At the core of coherence lies a fast network engine based on bitstreams and a data-oriented architecture, with numerous optimization techniques like delta compression, quantization and network LOD-ing ("Level of Detail") to minimize bandwidth and maximize performance.
The network engine supports multiple authority models:
Server authority with client prediction
Authority handover (request, steal)
Distributed authority (multiple simulators with seamless transition)
Deterministic client prediction with rollback ("GGPO"): coming in a future release
coherence supports persistence out of the box.
This means that the state of the world is preserved no matter if clients or simulators are connected to it or not. This way, you can create shared worlds where visitors have a lasting impact.