[ODE] Accelerations

[Kenneth Holmlund]

Hi,
There are many attempts to solve the problems of distributed simulation, 
but not many have
tried to distribute full fledged mechanical systems.
A useful reference is: Distributed Networked Environments by Michael 
Zyda & Sandeep Singhal.
It is easy to be pessimistic about distributing full physics 
simulations, but I'm actually rather
optimistic since information about the physics should be more efficient 
to use, than
just pushing transformations and using dead reckoning.
Ultimately we would like the simulation to scale such that it becomes 
more and more efficient
the more cleints that connect to the system!
At the developer pages at www.havok.com there are a couple of 
interesting white papers
about "network dynamics".

One of the most efficient implementations of networked physics around 
can be found in
the games Clusterball and X-team-racing (see  www.daydream.se).
They can host a server on a computer with a GSM modem and attach up to 5 
clients on
9.6kbps! They do local simulations on each client and distribute updates 
dynamically
depending on game context.

/Kenneth

Adam Moravanszky wrote:

>>Say for example that a rocket is launched. By simulating it on the server,
>>we will know if the rocket will affect anything within the game. If it
>>doesnt; simply run the simulation locally without applying any effect. If
>>the server determines that something actually blows up, alert the clients.
>>Since the differences between each client will (hopefully) be minimal, no
>>serious artefacts should occur.
>>
>
>
>Sure sure.  Rockets and such.  That is the simple stuff older games were
>made of.  What I'm talking about is an articulated creature with arms and
>legs and so on, all of them separate entities.  Our tolerance here is much
>smaller since a disjointed knee is much more visible, even if the error is a
>few centimeters, than a rocket that is off even by a much greater distance.
>
>And all the nice (and hopefully accurate) dynamics simulation isn't worth
>much if it can't be percieved (or can only be 'roughly' perceived) on the
>clients.
>
>--Adam
>


-- 
Kenneth Holmlund
Director VRlab
HPC2N, Umeå University, Sweden
holmlund@hpc2n.umu.se
T: +46-90-786 9655
C: +46-70-631 5520