Parrallel-Version FPS game based on Bullet

[michaelth]

I have made a FPS game reccently.The most import characteristic is

that its Server is running on 5 PCs.So more than 100 players can

easily play on ONE map simultaneously.Bullet on ervery sub-server

(PC) just provide calculations for Raycast and Bomb&Wall collison

detection.Because the player just attack another guy by Shooting

or throwing a bomb.Shooting can be easily implemented by Raycast

of Bullet, and Bomb&Wall collison detection as well.
Every sub-Server can hold more than 20 players.It receive and

process action data(KeyBoard&Mouse message) from player-clients

connected.Every player-client can receive all game-state data

(player's state action,bomb movement)from all sub-Servers and

these data are tansmitted through UDP protocol. So 3D-Rendering

Engine of every player-client can render the entire game.If one

Player connected to A sub-Server shoot another guy on B sub-

Server, A sub-Server just need send a Shoot Message(including

shoot-from point(btVector3 type),shoot-to point,who shoot,hurt

power etc) to B sub-Server.Therefore,B can get adequate data to do

a simple Raycast.Generally,2 bombs of different sub-Server will

have rare collision,so I just do Bomb&Wall collison detection on

every sub-server.
The communication delay is the bottleneck.And data transmission
between sub-Servers will increased as player's Number becomes

bigger.It's a big problem .

[Erwin Coumans]

I have made a FPS game reccently.The most import characteristic is

that its Server is running on 5 PCs.So more than 100 players can

easily play on ONE map simultaneously.Bullet on ervery sub-server

(PC) just provide calculations for Raycast and Bomb&Wall collison

detection.

That sounds interesting, networked collision detection. Do you have a link or further info?

The communication delay is the bottleneck.And data transmission
between sub-Servers will increased as player's Number becomes

bigger.It's a big problem .

How many packages of data are send, and how many bytes?

Thanks,
Erwin

[michaelth]

I have made a FPS game reccently.The most import characteristic is

that its Server is running on 5 PCs.So more than 100 players can

easily play on ONE map simultaneously.Bullet on ervery sub-server

(PC) just provide calculations for Raycast and Bomb&Wall collison

detection.

That sounds interesting, networked collision detection. Do you have a link or further info?

The communication delay is the bottleneck.And data transmission
between sub-Servers will increased as player's Number becomes

bigger.It's a big problem .

How many packages of data are send, and how many bytes?

Thanks,
Erwin

Networked collision detection has not been implemented completely in my program.Because the servers just do Networked raycast.And the bomb&wall collision detection will be done on every sub-server.If shooting and throwing bombs between different sub-servers increases greatly,the server-server UDP packets will increases greatly as well while sub-server also process packets from clients connected.So that's why I say The communication delay is the bottleneck.
For every second of the game,i assume there are 100 frames passing.In this second,the total communication data that one sub-server need to process consists of 3 part.
First part.In every game frame ,every sub-server firstly broadcast update message(include all connected players's state data(100 byte per player)) to all clients.100*100*20=200K
And meanwhile,20 players on every client send action data to server,the number of total data is unsure. I assume that every player moves and shoots uninterrupted.I don't know how many times my keyboard and mouse samples in 1 second. Assume it 100,so the total action data that one sub-server will receives in 1 second is :100*10*20*5=100K
The last part.I assume in 1 second,the sub server receive 100(mouse sample rate)*80(other 80 players)shooting message from other sub servers.The total data is: 100*80*20=160K.
So the total data per second is 460K.And I think there will be a good algorithm to reduce the total data.