Phoenix SS - Tween and Client Prediction

[Vidiot_X]

Phoenix SS - Tween and Client Prediction

 

The yellow ship is a remote client shown on a server.

http://www.youtube.com/watch?v=gg-5EYyTcEM&feature=youtu.be

 

I think I am finally getting a handle on implementing a client prediction model for networking Phoenix SS ( Thanks Mark Sibly ) . I am able, with a variable timing tween system and the exact same physics, collision and tween frameworks for the server and client, render a very pleasing smooth render of the remote client.

 

Both the server and clients are sampling net traffic at once every 100 milliseconds. The system interpolates the missing frames for physics and collision by having all the clients do their own physics and collisions and the server does as well. The server will periodically check the results of the clients to make sure no cheating is going on. This is a very raw test with no packet checking, but the absolute minimization of latency using this method is very cool.

 

Because the clients and server both use the same exact frameworks the client predictions are deadly accurate and the server will ideally be checking the results of its map against the clients map, periodically checking the accuracy of the client results.

 

That's the basic idea.

 

- Rich -

[Cheese]

that was fast

looks good as well

 

also, you ended up doing exactly what this game does

lol

[Vidiot_X]

also, you ended up doing exactly what this game does

 

Yup. I looked around for a better model but I could never find one. It really is just the best way to approach it.

[Vidiot_X]

Hi,

 

I have been running simulations and optimizing code recently. It looks as if my baseline system (duel core low-end Vcard) should host up to 50 clients with about 100 bullet/bomb objects per client in a server playable mode (can play while hosting). As an unplayable (can 'not' play while hosting as a server) dedicated server it should be able to host 200 clients or more. It is hard to say ultimately how many clients can be hosted by a server. It will depend on the upstream bandwidth and CPU processing power.

 

There is also an issues related to the display view and the level of zoom. If the client display is zoomed way out it may need to keep track of more then 50 players at a time. 50 players is about the max a client can handle at this moment but may increase with more optimization. So it may end up that a server might have have several zones with a 50 player max limit. I am hoping to increase this limit and I am working on it.

 

All of the simulations I have run are at 100% max client usage. In the real world it is not likely that each client would be using all 100 bullet/Bomb objects at one time, so there is some headroom. Real world testing will flesh this out.

 

So, I am getting close to a working server and client. The next few releases will be ironing-out the bugs and developing the client prediction model and server as well as fixing changes to the core frameworks to accommodate networking. So close to getting playable.

 

Regards,

- Rich -

[Vidiot_X]

Hi,

 

I had three clients and the server all flying around the map today. It looks as if I am on the right track to network this game. I should have bullets whizzing around and exploding ships tomorrow. I am hoping to have a 'basic' implementation of a server/client this weekend where you can at least fly around and shoot at and kill other players.

 

Fingers crossed:)

- Rich -