mining second life: characterizing user mobility in a ... · introductionmining second...

Introduction Mining Second Life Measurement Methodology Results Conclusion

Mining Second Life:Characterizing User Mobility in a Popular

Virtual World

Chi-Anh La - Pietro Michiardi

ACM WOSN 2008Seattle, WA, U.S.

!""#$%&'(#$%)*+&'$,)-.)/&0)1$%.2$

3*+4'&5)67$/60$8*-9&)+:$2*;+#$

<6"+)+9+$=9&*-'>1$8';.)/$!6+);'?)"$

@&/6-*$

$

$

$

A'$B.'>$<+$,/:$C'6-*&61$

<$ ./D*$ /$ 7&*/+$ ;?*/"9&*$ )6$ 4&)+)67$ +.)"$ &*-'>>*60/+)'6$ ?*++*&$ /E'9+$ F*'&7)'"$

8>/&/70/5)"#$ <$ />$!"")"+/6+$ %&'(*""'&$ /+$ <6"+)+9+$ =9&*-'>$ ")6-*$ G/69/&:$ HIIJ$

/60$-/>*$ +'$56'4$F*'&7)'"$ )6$@*E&9/&:$HIIK$ )6$ +.*$ -'6+*L+$'($>:$D)")+$ +'$ +.*$

C'>;9+*&$ 8-)*6-*$ 2*;+#$ )6$ M'"+'6$ N6)D*&")+:1$ 4.*&*$ F*'&7)'"$ )"$ ;9&"9)67$ .)"$

%.2$ "+90)*"#$ F*'&7)'"$ ;&'>;+?:$ /++&/-+*0$ >:$ /++*6+)'6$ +'$ 6'D*?$ &*"*/&-.$

;&'E?*>"$+./+$*60*0$9;$)6$/$-?'"*$-'??/E'&/+)'6$")6-*$>:$()&"+$D)")+#$A.*$(&9)+$'($

"*D*&/?$ >'6+."$ '($ 4'&5$ )"$ /$ &*"*/&-.$ ;/;*&$ +./+$ ./"$ E**6$ /--*;+*0$ ('&$

;9E?)-/+)'6$)6$<===$<6('-'>$HIIO#$

F*'&7)'"$ )>;&*""*0$ >*$ /"$ .*$ ./"$ +.*$ +/?*6+$ '($ E*)67$ -&*/+)D*$ /60$ D*&:$

>*+)-9?'9"P$.*$-/6$"'?D*$0)(()-9?+$;&'E?*>"$E'+.$(&'>$/$+.*'&*+)-/?$/60$;&/-+)-/?$

"+/60$ ;')6+$ /60$ ./"$ /6$ *D*&Q?/"+)67$ *6*&7:$ +./+$ .)"$ 7&'9;$ /60$ >:"*?($ &*/??:$

*6R':*0$4.)?*$4'&5)67$+'7*+.*&#$$

29&)67$.)"$4'&5$4)+.$>*1$F*'&7)'"$./"$ ('-9"*0$'6$+.*$0*")761$ +.*$/6/?:")"$/60$

+.*$ ;*&('&>/6-*$ *D/?9/+)'6$ '($ *(()-)*6+$ 'D*&?/:$ 6*+4'&5$ +';'?'7)*"$ ('&$

E&'/0-/"+)67$ ?/&7*$ E9?5$ ()?*"$ />'67$ /$ ?/&7*$ "*+$ '($ 9"*&"#$ A.)"$ 4'&5$ "+*>"$

F*'&7)'"S"$'&)7)6/?$4'&5$'6$T"*?()".$6*)7.E'&$"*?*-+)'6U$+./+$./"$E**6$;9E?)".*0$

)6$ "*D*&/?$ +';$ -'6(*&*6-*"#$ F*'&7)'"$ 6'+$ '6?:$ ".'4*0$ .)"$ ?*/0*&".);$ "5)??"$ E9+$

/?"'$ 0*>'6"+&/+*0$ +'$ E*$ D*&:$ -'>;*+*6+$4)+.$ +.*$>/6:$ (/-*+"$ '($ +.*$ ;&'E?*>$

960*&$ )6D*"+)7/+)'6#$ !+$ +.*$ +.*'&*+)-$ ?*D*?1$ F*'&7)'"$ -'6+&)E9+*0$ +'$ +.*$

('960/+)'6"$ '($ +.*$ ';+)>)V/+)'6$ (&/>*4'&5$ 960*&?:)67$ '9&$4'&51$ 4.)?*$ /+$ +.*$

;&/-+)-/?$?*D*?1$.*$/-+)D*?:$-'6+&)E9+*0$+'$+.*$0*D*?';>*6+$'($/$-9"+'>$")>9?/+'&$

*6D)&'6>*6+$ /60$ /$ ?/&7*Q"-/?*$ +*"+QE*0$ ('&$ +.*$ ;*&('&>/6-*$ *D/?9/+)'6$ '($ +.*$

":"+*>$4*$4*&*$"+90:)67#$$

F*'&7)'"$)"$)6+*??)7*6+1$./&0$4'&5)671$';*6Q>)60*0$/60$-/6$*L;&*""$.)>"*?($D*&:$

-?*/&?:$ E'+.$ )6$ '&/?$ /60$ 4&)++*6$ ;&*"*6+/+)'6"#$ <$ ./D*$ ('960$ .)>$ +'$ E*$ /6$

*L-*??*6+$ -'Q4'&5*&1$ ./D)67$ /$ 7&*/+$ V*/?$ ('&$ '&)7)6/?$ &*"*/&-.$ /60$ E*)67$ D*&:$

-'';*&/+)D*$ )6$ 7&'9;Q/-+)D)+)*"#$ <$ 4'9?0$ -*&+/)6?:$ &*-'>>*60$ .)>$ ('&$ .)"$

/;;?)-/+)'6$ +'$ ,)-&'"'(+$ W*"*/&-.1$ E*?)*D)67$ +./+$ .*$ *6R':"$ /??$ +.*$ 6*-*""/&:$

X9/?)+)*"$+'$"9--**0$)6$.)"$"-)*6+)()-$;9&"9)+"#$

<$+&9?:$4)".$.)>$+.*$E*"+$'($"9--*""$)6$/??$.)"$(9+9&*$;&'(*"")'6/?$*60*/D'&"#$

<($:'9$./D*$(9&+.*&$X9*"+)'6"1$;?*/"*$0'$6'+$.*")+/+*$+'$-'6+/-+$>*#$

M*"+$&*7/&0"1$

%)*+&'$,)-.)/&0)$


Outline of the talk

1 Introduction

2 Mining Second Life

3 Measurement Methodology

4 Results

5 Conclusion


Characterizing human mobility:

Objectives of this work:Define a novel methodology to carry out experiments on humanmobility with the following goals:

Affordable experimentsNo logistic organizationWireless technology independentScalability of experiments


Related works

Objectives of prior works:Build mobility models from tracesPerformance evaluation of forwarding strategies in DTNs

Chaintreau et. al.: IEEE Trans. Mobile Computing 2007Karagiannis et. al.: ACM Mobicomm 2007Rhee et. al.: IEEE Infocom 2008


Related works: Experimental Methodology

Select hardware→ exhausting task

Neighbor discovery→ hard for wifi in ad-hoc mode

Prepare / finalize the experiment→ logistic problems


Related works: Restrictions

Available traces are difficult to use (and debug)

Experiments are bound to specific wireless hardware

In general, only “temporal” information is available

GPS-based experiments only for out-door scenarios

Number of participants to experiments is fixed


The idea

Exploit Virtual WorldsNetworked Virtual Environment are a tremendously popular con-cept of on-line communities:

User interaction is synchronousContrast with Social-Networking applications such asFaceBook: asynchronous interaction

In this work we use Second Life and capture user interaction aswell as user spatial distribution.


Our Playground: Second Life

Second Life architecture:Flat, Earth-like world simulated on a large server farmWorld is divided into 256x256 m “lands”, one server perland

→ Limitation on number of concurrent users on each land

Each land has attributes:privatepublicsandbox

→ Limitations on user-generated content deployment


Monitoring Architectures

Measurements in Second Life can be approached underdifferent angles

Use Second Life to build and deploy monitoring probesUse Second Life to mimic real world experimentsSystem approach: connect to Second Life and get data

We built a lightweight client wich crawls a selected land

Input:

Valid Login/passwdTarget LandMeasurement granularityMeasurement duration

Output:Anonymized user ID(x , y , z) of every user on the target land every τ seconds


The Crawler Approach

Observations:The crawler is a user→ should not introduce bias inexperiments

One crawler per land is sufficientAll users concurrently connected to the target land can betracked: we override a method used to build mapsMultiple lands can be tracked using an “army” of crawlers

Limitation: maximum number of concurrent users


Measurement Methodology

We present results for the following lands:Open Spaces:

Apfel Land: a german-speaking arena for newbiesIsland of View: Valentine’s day event

Confined areas:Dance Island: a virtual discotheque

Note:Selecting lands is a tedious manual exerciseAutomate the process


Using SecondLife Traces

How do we use the traces?Using the coordinates of users connected to a target land wecreate several snapshots of radio networks

Given a communication range r , a link between two usersui ,uj exists if their distance d(ui ,uj) ≤ rWe build snapshots every measurement interval τ = 10secr ∈ {rb, rw}, where rb = 10 m (bluetooth) and rw = 80 m(WiFi at 54 Mbps)


Metrics

Temporal:Contact TimeInter Contact Time

Spatial:Node degree distributionNetwork diameterClustering CoefficientZone occupation

Mobility:Cumlative traveled distanceLogin time


Results: Some Numbers

24-hours traces

Apfel Land:Unique visitors: 1568Average concurrent users: 13

Dance Island:Unique visitors: 3347Average concurrent users: 34

Isle of View:Unique visitors: 2656Average concurrent users: 65


Results: Temporal Analysis (1)

101 102 103 104 105

0.1

0.51

Time (s)

1−F(

x)

Contact Time CCDF, r=80m

ApfellandDanceIsle Of View

Contact Time = transferopportunities betweenusersLarge values are good

101 102 103 104 105

0.1

0.51

Time (s)

1−F(

x)

Inter−Contact Time CCDF, r=80m


Inter Contact Time = timeto wait before a pair meetsagainLarge values aresupposedly bad


Results: Trip Characteristics

0 500 1000 1500 2000 250000.10.2

0.40.50.6

0.80.9

1

Length (m)

F(x)

Travel Length CDF


Users do not exercise a lot!Closed vs. open spaces

0 5000 10000 15000 2000000.10.2

0.40.50.6

0.80.9

1

Time (s)

F(x)

Travel Time CDF


Max on-line time ∼ 4 h90-th perc. on-line < 1 h

Our explanation:Quite obvious (and similar to real world): users do not movewhen they chat!


Results: Spatial Distribution

0 5 10 15 20 250.8

0.85

0.9

0.95

1

Number of users per cell

F(x)

Zone Occupation CDF, L=20m


Not a uniform distributionMost of the users aregroupedClosed vs. open spaces

0100

200

0

100

2000

2

4

6

8

X

Zone Distribution, Isle Of View, L=20m

Y

Num

ber o

f use

rs p

er c

ell

0100

200

0

100

2000

2

4

6

8

X

Zone Distribution, Dance, L=20m

Y

Num

ber o

f use

rs p

er c

ell


Concluding remarks

Novel approach to study mobilityDo real people walk like avatars?

Beyond mobility analysisEpidemiologySociologyVirtual playground to test applications


Thank you!

Need traces?Contact: [email protected]

Web: www.eurecom.fr/∼ michiard

mailto://[email protected]

http://www.eurecom.fr/~michiard

mining second life: characterizing user mobility in a ... · introductionmining second...

Documents