Consistency Maintenance in Multiplayer Games

1
Consistency Maintenance in Multiplayer Games

• Rob Fletcher
• CISC 897
• November 15, 2006

2
Outline

• The Problem
• Our Solution
• Why Our Solution Works

3
The Problem

• What is consistency maintenance?
• Why is it important?
• Why is it hard?
• What are the considerations?

4
Consistency Maintenance

• Consistency maintenance deals with keeping sets
  of replicated data equivalent
• In client-server games, the canonical worlds
  game state is stored on the server
• Some or all of the game state is replicated on
  each of the clients in order to be rendered and
  represented to the player

5
Consistency Maintenance

• Players interact with their replicas of the game
  world
• Any changes to the game world are sent back to
  the server
• Server then updates all the client replicas with
  these new changes

6
Why is it Important?

• Consider the following scenarios

7
Violent Action Game
8
Violent Action Game
9
Adventure Game
10
Adventure Game
11
Dress Up Games
12
Its Hard

• Many cheats in multiplayer games exploit
  consistency maintenance flaws
• Object duplication (duping)
• Perfect consistency algorithms do exist, however
  in games they are prohibitively expensive.
• If you dont get consistency maintenance right,
  players get angry, leave.

13
Its Hard

• Concurrency and Consistency maintenance
  algorithms (such as those using locking) ensure
  that all data stays the same
• Imagine if every movement you made in an FPS
  required several exchanges between the client and
  server
• Slower transactions are acceptable in a trading
  situation, inappropriate in high feedback
  interactions

14
Aspects of the Player Experience

• Choice of consistency maintenance scheme affects
  player experience
• Some of these aspects are fidelity, feedback,
  rate of animation, and warping.

15
Fidelity
16
Feedback
17
Rate of Animation
18
Warping
19
Trading off Aspects of the Player Experience

• In action games, you want very fast feedback
• You can trade fidelity for feedback by using
  predictive algorithms
• Predictive approaches are optimistic algorithms

20
Trade-Offs

• When high fidelity is required (such as in
  trading situations) you can exchange feedback for
  fidelity
• These are usually referred to as pessimistic or
  conservative algorithms

21
Previous Work

• There is a lot of research into consistency
  maintenance approaches
• Particularly in groupware research
• Multi-user collaborative drawing programs
• Multi-user collaborative text editors
• They understand trading off aspects of the user
  experience

22
Previous Work

• Pessimistic algorithms using locking
• Optimistic algorithms
• Rollback
• Operation Transform (Ellis and Gibbs)
• Timewarp (Edwards and Mynatt)

23
Our Solution

• Plug-replaceable consistency maintenance
• Workspace model provides us with the
  infrastructure
• Separate game logic and consistency maintenance
• Lets us quickly benchmark a handful of different
  consistency maintenance schemes

24
Quick Introduction to Workspace

• Proud product of the EQUIS lab
• Allows for a high level, diagrammatic definition
  of the intended behaviour of a distributed
  groupware system
• Refines it to an implementation level
  architecture at runtime based on available
  resources

25
Conceptual Level Architecture
26
Possible Replicated Refinement
27
CCCM

• Consistency and Concurrency Control Module
• Intercepts all calls to synchronized components
• Communicates with fellow CCCMs
• Embodies a consistency maintenance scheme

28
Many CCCMs

• We can create a handful of different CCCMs, each
  using a different consistency maintenance scheme
• Through the use of refinement hints, we can
  choose which CCCM to use for each set of data
• We can switch between them without recompiling to
  compare their behaviour

29
Why Our Solution Works

• We conducted a case study on a very simple game
• This game required a villager to traverse a
  series of waypoints
• Villager movement was controlled by the server,
  client simply observed
• Measured fidelity, rate of animation, and warping

30
Our Game
31
Controlled Variables

• We ran the experiment nine times, 15 minutes each
  time
• We used 3 different CCCMs under 3 different
  latency situations
• No Latency (0 ? 0ms)
• Low Latency (100 ? 10ms)
• Medium Latency (500 ? 100ms)
• Server sends update every 500ms

32
Our CCCMs

• Non-Predictive CCCM
• Client updates every message from server
• Client does not attempt to extrapolate villager
  position

33
Our CCCMs

• Two predictive CCCMs
• Extrapolates position between server messages
• Optimized Predictive CCCM estimates latency as
  well for higher fidelity

34
(No Transcript)
35
Conclusion

• Plug-replaceable CCCMs provide many advantages in
  game development
• Separation of concerns between consistency
  maintenance and game logic
• Ability to quickly compare various approaches
• Facilitates the development of a library of
  reusable CCCM components

36
References, Further Information and Questions

• R. D. S. Fletcher, T. C. N. Graham and C. Wolfe.
  Plug-Replaceable Consistency Maintenance for
  Multiplayer Games, in NetGames 06 Proceedings
  of the 5th Workshop on Network and System Support
  for Games, pages TBA, New York, NY, USA, 2006.
  ACM Press.
• fletcher_at_cs.queensu.ca