[Opensim-dev] A modest proposal

[dan miller]

on physics:  I am strongly of the opinion that we should *not* globalize
physics at this point (ie have one physics scene for multiple regions)
.  I have some ideas on how to do it but it's a major undertaking.  As it
stands, spreading out the physics work to threads (in standalone) or
processes (in grid mode), associated with a single region, is our best bet
for scalability.

> Is there any possibility to start thinking about non-uniform regions at
> this point (i.e. 2 128x regions next to a 256x region), or is that still
> too embedded in the protocol?

If greater scaling & more efficient load distribution is the goal, I'd like
to propose thinking more about parallelization.  ISTM that the big problem
with the whole shebang is that when someone throws a party, you get 30 or 40
avatars in one place, and everything grinds to a halt.  Solving that problem
(at least on the server side) seems to be the holy grail.  Subdividing
regions into smaller units I think is likely to add more troubles -- border
crossing issues; inability to build across borders.  And in the end, if
everyone jumps into the mosh pit which happens to be in a particular region,
you're back where you started.  The resources required are proportional to
the number of objects in a scene, not the square footage.

Tedd's idea of a scripting API that allows scripts to run on other machines
is very appealing.  In a sense, you could think of physics as a very
specialized form of scripting, involving the interaction of entities in the
world according to formal rules based on information they exchange, such as
position, shape, and velocity.  I'd like to think about the physics API
being amenable to a remote interface as well.  Physics can also be
parallelized, meaning a high-traffic sim could take advantage of specialized
hardware or a cluster of machines to offload the physics calculations.

Just some late night thoughts...  I may put more detailed ideas on my blog.

-danx0r