Hey everyone,

I'm a little indecisive when choosing how to render the gamestate on the client.

I can forward extrapolate the other clients to where they would be on the server, but then things like playing animations would need to be forwarded to compensate. As a result, the animation would "jump" into a frame; such that if the latency was half a second and the animation framerate was 60 frames per second, 30 frames would be skipped.

How should I deal with the timing issues?

1. Extrapolate forwards by the downstream latency to get the current position of the object
2. Render at time of receipt but still run client itself ahead using prediction

Forgetting about the jitter buffer, I'm concerned about hit detection; The first option would potentially be out by some margin, but the latter option would technically be incorrect as it was aiming at the client's old positions. Any thoughts?

So, for the sake of ease it is best to treat the latest incoming data as the current data.

How would one run a jitter buffer for this without doing something involving clock sync?

The predictions I send to the server will be the results of the inputs, therefore I'd only need to timestamp them to help prevent speed-hacking (I'd only need to know how long the inputs were processed for.)

In light of the fact that I'd need to re-simulate physics, I've decided to temporarily switch to another Python-based engine which gives me this ability, so that I can get the foundations working. Does this look like a permissible gameloop?

1. Receive network data.
2. Apply any corrections to physics data - there would have to be "real" positions and "render" positions, although this would make things more complex so I may just use interpolated positions for non-authority gameobjects.
3. Tick physics (perhaps a number of times if there are a number of saved moves).
4. Run simulation from inputs.
5. Tick Physics
6. Send results to server (with timestamp) -> delta = last_timestamp - current timestamp

Here I'm wondering how Unreal does it. By default, the replication statements are applied as soon as they're received. This would include physics. However the physics system scrapes the replicated data so It may choose to implement a jitter buffer. Would it then be sensible for the Physics "system" (that which iterates over the actors and scrapes the data) to implement the jitter buffer; sending a timestamp with the attributes?

How would one compensate for the scenario when physics are deliberately omitted from the replication (e.g when the object isn't moving)?
The two solutions I foresee would be a boolean "isIdle" that is replicated with the physics, or just sending it every replication regardless (or just reading the velocity value).

Furthermore, have you any suggestions on how to create the buffer? Because my attempts haven't been altogether succesful.