[ODE] ODE Physics Benchmark Setup

[cris@cris.hspace.name]

Date: Thu, 25 Oct 2007 15:18:41 +0800
From: Adrian Boeing <junk at adrianboeing.com>
Subject: [ODE] ODE Physics Benchmark Setup
To: ode at ode.org
Message-ID: <47204351.80809 at adrianboeing.com>
Content-Type: text/plain; charset=ISO-8859-1; format=flowed

> Hi,

> I was wondering what the best ODE configuration would be
in order to 
> achieve:
> a) The best accuracy (in terms of following the ideal
physical models)
> b) The best performance (in terms of computational
efficiency).

> In particular, I am referring to:
> 1. Which "stepper" to use (at the moment, I use WorldStep)
> 2. How the dSpaceCollide callback should be structured.

Hi,
first of all, I think you shouldn't use the WorldStep. It's
the old "big matrix" solver, and this could explain why in
some of the tests the performance drop dramaticaly down!
(e.g. the stack test...). I believe that all the other
engines use some sort of "iterative" solver.
Secondly, for my experience the "best" setup is really hard
to be defined... it depends on many factors of the scene, of
course, but using the "QuickStep" (iterative solver) you
will notice difference in accuracy/performance playing
around with the number of iterations.
Cristian