multiple physics bubbles

[Capt Snuggler]

So as i understand it. physics in KSP is single threaded. so it can only be processed on a single core of your CPU. right?

we mostly all have multi core CPUs but they are currently being underutilized.

my suggestion is when objects (dropped stages) move outside of the 2.3km "physics bubble" have the physics for that object start on a new thread and therefore another core. maybe even use a pic in pic mode so you can monitor the decent or what ever.

so, is this even a possibility? or have i grossly misunderstood how these things work?

thank you for your time.

[Xaiier]

This might be possible in theory, but not in practice. The "bubble" is the unity gameworld, of which there can only be one.

[Kegereneku]

Maybe they could do a "pseudo-bubble" ? They did solved the problem they had with coordinates Origin's (or something) after all.

I'm more concerned about how to limit and deal with other physics bubbles. Inevitably you need to limit their amount, but also make them Active or Inactive on demand easily.

And even after all this is done you'd have to create delayed task for things you want to retrieve, delayed parachute opening after booster separation for example.

Edit : by the way, it is a given that any multiplayer implementation will require to solve this problem. We just don't know when they'll get to it and how important it will be.

Edited November 5, 2014 by Kegereneku

[Xaiier]

Maybe they could do a "pseudo-bubble" ? They did solved the problem they had with coordinates Origin's (or something) after all.

The floating origin method of keeping the vessel at (0,0,0) and moving everything else is simple. Having two vessels at (0,0,0) but far from each other is impossible. It would require you to have a completely different instance of the game-world and physics engine in order to process each ship.

Edit : by the way, it is a given that any multiplayer implementation will require to solve this problem. We just don't know when they'll get to it and how important it will be.

Multiplayer bypasses this by having multiple clients running

Each client can process it's own primary game-world and send that data to others for them to display at a distance.

[David H]

So as i understand it. physics in KSP is single threaded. so it can only be processed on a single core of your CPU. right?

we mostly all have multi core CPUs but they are currently being underutilized.

my suggestion is when objects (dropped stages) move outside of the 2.3km "physics bubble" have the physics for that object start on a new thread and therefore another core. maybe even use a pic in pic mode so you can monitor the decent or what ever.

so, is this even a possibility? or have i grossly misunderstood how these things work?

thank you for your time.

Is this why when running most games on my 7 core computer it runs most over 3gb. Even some of the older games. That were made before duel core was possible.

Then about every 10-15 mins. My screen freezes.

Should I force the game to only use 1 core.

If so. I fogot how to do that.

Help appreciated.

My older 4 core 4gb ram yr old laptop out performs this new 3 month old 7 core high end video card 12 GB of ram all the bells and whistles and still junk.

Edited November 6, 2014 by David H

[Red Iron Crown]

The floating origin method of keeping the vessel at (0,0,0) and moving everything else is simple. Having two vessels at (0,0,0) but far from each other is impossible. It would require you to have a completely different instance of the game-world and physics engine in order to process each ship.

So...that would give us multicore physics then?

Not totally kidding, on a multicore machine with much more RAM than KSP can use multiple game instances might actually be a useful solution. Would enable other things like multiple simultaneous views, too (IIRC about how that mod works).

[Murph]

There's a couple of issues I have with this thread so far. You're not really fully stating the problems that the multiple physics worlds would solve. It wouldn't begin to address the lag from large and complex craft, which is where we really need multi-threaded physics (i.e. what we really need is multi-threading of the physics for the current single 2.5km sphere).

Secondly, if there's some desire to solve the recovery of spent stages, we don't need a full physics simulation for that, just a very rough approximation of whether an object is recoverable or not, and then to auto-recover or auto-land it at the point where it's currently auto-deleted. Is there any reason that a vastly simplified yes-or-no approximation would not be sufficient for that? Adding a full physics sim just for spent stages is frankly a terrible idea, as it could cause horrible and entirely unnecessary lag for people on lower spec systems. Remember that there could potentially be many separate objects falling, possibly in many distinct 2.5km spheres.

If those are not the motivations for this, please expand on what the actual problems are that you're trying to solve, as that would probably help the discussion.

[Red Iron Crown]

Secondly, if there's some desire to solve the recovery of spent stages, we don't need a full physics simulation for that, just a very rough approximation of whether an object is recoverable or not, and then to auto-recover or auto-land it at the point where it's currently auto-deleted. Is there any reason that a vastly simplified yes-or-no approximation would not be sufficient for that? Adding a full physics sim just for spent stages is frankly a terrible idea, as it could cause horrible and entirely unnecessary lag for people on lower spec systems. Remember that there could potentially be many separate objects falling, possibly in many distinct 2.5km spheres.

While I agree for simple rocket stages, not all spent stages are created equal. Imagine a SpaceshipTwo-style craft, with an atmospheric mothership that detaches a rocket spaceplane that goes to space. Multiple bubbles would permit the mother ship to keep cruising along until the player switches back to it to land and recover. I'm not sure it's trivial to check if a plane can be returned to land, nor if it is even desirable to have the mothership automatically recovered at all.

[Capt Snuggler]

There's a couple of issues I have with this thread so far. You're not really fully stating the problems that the multiple physics worlds would solve. It wouldn't begin to address the lag from large and complex craft, which is where we really need multi-threaded physics (i.e. what we really need is multi-threading of the physics for the current single 2.5km sphere).

Secondly, if there's some desire to solve the recovery of spent stages, we don't need a full physics simulation for that, just a very rough approximation of whether an object is recoverable or not, and then to auto-recover or auto-land it at the point where it's currently auto-deleted. Is there any reason that a vastly simplified yes-or-no approximation would not be sufficient for that? Adding a full physics sim just for spent stages is frankly a terrible idea, as it could cause horrible and entirely unnecessary lag for people on lower spec systems. Remember that there could potentially be many separate objects falling, possibly in many distinct 2.5km spheres.

If those are not the motivations for this, please expand on what the actual problems are that you're trying to solve, as that would probably help the discussion.

wow, sorry if i have offended you.

I said dropped stages as it was the simplest example of a situation where multiple physics bubbles would come in to play.

My question was more focused on could it work, not why.

but another situation could be dropping probes or landers from a mother ship during aero breaking.

or as Red Iron Crown said a SpaceshipTwo or X-15 scenario where a rocket plane is released from a lifter aircraft. with two physics bubbles your lifter could cruse along safely while your rocket plane powers up and out of the atmosphere.

Or a SpaceX reusable launch system.

currently these scenarios range from very difficult to impossible.