There has been an ongoing discussion on physical prim maximum size. Lani Global
did some testing and published a graph of her results at
http://opensimulator.org/wiki/User:LaniGlobal#PhysicalPrimMax_Size_Testing .
Her testing showed that, for ODE, a max size of 32m resulted in a livable frame
rate degradation.
I performed the same tests with BulletSim and discovered that a single sphere
(native physical shape) or a single torus (mesh/hull shape) did not reduce
simulator frame rate no matter the size of the object.
So, I created a single, standalone region with lumpy terrain and dropped
different quantities of large shapes. The resulting frame rates were:
Number Number
SPHERE 1 4 8 16 32 TORUS[2] 1 4 8 32 64
+-------------------- +--------------------
10 | 55 55 55 55 55 10 | 55 55 55 55 55
S 20 | 55 55 55 55 55 S 20 | 55 55 55 26 13
i 32 | 55 55 55 55 55 i 32 | 55 55 40 16
z 64 | 55 55 55 49 49 z 64 | 55 50 20
e 100| 55 54 49 47 [1] e 100| 40 15
[1] My setup wasn't large enough to hold 32 100m spheres.
[2] The torus height is about one third of the width because I preferred the
doughnut shape.
A few pictures of the experiments (one region with walls):
https://dl.dropbox.com/u/86260377/20121107_009.png
https://dl.dropbox.com/u/86260377/20121107_012.png
https://dl.dropbox.com/u/86260377/20121107_017.png
https://dl.dropbox.com/u/86260377/20121107_019.png
https://dl.dropbox.com/u/86260377/20121107_022.png
https://dl.dropbox.com/u/86260377/20121107_026.png
It looks like, for BulletSim, a physical object max size of 32m is acceptable
and 64m would probably be OK.
The physics frame rate is a function of the number and size of large, physical
prims. The above numbers are for multiple of the same object. The next tests
should be on large sized linksets (for instance, two 5m cubes that are 32m
apart and linked and physical).
-- ra
_______________________________________________
Opensim-dev mailing list
[email protected]
https://lists.berlios.de/mailman/listinfo/opensim-dev
