacts in
>>> the collision callback. The trimesh is very small, only ~60 tris. Its
>>> size in-game is about 0.1 units. In other games, I've used trimesh
>>> terrain and it worked wonderfully, and the terrain had thousands of
>>> triangles.
>>> _______________________________________________
>>> ODE mailing list
>>> ODE at q12.org
>>> http://q12.org/mailman/listinfo/ode
>>>
>>>   
>>
>>
>>  
>>
>
> _______________________________________________
> ODE mailing list
> ODE at q12.org
> http://q12.org/mailman/listinfo/ode
>
>
>