I've built an online game with Unity with a very large open world. The server (which can be a local player, or run as a standalone application without a local player) simulates everything around players - so for example if there is an object or NPC at coordinates (100,34,20) but there aren't any players nearby then the object is not simulated. Objects are streamed dynamically from a database or generated procedurally depending on the object.
Problem
I want to enable players to go to other worlds. Of course it is impractical if not impossible to have global coordinates shared across all of these different worlds due to floating point limitations. I'll have to share the limited coordinate system for each simulated world.
The issue is the server would still need to render and handle physics of multiple worlds but could now potentially have conflicts with objects and/or people on one world appearing in other worlds. Essentially I need a way to divide up these "worlds" into parallel dimensions that are executed simultaneously on the same server (in one Unity scene).
Parallel Dimensions
A partial solution would be to have some sort of indicator, such as an integer (I'll call it "worldIndex") on each object to indicate which "world" that object exists in. This would work for clients since the server would be able to tell the clients only about objects that are on their world.
There would still be a problem on the server itself. Every world needs to be simulated in the same coordinate space. For a local player I could use the "worldIndex" to hide objects that aren't in the local player's current "world". But I still need to simulate physics for every world.
Unity's Layers could be used to mask physics interactions between objects. However I would then need to have at least one layer per world - and I already use multiple layers for various different interactions.
My thoughts are I would need something like a layer I could specify on each GameObject or some way to group a bunch of GameObjects together and instruct the physics system to only allow interactions within the group. But as far as I know no such mechanism exists within Unity.
Subspace Pockets
Another thought I've had is rather than using the Unity coordinate space directly, I could slice it up into "subspaces" or pockets. Each player would be in their own subspace, unless a player physically enters the same space that is already being simulated by another player, in which case that player would join the same subspace. But then I would have shift around the coordinate systems and it would increase complexity of merging players into and out of these subspaces. I would also have to generate subspaces dynamically as players enter new areas that are not currently being simulated.
Even if each world is extremely large, it is not necessary to use the entire Unity coordinate system around each player - a few thousand kilometers is more than enough.
Overall I think something along these lines is probably the best solution.
Multiple Servers
I could run one server per "world" and have a hand off process to make it transparent to players. But then I'd have to run multiple server instances and players would not be able to host their own server - at least not without a lot of additional overhead.
Question
So my question is: what is the best method for essentially running multiple, parallel simulations within a single Unity scene?