Transcript Online games

Internet Measurement
Online Games
1
Why Online Games?

One of the fastest growing areas of the Internet

More recently non-sequential and interactive
gaming has become popular

On-line games are big business


60% of all Americans play video games
Hosting games very costly (30% of revenue)
2
Online Games
Games and
Measurement
Properties
Networked
Games
Measurement
Challenges
State of the Art
3
Online Games
Networked
Games
Measurement
Challenges
State of the Art
4
Game Measurement Properties

Different genres:

First Person Shooters (FPS)

Most popular type of online gaming

High real time requirements

Real Time Strategy (RTS)

Massive Multiplayer Online Role Playing Games
(MMORPGs)

Different Platforms: XBOX, PS, Nintendo
5
Game Measurement Properties
Measured property
Why measured
Where measured
Traffic characterization
Growth patterns,
popularity
Across internet
Game system
architecture
Differences in
architecture
Across internet
Scalability
Provisioning,
performance
Varies with game genre
Real-time requirements
Game viability and
latency limitations
Game client
Manner of access
Mobility constraint
Client and server
locations
Session duration
State maintenance
server
6
Game Measurement Properties
Measured property
Why measured
Where measured
Traffic characterization
Growth patterns,
popularity
Across internet
Game system
architecture
Differences in
architecture
Across internet
Scalability
Provisioning,
performance
Varies with game genre
Real-time requirements
Game viability and
latency limitations
Game client
Manner of access
Mobility constraint
Client and server
locations
Session duration
State maintenance
server
7
Traffic Characterization

First property: Fraction of Internet traffic and individual popularity of
games

Sample traffic flowing to and from port numbers of games (like P2P)

Individual game characterization

Traditional metrics like size, inter-arrival time of packets

Behavioural differences between clients and server

Large amount of games trafiic take place over proprietary networks –
surveys used in these cases.
8
Game Measurement Properties
Measured property
Why measured
Where measured
Traffic characterization
Growth patterns,
popularity
Across internet
Game system
architecture
Differences in
architecture
Across internet
Scalability
Provisioning,
performance
Varies with game genre
Real-time requirements
Game viability and
latency limitations
Game client
Manner of access
Mobility constraint
Client and server
locations
Session duration
State maintenance
server
9
Online Games
Games and
Measurement
Properties
Networked
Games
Measurement
Challenges
State of the Art
Game
System
Architecture
Centralized
Decentralized
Hybrid
10
Online Games
Games and
Measurement
Properties
Networked
Games
Measurement
Challenges
State of the Art
Game
System
Architecture
Decentralized
Hybrid
11
Centralized Architecture 1

All interaction requests sent through a
central server

All clients not required to know movements
of all other clients at any given instant

Server decides what each client needs to
know
12
Centralized Architecture 2

Server requirements:
•
High processing capability
•
High reliability
•
Low latency/packet loss between clients and server

Used to prevent cheating amongst clients

Most commonly used architecture today
13
Centralized Architecture
Serve
r
14
Server responsibilities

Authentication

Updating positions

Maintaining scores/information about players
and teams

Managing forming of teams
15
Online Games
Games and
Measurement
Properties
Networked
Games
Measurement
Challenges
State of the Art
Game
System
Architecture
Centralized
Hybrid
16
Decentralized architecture 1

Clients interact with each other directly

Proposed decentralized architectures:

MiMaze

Mercury

P2P-Support

Zoned Federations
17
Decentralized architecture 2

Partial decentralization
•
Partitioning players and associated responsibility
into regions

Complete decentralization
•
Any peer in P2P network can carry out
authentication requirements to eliminate
cheating
18
Decentralized Architecture
19
Online Games
Games and
Measurement
Properties
Networked
Games
Measurement
Challenges
State of the Art
Game
System
Architecture
Centralized
Decentralized
20
Hybrid Architecture

One example: Mirrored server
•
Each game has several distributed servers
•
Clients only communicate with one of these
•
FreeMMG
21
Hybrid Architecture
22
Game Measurement Properties
Measured property
Why measured
Where measured
Traffic characterization
Growth patterns,
popularity
Across internet
Game system
architecture
Differences in
architecture
Across internet
Scalability
Provisioning,
performance
Varies with game genre
Real-time requirements
Game viability and
latency limitations
Game client
Manner of access
Mobility constraint
Client and server
locations
Session duration
State maintenance
server
23
Scalability

Number of players that can simultaneously participate in a
networked game


Typical numbers

<10 for RTS (Real Time Strategy)

10-30 FPS (First Person Shooter)

Thousands in MMOGs (Massively Multiplayer Online Games)
Increased users increased delays
24
Game Measurement Properties
Measured property
Why measured
Where measured
Traffic characterization
Growth patterns,
popularity
Across internet
Game system
architecture
Differences in
architecture
Across internet
Scalability
Provisioning,
performance
Varies with game genre
Real-time requirements
Game viability and
latency limitations
Game client
Manner of access
Mobility constraint
Client and server
locations
Session duration
State maintenance
server
25
Real-time requirements

Often the limiting factor in viability of a game

Varying requirements for latency and packet loss

Even within a single networked game, different
objects may require different real-time
standards

e.g., high accuracy sniper rifle vs. machine gun
26
Lightening Gun in Unreal Tournament
27
Game Measurement Properties
Measured property
Why measured
Where measured
Traffic characterization
Growth patterns,
popularity
Across internet
Game system
architecture
Differences in
architecture
Across internet
Scalability
Provisioning,
performance
Varies with game genre
Real-time requirements
Game viability and
latency limitations
Game client
Manner of access
Mobility constraint
Client and server
locations
Session duration
State maintenance
server
28
Manner of access
Wired and Mobile Environment

Physical location of client can be used

Increasing popularity of GPS and Bluetooth devices

Require accurate client location abilities

Active Bat, Cricket (indoor location systems), Human
Pacman

Most games require wired environment for lower
latency/packet loss
29
Human Pacman
30
Game Measurement Properties
Measured property
Why measured
Where measured
Traffic characterization
Growth patterns,
popularity
Across internet
Game system
architecture
Differences in
architecture
Across internet
Scalability
Provisioning,
performance
Varies with game genre
Real-time requirements
Game viability and
latency limitations
Game client
Manner of access
Mobility constraint
Client and server
locations
Session duration
State maintenance
server
31
Single session vs. Multi-session

Single session



User connects, plays, then exits game
More common among older games
Multi-session gaming

User logs in, plays, stalls session until next game

Value of character can grow or ebb during time

Increases necessity for network performance
32
World of Warcraft
33
Online Games
Games and
Measurement
Properties
Hidden Data
State of the Art
Hidden
Layers
Tools
34
Networked Games Measurement Challenges

Measurement challenges are significantly
different from DNS or Web  High
interactivity, lower tolerance for errors and
delays

Harder to simulate user traffic via programs 
Diversity of possible user action

Less datasets
35
Online Games
Games and
Measurement
Properties
State of the Art
Hidden
Layers
Tools
36
Hidden Data

Skill levels of users



impacts importance of latency/packet loss/etc.
No uniform way to measure impact of network problems
Information about game server rarely public, difficult to
reverse engineer (e.g. Algorithm server use to arbitrate jobs)

Downloading of new content can effect performance

Fortunately, usually no intermediaries between client and
server
37
Online Games
Games and
Measurement
Properties
Hidden Data
State of the Art
Tools
38
Hidden Layers 1

Games typically involve authentication, setting up
parameters, playing, and quitting

One or more steps may be avoided by suspending the state and
continuing from a previous session

Authentication generally done via TCP handshake

Game actions usually sent over UDP or TCP

Downloads of game updates sent over TCP

Less complex than short session applications (e.g., Web)
39
Hidden Layers 2

Quality of game effected by: Network, Client, Server, I/O
Devices

Delays cause different users to react differently

Delays on server end factored into measuring delays from
player's view

Team games add more complexity to measurements

Time of game effects impact of adverse network conditions

Location of player changes effect of network problems
40
Online Games
Games and
Measurement
Properties
Hidden Data
State of the Art
Hidden
Layers
41
Tools 1

Ping used to measure latency, latency radius
(number of active players within latency threshold)

Tcpdump for passive recording of game traffic

Geographic mapping tools used to locate game
servers

RTT measured at time of special events such as a
player dying
42
Tools 2

Traditional passive measurement
•
•
•
•
•

Average bandwidth
Packet inter-arrival time
Packet count and size
Number of attempted/successful connections
Unique clients
Non-traditional measurement tools
•
•
GameSpy tool used to report number of players
associated with game server
Qstat which display game server status
43
Online
Games
Games and
Measurement
Properties
Networked
Games
Measurement
Challenges
Architectural
Issues
Traffic
Characterizatio
n
Impact of
Network
Effects
Synthetic
Traffic Models
Generated
Mobile Game
Environments
44
Online
Games
Games and
Measurement
Properties
Networked
Games
Measurement
Challenges
Traffic
Characterizatio
n
Impact of
Network
Effects
Synthetic
Traffic Models
Generated
Mobile Game
Environments
45
Architecture Issues
Decentralized Game Architecture

MiMaze Game: Decentralized server
research

IP Multicast used for player moves

Latency limited to 100ms

Not a popular architecture

Presence of cheating

Lack of a synchronous view by all players
46
Online
Games
Games and
Measurement
Properties
Architectural
Issues
Networked
Games
Measurement
Challenges
Impact of
Network
Effects
Synthetic
Traffic Models
Generated
Mobile Game
Environments
47
Traffic Characterization
Some Characteristics / Quake and UT

Examined characteristics




Inter-arrival time of packets
Size of packets
Differences between client and server were examined
Data gathered passively using DAG cards (packet
capturing hardware)

Client packets were more numerous but smaller
than server packets in Quake
48
Traffic Characterization
Resource Requirements of Game Server / Counter Strike

Half a billion packets captured in 1 week from ~6000 players

Showed that updates must be predictable to compensate lag

Client/server packets maintained properties from Quake study

Regular traffic bursts are quite likely

Packets size vary

Active clients sent relatively uniform load
49
Traffic Characterization
Player behaviour / Half-Life and Quake

Player behaviour studied across a few thousand servers

How many players

How long they connected

Time-of-day effects game traffic

Players joined games with higher numbers of players

Duration a player continue to play appears to be independent
of number of players

Inter-arrival times of player follows a heavy-tailed
50
Time-of-day effects game traffic
51
Traffic Characterization
Session time / Counter Strike

Traces of one-week were gathered with GameSpy




Number of players
Session
Duration
Unlike most applications, session times followed a Weibull
distribution

Most players played for short durations

Testing new features of game

Study showed difficulty of generalizing network games
52
Session Time/Probability Chart
53
Online
Games
Games and
Measurement
Properties
Networked
Games
Measurement
Challenges
Architectural
Issues
Traffic
Characterizatio
n
Synthetic
Traffic Models
Generated
Mobile Game
Environments
54
Impact of Network Effects
Network Latency / Quake 3

Used servers in California and London

Intentionally masked London server as
California location

Found players chose servers closer to them
geographically

Bottleneck last mile between user and ISP
55
Impact of Network Effects
Unreal Tournament

Emulating packet loss and latency according to
live server data

Found no significant difference in ability to
move due to packet loss (prediction
compensation)

Even 100ms latency caused drop in perceived
performance
56
Online
Games
Games and
Measurement
Properties
Networked
Games
Measurement
Challenges
Architectural
Issues
Traffic
Characterizatio
n
Impact of
Network
Effects
Mobile Game
Environments
57
Synthetic Traffic Models Generated

Each game must be examined and synthesized separately

Representative set of players must be found and data
captured over a period of time


Skill of players will effect data
Typical information gathered




number of packets
packet length
Inter-arrival time
server response time
58
Online
Games
Games and
Measurement
Properties
Networked
Games
Measurement
Challenges
Architectural
Issues
Traffic
Characterizatio
n
Impact of
Network
Effects
Synthetic
Traffic Models
Generated
59
Mobile Environment

Human Pacman, Can You See Me Now?

Few measurements so far

Study on GAV (GPL Arcade Volleyball) game
ported to PDA found that wireless
environment could not support real time
requirements of GAV
60