Online Games:Networking and Data Management

Why do we “network” games?

• Playing by yourself is NO FUN!• Adds social dimension to gaming• But … we have the complexities of distributed

computing• From a systems perspective:

– “How do we cope with these physical restrictions, while developing and running our game?”

Taxonomy

Twitch games

Doom, etc.

Cooperative

Halo, Battlefield 2

MMOGs

Everquest, WoW

Turn-based

Civ, Poker

Deg

ree

of d

istr

ibut

ion

Degree of interaction

Lecture Outline

• LAN + Internet play– Latency– Scalability

• Data management– Tools

• Data-driven games

Networked Game “Data”

Shared game world

LAN Game Anatomy

Client Server

Physics

AI

UI

Physics

AI

Snapshot

…

“The Duel”

“The Duel”Server

time

“The Duel”Server

work

time

“The Duel”Server

work

time

“The Duel”Server

work

time

“Massively” Multiplayer

• Network latency– E.g. Current TCP/UDP

• NYC => Seoul, Korea ~ 150ms one-way

– At best speed of light constraint between players:• NYC => Seoul ~= 11000km/c ~= 36ms

• Scalability issues– Load balancing

– Data dissemination

Network Latency

• Two components:– Last-link latency– Server-server latency

• Last-link latency is hard to cope with:– Cable-modem => exchange– But no other route…

Dead Reckoning

• Predict to overcome latency!

• Prediction models:– Newtonian (linear)

• Not everything can be predicted …– Developer must think about consequences of

failed predictions

“The Duel” RevisitedServer

time

“The Duel” RevisitedServer

time

predict

Dead Reckoning (in more depth…)

• Videos

Data Consistency

• Client “cache”

• Static data consistency is easy….

• Dynamic consistency is harder…– Ownership issues– “Interest” management

Fairness Models

• Who is right, the shooter or the victim?

• What are other scoring models?– E.g. everything hurts a lot at first…

Growing the System

• Recap the LAN setup:

• Suppose we want 1000 players…

Client Server

Scalability: Load Balancing• Zone partitioning

• Each server may be distributed as a cluster• Server placement and geographical

partitioning helps…

Scalability: Data Dissemination• All about saving bandwidth• Tree network between zones

– i.e. pub/sub system

• Zone transitions:– Artificial impediment to gameplay

– “Content-based routing” and “profiles”

Data Dissemination: “Hotplugging”

• Open, dynamic systems

• Recently attempted by:– WoW– Star Wars Galaxies

• Should be seamless!

PlanetLab

645 nodes, 305 sites

Data Management

• “What kinds of data are used to build and run games?”• Content:

– Visual: models, animations, artwork– Audio: effects, speech, languages– Contextual: plot, missions

• Game logic:– Player profiles, inventories, etc.– NPC AI state.

• Transient:– Chat, VoIP

Data Management Tools

• Expressive, “declarative” approach– High-level triggers and rules

– Distribution transparency

– Clear-cut specification of game behaviour

– Modular frontend and backend extensibility

• Automated resource optimization– Scalable

– Adaptive

– Tolerant to failures

Dataflow Programming

• Flow-based representation of program– Based on events and triggers

• Game logic defined as triggers– E.g collision induced events

• Gameplay from “queries” over events– Supports roles– E.g. “tell me when there are 5 players near my

flag”

Dataflow Programming

• Borealis Demo

• Other distributed middleware kits:– Demonware, Quazcal, ReplicaNet– Opensource: MASSIV, Worldforge

Dataflow Programming

• What are good primitives for games?– E.g., collision detection, path planning, etc.– Class-specific primitives?

• E.g buy/sell items

– Come up with a high-level game template

Data-driven Gaming

• Let’s use the data out there!

• Content, plot and script generation.

• Game “embedding”– Splinter Cell, Battlefield 2– The Sims– Civilization– “Alternate reality” gaming

Mobile Gaming

• Sensors/actuators– Power concerns

– Wireless connectivity• Connection range

• Interruptions

• Data mules

• GPS• Camera phones• Google Maps

Data-driven Gaming

Pirate raid

Coastal mapMission objective

classify

Game entitiesGame level

Data sources

{

Data-driven Gaming Pipeline

• Acquisition– News feeds (RSS/Atom)

• Integration– Duplicate elimination– Event classification

• Morphing– Introducing “dissimilarities”– Consistency checking

• Deployment

Conclusions

• Latency causes inconsistencies in any distributed game.

• Scalable infrastructures to sustain game growth

• Data management tools, describing a high-level game representation

• Sources of game data

Thanks!

yna@cs