Creación de Videojuegos
Networked Games
Networking
Why Networking?
Players enjoy interacting with other players
Killers like messing with people
Achievers like having leaderboards and showing off their achievements
Socializers like interacting with other people
AI is not (yet?) perfect
Networking Requirements
Fast: Should not adversely affect gameplay
Reliable: (Small) connection problems should not ruin the game
Fair: No player is preferred/affected by other player's connection problems
Secure: Players can't cheat
Peer-to-Peer Games
In the olden days, games would simply connect all players with each other
Every player runs the game on their machine
When someone moves, the move is sent to all players and their copy of the game state performs it
How can we synchronize the moves, so that all players have the same game state?
Lockstep Networking
Let's assume our game has "turns"
When it is a player's turn, they send the action they perform to everyone else
Players synchronize game states using turn order
Note: Real-time games can use this, too, by using "very short turns" (e.g. 10ms)
Lockstep Networking: Limitations
The game needs to be deterministic
Connection problems produce noticeable wait times
Worse, any player's connection problems affect every player
Also, what if the players do not agree on the game state (perhaps because someone is cheating)?
Client-Server Games
Instead of letting every player keep track of the game state themselves, let's use a central server
The server is the authority on the correctness of the game state, the clients only display it
When a player wants to perform an action, a request is sent to the server, and the action is only performed if the player is allowed to do so
Client-Server Games: Challenges
The server needs to respond to client requests quickly
However, there can still be a delay between the request and the answer
Clients should (ideally) not know anything the player should not know
What if there are network delays?
A word from John Carmack
"While I can remember and justify all of my decisions about networking from DOOM through Quake, the bottom line is that I was working with the wrong basic assumptions for doing a good internet game. My original design was targeted at < 200ms connection latencies. People that have a digital connection to the internet through a good provider get a pretty good game experience. Unfortunately, 99% of the world gets on with a slip or ppp connection over a modem, often through a crappy overcrowded ISP. This gives 300+ ms latencies, minimum. Client. User's modem. ISP's modem. Server. ISP's modem. User's modem. Client. God, that sucks. Ok, I made a bad call. I have a T1 to my house, so I just wasn't familiar with PPP life. I'm addressing it now."
Client-Side Prediction
In the original model, if you press a button, you have to wait for the server to answer to actually move
But the client could know/expect/predict that the button press will cause movement
Then, when the server confirms the coordinates, the client synchronizes with that information
Problem: The client tells the server about updates "in the past"
Dead-Reckoning
Remember Newton's first law? Objects in motion will stay in motion (unless a force acts on them)
The client can assume that for non-player objects
When the server sends another update, the client reconciles the actual information with their prediction
For example: If a character is moving forward, the client can keep moving them forward until the server tells them otherwise
The client can also use the game mechanics to predict changes in movement (e.g. through collision)
Example: With 100ms delay
t = 0: Player 1 sees player 2 in crosshair and presses "fire" button, client sends the "shoot" action to the server; client 1 predicts that player 2 was hit and killed
t = 1s: The server receives the "shoot" action from player 1, but player 2 is no longer in the crosshair
t = 2s: Client 1 gets response from server; Player 1 missed the shot, and player 2 isn't dead after all
Player 1 feels bad, because they "definitely" hit player 2
Lag Reduction
The server keeps track of past game states
When a player performs an action, the server applies it to the game state at the appropriate time in the past
This means the server must also re-apply other players' actions that happen afterwards
Advantage: Most of the time, a player's action will have the effect they see on their screen
Example: With 100ms delay
t = 0: Player 1 sees player 2 in crosshair and presses "fire" button, client sends the "shoot" action to the server; client 1 predicts that player 2 was hit and killed
t = 1s: The server receives the "shoot" action from player 1 with the timestamp information, and applies it to the game state at t = 0, resulting in a new game state in which player 2 is dead
t = 2s: Client 1 gets response from server, they hit player 2 and killed them
But what about player 2?
t = 0: Player 2 is visible to player 1
t = 1s: Player 2 is behind a wall
t = 2s: Client 2 gets a message from the server telling them that player 2 died (at t=0)
Player 2 feels bad because they were hit "while they were behind a wall"
Trade-off
This system is actually fair: Both players have the same chances
However, it favors active behavior, because a player's own actions are reflected in the game state more readily than actions that are performed on the player
In many cases, especially in shooters, players are running towards or away from their enemies
Often the delay is also not large enough for players to actually notice, since they also need some time to process what is happening to them
Keeping track of historical states is expensive for many players
Trust the Client?
As we discussed earlier: The client can not be trusted
How about Hybrid Hit Detection (such as in Battlefield 3)?
The client says when they hit someone
The server performs a plausibility test
Cheating
The server is the authority on the game state
However, to be able to predict movement, the client needs to know the positions of other players
By extracting information from the binary game state, cheat programs can "see" enemies through walls
Other cheating methods include disabling or manipulating the z-Buffer to draw hidden enemies
Anti-Cheating measures
There is no method that is 100% effective against all forms of cheating, but a few things that are done are:
Scan game binaries for modifications
Look for known cheat software that is running
Run game in a sandbox that can't be tampered with
Check for graphics card driver modifications/settings
Player Analytics
Why Player Analytics?
Games are expensive to make
Players often abandon games quickly
We want to know what players do in our games to keep them engaged
Related: We also want to make money
Player Analytics
We can collect data while players play the game
This data can then be analyzed to answer questions/confirm hypotheses
The answers to these questions can then inform game design
Typically, you want the players to keep playing your game, and use all your content
Or you might want to know what kind of content you should make more of
Player Retention
Player Retention
Good first impression/new player experience
Good challenge curve/flow
Delayed rewards (daily log-in bonus, loot chest with a key coming later, etc.)
Give player a personalization/progression choice
Social features, invite and team play rewards (see also: player acquisition)
Madden NFL
Madden NFL is a game where you play American football
Researchers (in collaboration with EA) analyzed game log data
They built a statistical model to predict how many games a player would play
Using this model, they identified features which result in higher player retention
Game design can focus on improving these features
Madden NFL: Features
Player Modeling
You can personalize content, or just focus on more common types of players
How do you determine "types" of players?
Your game can track player actions and report back!
Especially in open world games, determining what players spend their time on is beneficial to focus future development or guide players to underexplored areas
Player Profiles
On the other hand, you also want to know who plays your games
Geographic region, age range, maybe personal details, personality
Why? So you can target ads
Play patterns are also important: time of day, session length, number of games per week
Player Profiles: How do you get this data?
IP Geolocation tells you the region
Provide a birthday bonus to incentivize players to tell you their birthday
Connect to facebook to get personal details
Statistical models: Play style and personality have been shown to be related
Machine Learning for Player Analytics
Some machine learning techniques can be used to analyze collected data
For example, if we want to know which "types" of players exist, we can use clustering
If we want to predict what the player is likely to do, we can use supervised learning (e.g. a neural network)
We can also learn a player's preferences (e.g. which weapon they are likely to favor)
How to use this information in Game Play
Marketing is nice and important, but can we use this for actual mechanics?
Of course!
Say a large portion of your players typically play from 7-10pm
Put large in-game events at 8pm
You can cater the type of event to the typical player population
How to use this information in Game Play
If you know your players are going to favor a certain kind of weapon, you can focus on making more of these weapons in the future
You can even use the learned models during development of future versions to predict where players might go more
Player types can inform the kind of content you should focus on
Game Play Statistics
So you have your game and you collect data
Then you analyze what content which players are consuming and when
But you already invested all this time and effort?
Let's look at an example
Project Gotham Racing 4
Project Gotham Racing 4
Single player and multi player
10 different game modes: career, arcade, network race, tournaments, time attack, etc.
29 "event types": street race, elimination, cat and mouse, etc.
134 cars with 7 different ratings (easy to hard to control)
9 cities with a total of 121 race tracks
Development cost: "a fortune" (per Business Director Brian Woodhouse)
Project Gotham Racing 4: Analysis
4 researchers (including one from the publisher) analyzed 3.1 million game logs
Overall, 30-40% of the content was used in less than 1% of games
Game modes: Network tournaments were used in less than 0.1% of games
12 of the 29 event types were used in less than 1% of games
50 cars were used in less than 0.25% of races
47 race tracks were used in less than 0.5% of races each (13% together)
Project Gotham Racing 4: What to do?
First observation: More is not always better
The data can be used when developing a successor
But there would also have been a better way to develop this game:
- Release a small/core version of the game with less content
- Add content where it is needed
- You will spend less money overall, even though you provide "free content updates"
Claim: An analysis like this might cost $50-$100k, but will save $0.5-$2 million
Player Acquisition
Marketing!
For now, let's talk about mechanics that encourage player acquisition
Once you have players, you want them to invite their friends
The power of social connections
Player Acquisition
Say you have a game in which you tend to your farm
You can make money that allows you to buy more things for your farm
Some (fancy) things you can only buy with real money, though
Unless your friend (who also plays the game) "gives" them to you (they don't have to pay)
Of course you also have to tend to your crops (or they die)
But a friend can help you out on your farm
Player Acquisition: Other tactics
Get a bonus for posting a status on facebook that allows your friends to perform some in-game action
Make content harder to beat alone
There is a fine line between designing mechanics for player acquisition and (overly) aggressive marketing
The actual game play should still be enjoyable
But we don't have any game yet!
Many of the analytics options are only available to the developers
But there are APIs to extract replays for many games
You can run analytics on these replays to find out what players do in other games to learn their preference
This can inform design decisions for your own game
Unity Analytics
Unity Analytics
Unity has built-in support for data collection
To enable analytics, open the Services window under Windows -> General -> Services and switch Analytics to "On"
By default it will collect:
- New installs
- Daily/Monthly active users
- Sessions (total/per user)
- Time spent
- Users by geographic region, OS, life cycle
There are many other standard events you can use, or create custom ones
Also integrates with monetization options
Unity Analytics: Funnels
There are many ways to analyze the collected data
One of the more common ones: Funnels
You measure the progress through some linear progression (e.g. the tutorial)
The "funnel" will show you how far players make it through the progression, i.e. where you lose players
AnalyticsEvent.LevelStart(thisScene.name, thisScene.buildIndex);Unity Analytics: Funnels
Unity Analytics: Heatmaps
You may also be interested where in your game things happen
Where do players go often, where do they die
There is a Heatmap extension for Unity analytics, where you send analytics events with a location
Requires Unity Pro :(
Unity Analytics: Heatmaps
The EU General Data Protection Regulation
Remember getting emails from every website telling you that they collect personally identifiable information?
That's because of the EU GDPR
Use the Unity Analytics Data Privacy Plug-in
This plug-in includes a button you can place in your game for players to opt out of data collection (and see which data was already collected)
Make sure, you don't send personally identifiable information in your analytics events