Client/Server Interpolation for Simulations

Talk at Vienna Dart meetup

Transcript

1. Client-Server Interpolation in dart by @slemgrim
2. How can we sync a simulation with multiple clients? It depends on your goals Number of players? Bandwidth? Latency?
3. Three approaches Deterministic Lockstep Snapshot Interpolation State Synchronization
4. Deterministic Lockstep The Simulation is running both on client and server Only player inputs are sent over the wire Simulation has to be deterministic Comparable physic states
5. Deterministic Lockstep
6. Pros Simple to understand Low bandwidth Bandwidth is not bound to number of objects
7. Cons Determinism is hard (float inaccuracy) Clients need to do physics themselves Gets harder with more players
8. Snapshot Interpolation Simulation runs only on the server States are sent to client Client interpolates and renders state
9. Client Interpolation
10. UDP over TCP No UDP in your browser WebSockets run on TCP Maybe in the future with WebRTC For now we forget about UDP, packet-loss and jittering
11. Interpolation „In the mathematical field of numerical analysis, interpolation is a method of constructing new data points within the range of a discrete set of known data points.“ - wikipedia
12. Linear Interpolation
13. Interpolation Buffer Client is a few snapshots behind of server (Latency)
14. Cubic polynomial interpolation
15. Interpolation Buffer Cubic interpolation needs control points. So we need more „future“
16. Pros Client is stupid and fast Easy to maintain
17. Cons High bandwidth (introduces snapshot optimization) Needs buffer for jitter and bandwidth (latency) We have to send everything that should be visible
18. State Synchronization Mix of deterministic lockstep and snapshot interpolation Game runs on server and client send state and inputs as often as possible No interpolation needed
19. State Synchronization
20. Pros Updates can be prioritized Bandwidth can be optimized easily Minimum added latency Buffer only for jitter not for interpolation
21. Cons Difficult to maintain New object types are a pain
22. Input prediction Imagine you have to wait until your input has been processed by the server Instead of waiting for the server to update your own position, the local client just predicts the results of its own user commands If server returns a different result, a correction must be made
23. Conclusion Deterministic lockstep for deterministic games with small player count (consoles) Snapshot interpolation for robust and easy and maintainable games (Which can afford the extra latency) State synchronization when low bandwidth and low latency is required
24. Sources http://gafferongames.com/ Networked physics by Glenn Fiedler https://developer.valvesoftware.com/wiki/ Source_Multiplayer_Networking Source Multiplayer Networking
25. Fin Examples: https://github.com/Slemgrim/syncball Twitter: @slemgrim E-Mail: slemgrim@gmail.com