Gameplay Networking Jacob Steinfort Importance of Multiplayer Games

Gameplay Networking Jacob Steinfort

Importance of Multiplayer Games • If gamers had to choose either a single-player game or a multiplayer game, most people will choose multiplayer • Why? – Beating your friends is more fun than beating an AI – Get a different experience every time you play the game

USA 2011 Top Unit Sales Multiplayer Single-Player

Importance of Getting Multiplayer Right • If developers provide a bad experience, people won’t buy the game

How do we get multiplayer right?

Computer Networking Review Part 1 • CS 3830 - Data Communications and Computer Networks – Latency = Delay, amount of time it takes a packet to travel from source to destination RTT • RTT (Round Trip Time)= source -> destination ->source • Usually measured in milliseconds (ms) – Bandwidth = amount of data that can be transferred per unit of time • Usually measured in Megabits / second (Mbps) Bandwidth

What is a perfect connection? – Ideally: • Infinitely small latency (0 ms) • Unlimited bandwidth (99999… Mbps) – Realistically:

Computer Networking Review Part 2 • Socket = bidirectional communication endpoint for sending and receiving data with another socket. • Two main types of sockets: – TCP (Transmission Control Protocol) – UDP (User Datagram Protocol)

TCP vs. UDP TCP UDP Connection based (handshake) No concept of connection, have to code this yourself Guaranteed reliable and ordered No guarantee of reliability or ordering of packets. They may arrive out of order, be duplicated, or not arrive at all! Automatically breaks up your data into packets for you You have to manually break up your data into packets and send them Makes sure it doesn’t send data too fast for the internet connection to handle (flow control) You have to make sure you don’t send data too fast for your internet connection to handle Easy to use, you just read and write data like its a file If a packet is lost, you need to devise some way to detect this, and resend that data if necessary Slow Fast

Why UDP? • Real-time requirement – For most parts of a game, it doesn’t matter what happened a second ago, you only care about the most recent data

Gameplay Networking • What is it? – Technology to help multiple players sustain the belief that they are playing a game together – Is it possible to achieve this? • Challenges: – Latency – Bandwidth – Dropped Packets – Cheaters – Joining/Quitting Games in progress

What type of games will I be talking about? • Action Games – Emphasizes physical challenges, including hand– eye coordination and reaction-time Best Example: SHOOTERS!

First technique of Gameplay Networking: Peer-to-Peer Lockstep • Each computer exchanging information with every other computer • Process: extract a game into series of turns and a set of command messages – Example: turn = 50 ms; set of commands = {move unit, attack unit, construct building, …} • What happens during a turn on one machine? 1. 2. 3. 4. Receive other player’s commands Evolve the game state Start recording commands Stop recording commands and send them to other peers

Peer-to-Peer Lockstep • Was created for RTS (Real Time Strategy) games – Game state was too large to transmit • Had to settle for transmitting changes only

Peer-to-Peer Lockstep • Deterministic – Will always produce same output when given same input – Synchronized: each machine is at the exact same state at any given time

Problems with Peer-to-Peer Lockstep 1. Game could become out of sync – One small change could destroy synchronization 2. Doesn’t support late join – Everybody needs to start from the same state 3. Everybody’s perceived latency is equal to the slowest latency – Necessary for consistency

Peer-to-Peer Lockstep • Does it work for action games? – Maybe on LAN, definitely not over the Internet • Problem: input latency Doom (1993): First action game to attempt to implement peer -to-peer lockstep

Computer Networking Review Part 3 • Client/Server model

Solution: Client/Server • Each client has only one connection: to the server • Clients turned into “dumb” terminals – Input sent to server (real-time) to be evaluated – Server sends updated game state (their player and all other players) to the client

2 Types of Client/Server: • Dedicated Server – Clients are the only players • Non-dedicated Server – Server is also a player – Server player is called “Host”

Client/Server Benefits 1. No more turns = Less latency – Other player’s latency will not slow you down Process changes

Client/Server Benefits 2. No more consistency issues – Game is only being simulated on the server (Peer-to-Peer Lockstep: game simulated on all machines)

Client/Server: Small Problem • Frame rate on client is limited to how frequently the server sent the game state to the client • Solution: Entity Interpolation!

Entity Interpolation Interpolated Game State / Rendered Frame Game State 2 Game State 1

Entity Interpolation with dropped packet safeguard ?

Client/Server with Entity Interpolation Benefits: • Provides a very smooth experience (unlimited framerate) that is much better than Peer-to. Peer Lockstep • Clients still run minimal code (no physics/collisions)

Client/Server: Big Problem + = LAG! (user-perceived latency) • Client 1 has an unfair advantage • If not dedicated server, huge host advantage

New Solution: Client-Side Prediction • As soon as user changes input, their machine predicts where the server is going to put them – Push forward on keyboard, instantaneously move forward on screen • Client needs to run more code – Client needs to be aware of physics and collisions (don’t want to run through a wall)

Client-Side Prediction (This is where it gets complicated) • After RTT has passed, the client gets the updated game state from the server and verifies that its game state WAS equal to what it predicted – If not, client performs correction User perceives lag Client Prediction Server Correction Client’s Modified Prediction

Client-Side Prediction How to implement? • Need a circular buffer on the client to store the user’s last few commands – When correction comes in from server, client performs new predictions based on the saved commands Command Client Prediction Predicted Actual Location Predicted Location Actual Location - Location Forward - 3 seconds (0, 3) (0, 2) Forward - 3 seconds (0, 3) Right - 4 seconds (3, 4) - (0, 2) Right - 4 seconds (4, 3) (4, 2) - (4, 2) Forward - 2 seconds (3, 6) Forward - 2 seconds (4, 5) (4, 4) … … -

Gears of War 1 (2006) and 2 (2008): Host Advantage 0: 57 Host 1: 05 Non-host

Another problem with Client/Server 1. I shoot another player and know it is a hit 2. Client sends shoot command to server – Contains point of origin, and direction 3. Server gets the command evaluates the shot – Server: client did not get the hit Why? The position of other players on the client’s machine is always out of date. You only see a “ghost” of other players.

Possible Solution: Give Client authority over shots • Client performs collision check right after shot – If hit, send a “hit confirmed” message to server • Problem? – CHEATERS! – Man-in-the-middle attack

Actual Solution: Lag Compensation 1. Client sends shoot command to server – Contains point of origin, and direction 2. Server gets command a) b) c) d) Estimates when the client executed the command using RTT Rolls back the game to the time of shot Simulates shot, checks for collision Server updates current game state Counter Stike: • View of Player 1’s point of view from the server • Player 2 is running to the left Player 2’s actual position on server Player 2’s position seen from Player 1’s client

Review of Basic Techniques • Peer-To-Peer Lockstep – Simple, works well for RTS games • Client/Server – Much better for action games, frame rate limited • Client/Server + Interpolation – Unlimited frame rate, still has large input latency • + Client-Side Prediction – No more input latency, interacting with other players is lagged though • + Lag Compensation – No lag with other players – Amazing!

Battlefield 3 Getting Gameplay Networking Right

Extra Tricks In Use Now • Game: • Developer: Halo: Reach Bungie Studios

A closer look at Client-Side Prediction • This technique works great for the character that the client is controlling • What about the other objects in the game (e. g. a grenade)

Entity Extrapolation

Entity Extrapolation • This is how it looks on a Halo: Reach client

Entity Extrapolation • Only makes sense for objects that have predictable paths – Grenades, rockets, anything not being controlled by a user • Doesn’t make sense to use on other players – Unpredictable direction changes – Have to stick with Interpolation for other players

Entity Extrapolation • It’s so complicated, why use it? • To reduce lag

Another Trick: Prioritization • A Halo game can have hundreds of objects – Some objects are less important to update on one client and more important to update on another • Solution: Real-time prioritization – Each object for each client has an update priority

Prioritization example 0. 22/0. 97/127 0. 50/1. 00/0 Legend: Final priority / relevance / desired update period (ms)

Bungie’s Networking Stack Internet Protocol Stack Layer Purpose Game Runs the game Game Interface Extract and apply replicated data Prioritization Rate the priority of all possible replication options Replication Protocols with various reliability guarantees Link Channel Manager Flow and congestion control Physical Transport Send & receive on sockets Application Transport Network

Check out Bungie’s Presentation • I Shot You First: Networking the Gameplay of HALO: REACH – Game Developers Conference, 2011

Sources • Aldridge, David. "I Shot You First: Networking the Gameplay of HALO: REACH. " GDC Vault. Game Developers Conference, 28 Mar. 2011. Web. <http: //www. gdcvault. com/play/1014345/I-Shot-You-First-Networking>. • Fiedler, Glenn. "Networking for Game Programmers. " Gaffer On Games. N. p. , 1 Oct. 2008. Web. 1 Sept. 2012. <http: //gafferongames. com/networking-for-game-programmers/>. • Kurose, James F. , and Keith W. Ross. Computer Networking: A Top-down Approach. Boston: Pearson/Addison Wesley, 2008. Print. • "Source Multiplayer Networking. " Valve Developer Community. Valve, n. d. Web. 1 Oct. 2012. <https: //developer. valvesoftware. com/wiki/Source_Multiplayer_Network ing>. • Steed, Anthony, and Manuel Fradinho. Oliveira. Networked Graphics: Building Networked Games and Virtual Environments. Burlington, MA: Morgan Kaufmann, 2010. Print.

Sources Videos: • Gears of War 2 Host Comparison – http: //www. youtube. com/watch? v=7 e. Tox. Vx. GO 9 k • Battlefield 3 – Jet vs Sniper – http: //www. youtube. com/watch? v=o 1 s 0 ED 51 Tic

Questions?