1. Managing and Mining Spatio-Temporal Data in Massive Multiplayer Online Games
Matthias Schubert
joined work with
Hans-Peter Kriegel and Andreas Züfle
Lehrstuhl für Datenbanksysteme
Institut für Informatik
Ludwig-Maximilians-Universität München

2. Outlook Spatio-Temporal Research and Computer Games
Managing Spatio-Temporal Game Data
Mining Game Data
Detecting Cheaters
Evaluating Game Balance
Conclusions

3. Spatio-Temporal Game Data
Player avatars or units move within a virtual spatio-temporal environment
A game server has to manage a unique valid representation of the game state
Movement and timing are essential aspects of most games
Similarity to other spatio-temporal applications: traffic management, simulations, surveillance systems etc.

4. Why do research for Computer Games ?
Computer games are big business (market volume 2010: gaming industry approx. 74 billion USD, relational databases approx. 20 billion USD)
There is data: spatial objects, events, trajectories, flocks,
There are applications: Modeling computer controlled entities, manage servers, analyze players, detect cheaters
It’s fun

5. Why is research interesting for companies ?
Trends in computer games:
stand alone games -> Online Games
1-10 players -> massive multi-player (1000+)
Console/PC -> Mobiles, Browsers ..
purchase price -> subscriptions , micro transactions.
Implications:
Games need distributed network technologies: Servers, P2P, synchronization, large data volumes, user management, spatial query processing
New threats: cheating, account hijacking, hacking…
New challenges: keep people investing time and money
Preview of the Diablo III auction house using real money

6. Management Challenges
Game Servers are high throughput
several thousand position updates in a tick (ca. 100 ms)
low jitter: a tick is required to take 100ms at most not on average
Large Games have to be distributed
a single server cannot maintain enough players
number of possible interactions increase quadratically
Distributing players depends on their spatial positions
fixed zoning via dynamic server relocation

7. Management Challenges
Persistency is mandatory
parts of the game state must be stored permanently
server must manage all relevant data (no local save games)
saving the game state must be done without extending tick processing times
(disk updates must be distributed over time several ticks)
Temporal Synchronization
low dependency on heterogeneous connection latencies
temporal uncertainty: Where is player A when her last position update arrived 2s ago.
reduce transferred data volume while maintaining a fluent game play (e.g. employ dead reckoning)

8. Mining Game Data Analyzing the player behavior to Detect Cheating
Bots (Programs playing the game for you: Farm bot)
Hacks (Modifications of the game client: Speed hack)
Exploits (Flaws in the game allowing unintended advantages: positions allowing to attack but not being attacked)
2. Evaluate Game Balance
Maintain a challenging but non-frustrating game play
Maintain a fair chance of winning for different character or faction types.

9. Why do players cheat? economical reasons
Sell ingame money or goods for real money: poker bots, goldfarming, account trading, item trading..:
Example: playerauctions.com
Over $1 billion USD in accumulated player-to-player trading value
Over 25,000 average daily transactions (nearly 20 per minute)
Over 700 supported Massively Multi-Player Online Games
Over 30 million accumulative transactions

10. Why do players cheat? Saving Time:
Example: AFK bots are self-running programs doing simple gaming tasks without user interactions like collecting ore or herbs.
Prestige:
Having success is often coupled with high prestige in the gaming community. Example: Reaching Masters League in Starcraft II
Fun:
Winning the game is simply fun even without the satisfaction that the success is well deserved
All types and motivations are a problem because
the gaming companies directly loses money (micro transactions)
the game becomes less attractive to other players which might quit (no fair competition)

11. Monitoring Cheats Challenges:
game state maintains current entity states
but: analyzing behavior requires recent states as well
monitoring might strain server resources
checking all player actions for a reasonable time period requires a lot of processing overhead
checking should be as generic as possible
(use rather outlier detection than supervized methods)
cheating players still have to be penalized (e.g. temporary ban)

12. Data Mining and Balancing
challenges:
detect interesting events like boss encounters in the logs
monitor encounter results
estimate player strength to remove the bias from statistics
formalize and cluster encounter tactics
=> More than one successful strategy indicates interesting game play
use text mining on community web sites to measure player happiness

13. Conclusions Computer games are an interesting application area
many games rely on a spatio-temporal virtual environment which is similar to monitoring and tracking systems
Recent developments will increase the need for managing and mining techniques
New challenges arise in real-time spatial querying, updating, persistency and server distribution
Cheat dection is a challenging task w.r.t. detection rates and throughput
Checking game fairness is statistically challenging
Measuring game difficulty and diversity requires combined consideration of game logs and community feedback

14. Join us on MAMIVE‘11 1st International Workshop on Managing and Mining Virtual Environments In conjunction with ACM SIGSPATIAL GIS in Chicago November the 1st, 2011 (Submission deadline September the 2nd)