Yes I must admit, considering the use of ragdoll would not be limited to death's, normal hits etc will have effects too. Really I am looking at areas where physics mught cause problems, namely in a multiplayer environment, and if that is the case then I would only use phsics in key areas, like post death animation, and other things. Also in a 64player map with say 100 standard physics items, how accurate is the system? what sort of variences if any can be seen? How does the engine determine priority, i.e we both shoot the same box in different directions at the same time, will it grant one over the other? Will both forces go into the calculation pot, and if so what if 3+ etc. As far as I understand phsyics there is not a true calculation yet that can accurately realise the bounce from 3 or more balls hitting together. This is a maths issue, not just a physics engine one.

So I am wondering really how reliable are any of the engines, is this something better avoided?

Anyway a new computer has just turned up so I have to oversee its delivery, lets hope their is a nice new shiny video card in there for me.....

Thanks for all your comments.

The three body problem you describe has to do with Gravitational attraction and the exact solution to it's equations of motion. There is no exact mathematical solution to this problem, though it can be numerically integrated with not problem.
Three balls colliding off each other has no math difficulties and from my experience with 3DGS PE, it's not a problem within the game.
All the other issues you mention are classic issues of MMOG design, not just focused around the physics engine. Do some research on MMOGs and Multiplayer Optimization and it might answer alot of these questions.

A dumb question but .... you mention how processor heavy the physics is. I was just wondering if it might not be possible to create a hybrid or blended use of physics. For example the player is struck and has the initial ragdoll effect which segues or blends into a standard death animation. In other words I have a standard animation for dying and so when player is hit he has a brief ragdoll effect which morphs to the start point for the regular animation -the physics is set up so the player rotation etc end matches the beginning of a standard animation.

Like I said may be dumb idea but what I mean is if you could have some of the realism of physics but reduce the overhead required by say reducing the actual physics calculations by half. Hope this was clear.

@bupaje: Yes that could be done, but it would be tough and you will probably see some popping from changing between a hinged ragdoll (separate models for arms, legs, etc) and a boned model.
I still think that Unreal Tournament 2003 has done it very well: they use animated or boned models and when critically hit convert it to a ragdoll and after a while disable physics on that ragdoll until respawn.

Realtime fluid dynamics in 2D is possible but wasteful since you can get good-looking results just by having a mass-spring grid and pretty textures. Coming back to UT2003, it seems that whenever something hits a water surface the closest vertex gets displaced and then the grid wiggles around until it reaches equilibrium.
Water grids and cloth and jelly could all be done in C-Script. Here's an excellent intro:
http://www.intel.com/cd/ids/developer/asmo-na/eng/segments/games/resources/physics/20413.htm

Going through this thread again, it is obvious to me now, that many of the questions I posted here, do not require a physics engine at all. The water and cloth (curtains, drapes, etc) can all be handled in a variety of ways, including bones, mesh-morphing and good old fashioned vertex-by-vertex mathematics. For example to create water that rises and falls as bodies break the water surface and move about within it: A simple spring maths system that returns all vertexs within the water mesh to a relative 0 z axis, bouncing between - & + z values, that adds to the z axis of the vertexs relative to the model that is influencing them, and the vertexs around them. (obviously the mathematics is a little more complicated than this). Combined with reflective shaders, bump mapping, and possibly a sqewed view mapped onto the water model, to create the natural light bending water illusion.

In short, many of my questions can be resolved without a physics engine, as Marco points out. Some of my questions are still valid though so I will push on.

Rag-Doll does not need to be used for normal movement/game play, it could only be used when a model is dead, in which case any variations of model position, would not effect the game play, and would not be exploitable. However we would like explosions and other forces like weapons fire, to be able to move phsyics objects, and for players to be effected in the same way. So models caught in an explosion, that survive, should be able to get back up, after the physics finished with the player, i.e player comes to a complete stop. So control would be past back to the player after this point.

So technically could we swap from mdl part/bones animation to rag-doll and then back again? And are there any issues with this, and which physics engine is best suited? Considering that we intend to build a 64 player multiplay option, so possibly 8-12 models getting effected by explosions, and other objects (hit by a car for example, I say this as we expect to have drivable-vehicles in our game)

Thanks again everyone

check out the new newton tech demo. it's very cool! -> www.physicsengine.com

I have already done that Ventilator, and as Newton is a good example of the current state or the...affordable......physics engines, my inclination to look to Havok, is more so. I appreciate that the Newton engine, can (almost) handle a lot of todays game physics, its techdemo is very buggy, and falls over easily. I tested the demo to breaking point and beyond, and found that although the basic engine worked ok, when you combined different complex physics Newton got confused and either had strange results, or just completely stopped working. I found that the swinging lights would disapear through the ceiling, the rag-doll bear on the fan, quickly got in a mess, and often once one part stopped, it all did.

Thats not to say that Newton is not impressive, far from it! Newton is very cool, its just not at a commercial level yet [IMHO], I still have not seen anything of the GS physics, but in truth that is becuase I am busy working on the games shader materials, (your wiki is very useful), and trying to understand this 9800 reflection bug, more on that in another thread!, so I have yet to spend enough time looking at physics engines.

So again I come to the question, Will Havok & GS work together?

Thanks for the interest Vent....

We are very sorry that Newton did not meet the high expectations of IndiGLOW. He has such very high standards that he should not use Newton, instead He should use something better (GS professional or Havok maybe).

Newton is for those that do not have the budget (more than hundred thousand dollars for the Havok case), the programming expertise, and a household brand for company name.

There are other alternative to Newton, but let me make this analogy. A physics engine is like a formula one car. They can run very fast but they are very temperamental and unstable, You need to be a really good pilot to drive them. This applies to any physics engine you try. But I think that IndiGLOW new that already. Let me tell you that I have evaluated many physics engines commercials and freeware. You can blame Newton for lacking some features (something that will be corrected soon), but when it comes to stability Newton come on top.

I am dying to see IndiGLOW Havok demos with Game Studio.
IndiGLOW we are sorry we lose you as a user. But we will still welcome you should you decide to come back.

Newton