1 Workshop on Online Social Networks Microsoft Research Cambridge Elizabeth Daly and Mads Haahr Distributed Systems Group, Computer Science Department.

1

Workshop on Online Social NetworksMicrosoft Research Cambridge

Elizabeth Daly and Mads HaahrDistributed Systems Group,

Computer Science DepartmentTrinity College, Dublin

Social Network Analysis for Routing in DisconnectedDelay-Tolerant MANETs

2

Introduction and Motivation

• Routing in a disconnected network graph– Traditional MANET Routing protocols fail– Store-carry-forward model used– Global view of network unavailable and

volatile

• Social Networks– Milgram’s ‘Small world’– Hsu and Helmy’s analysis of wireless

network

3

Related Work

• Deterministic– Assumes node movements are deterministic

• DataMULEs or Message Ferries– Assumes given nodes travel around the network

• Epidemic– Expensive in terms of resources

• History or Prediction– Captures direct and indirect social relationships– Problem:

• What if destination node is unknown to neighbouring nodes

4

Solution

• Exploit Social Network Analysis Techniques in order to:– Identify bridging ties

• Centrality

– Identify clusters• Similarity

5

Centrality Metrics [Freeman 1977,1979]• Degree centrality

– popular nodes in the network

• Closeness centrality– the distance of a given node to each node in the

network

• Betweenness centrality– the extent to which a node can facilitate

communication to other nodes in the network

N

kkiiD ppapC

1

),()(

N

kki

iC

ppd

NpC

1

),(

1)(

N

j

j

k jk

ijkiB g

pgpC

1

1

1

)()(

6

Ego Network Centrality Measures• Analysis of a node’s

local neighbourhood

N

kkiiD ppapC

1

4),()(s4

w6

w8 w7

w9 s2 i3 w4

w2

w3 iw5

w1 s1

Degree Centrality

ClosenessCentralityBetweennessCentrality

14

4

),(

1)(

1

N

kki

iC

ppd

NpC

N

j

j

k jk

ijkiB g

pgpC

1

1

1 3

1)()(

7

Egocentric Betweenness Correlation

Node SociocentricBetweenness

EgocentricBetweenness

w1 3.75 0.83

w2 0.25 0.25

w3 3.75 0.83

w4 3.75 0.83

w5 30 4

w6 0 0

w7 28.33 4.33

w8 0.33 0.33

w9 0.33 0.33

s1 1.5 0.25

s2 0 0

s4 0 0

i1 0 0

i2 0 0

w6

w8 w7

s4

w9 s2 i3 w4

w2

w3 i1w5

w1 s1

Marsden 2002

8

Similarity

• Social networks exhibit clustering• Increased common neighbours increases

probability of a relationship [Newman 2001]

• Similarity metric may be used to predict future interactions [Liben-Nowell,Kleinberg 2003]

• Represents similarity of social circles

)()(),( yNxNyxP

9

SimBet Routing

A B

HELLODeliver msgs

Exchange encounters

Add node encounters

Update betweenness

Update similarityCompare SimBet UtilityExchange Summary Vector

Add node encounters

Update betweenness

Update similarity Exchange messages

10

Betweenness Utility Calculation

0

1ijA

ijAA 12

• Node contacts represented in symmetric adjacency matrix

if there is a contact between i and j

otherwise

• Ego betweenness is given as the sum of the reciprocals of

w8 w6 w7 w9 s4w

8w6w7w9s4

0 1 1 1 1 1 0 1 1 0 1 1 0 1 1 1 1 1 0 1 1 0 1 1 0

=w8

* * * * * * * * * 3 * * * * * * * * * * * * * * *

w8 w6 w7 w9 s4w

8w6w7w9s4

=w82[1-w8][Everett and Borgatti 2005]

11

Similarity Utility Calculation

• Indirect Node contacts learnt during a node encounter is represented in and additional matrix

• Node similarity is a simple count of common neighbours

w8 w6 w7 w9 s4w

8w6w7w9s4

0 1 1 1 1 1 0 1 1 0 1 1 0 1 1 1 1 1 0 1 1 0 1 1 0

=w8

0 0 1 0 0

w5w8 w6 w7 w9 s4w

8w6w7w9s4

0 1 1 1 1 1 0 1 1 0 1 1 0 1 1 1 1 1 0 1 1 0 1 1 0

=w8

0 0 1 0 0

w5

12

SimBet Utility Calculation

)()(

)()(

dSimdSim

dSimdSimUtil

mn

nn

)()(

)()(

dBetdBet

dBetdBetUtil

mn

nn

nnn BetUtilSimUtildSimBetUtil )(

1

•Goal: to select node that represents the best trade off across both attributes

• Combined:

where

13

Simulation Setup

• Trace based simulation using MIT Reality Mining project data set– 100 users carrying Nokia 6660 for 9 months– Bluetooth sightings used as opportunity for

data exchange• Comparison

– Epidemic Routing [Vahdat and Becker 2000]– PRoPHET [Lindgren, Doria and Schelén 2004]

• Scenario 1: Each node generates a single message for all other nodes

• Scenario 2: Message exchange between least connected nodes

14

MIT Data set Egocentric Betweenness

15

Egocentric Betweenness Correlation

Pearson’s Correlation

16

Egocentric Betweenness

Friendship network Eagle and PentlandEgocentric Betweenness

17

Delivery Performance

18

Average End-To-End Delay

19

Average Number of Hops

20

Total Number of Forwards

21

Delivery Performance between least connected nodes

22

Conclusion

• Simple metrics for capturing network social structure suitable for disconnected delay-tolerant MANETs– Egocentric Betweenness – CentralitySimilarity

• Achieves comparable delivery performance compared to Epidemic Routing– But with lower delivery overhead

• Achieves delivery performance between least connected nodes

23

Questions…