decentralized routing in social networks

54
Geographic routing in social networks C. Herrera, T. Couronne, Z. Smoreda, C. M. Schneider, R. M. Benito, M. C. González

Upload: carlos-herrera-yaguee

Post on 11-May-2015

255 views

Category:

Education


5 download

DESCRIPTION

Slides for talk presented in Catedra Orange Network Course held in ETSIT-UPM, Madrid, November, 2012

TRANSCRIPT

Page 1: Decentralized routing in social networks

Geographic routing in social networks

C. Herrera, T. Couronne, Z. Smoreda, C. M. Schneider, R. M. Benito, M. C. González

Page 2: Decentralized routing in social networks
Page 3: Decentralized routing in social networks
Page 4: Decentralized routing in social networks

very short paths exist

Page 5: Decentralized routing in social networks

People were able to find them

Page 6: Decentralized routing in social networks

What’s been done so far?network models

Real-world experiments

Simulation on network

data

Page 7: Decentralized routing in social networks

What’s been done so far?network models

Real-world experiments

Simulation on network

data

Page 8: Decentralized routing in social networks

What’s been done so far?network models

Real-world experiments

Simulation on network

data

Page 9: Decentralized routing in social networks

What’s been done so far?network models

Real-world experiments

Simulation on network

data

Page 10: Decentralized routing in social networks

What’s been done so far?network models

Real-world experiments

Simulation on network

data

Page 11: Decentralized routing in social networks

What’s been done so far?network models

Real-world experiments

Simulation on network

data

Page 12: Decentralized routing in social networks

What’s been done so far?network models

Real-world experiments

Simulation on network

data

Page 13: Decentralized routing in social networks

What’s been done so far?network models

Real-world experiments

Simulation on network

data

Page 14: Decentralized routing in social networks

What’s been done so far?network models

Real-world experiments

Simulation on network

data

Page 15: Decentralized routing in social networks

What’s been done so far?network models

Real-world experiments

Simulation on network

data

Page 16: Decentralized routing in social networks

What’s been done so far?network models

Real-world experiments

Simulation on network

data

Page 17: Decentralized routing in social networks

What’s been done so far?network models

Real-world experiments

Simulation on network

data

Page 18: Decentralized routing in social networks

What’s been done so far?network models

Real-world experiments

Simulation on network

data

Page 19: Decentralized routing in social networks

What’s been done so far?network models

Real-world experiments

Simulation on network

data

Page 20: Decentralized routing in social networks

What’s been done so far?network models

Real-world experiments

Simulation on network

data

Page 21: Decentralized routing in social networks

What’s been done so far?network models

Real-world experiments

Simulation on network

data

Page 22: Decentralized routing in social networks

What’s been done so far?network models

Real-world experiments

Simulation on network

data

Page 23: Decentralized routing in social networks

What’s been done so far?network models

Real-world experiments

Simulation on network

data

Page 24: Decentralized routing in social networks

What’s been done so far?network models

Real-world experiments

Simulation on network

data

Page 25: Decentralized routing in social networks

Take home message

Routing in a real world network using only local information may be possible:

We need to understand the role of geography in the social links

Page 26: Decentralized routing in social networks

Pretty BIG Data3 phone networks country scope during 6 month

Mutual links considered (at least one interaction per direction)

User are geo-located to their most used tower

Page 27: Decentralized routing in social networks

Pretty BIG Data

1e−08 1e−07 1e−06 1e−05 1e−04 0.001 0.01 0.1 1

1 10 100 1000 10000k

P(k)

networkFrancePortugalSpain

Page 28: Decentralized routing in social networks

Long distance relationships?

-1.5<α<-1

Page 29: Decentralized routing in social networks

More likely short...

0.001 0.01 0.1 1

●●

●●

●●

●●

●●

●●●

●●●●●●●●

●●●

●●●

●●●

●●

●●●

●●●●●

●●

●●

●●●●●

●●

●●

●●

●●●●●●●●

●●●●●

●●●●●●

●●●●●●

●●●●●●●●●●

●●●●●●●●●●●●

●●●

●●●●●●●●●●●●●●●●

●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●

●●●

●●●●●●

●●●●●●●

●●●

●●●

●●●

●●●●●●●●●

●●●●

●●●●●

●●●●●●

●●●●●●

●●●●●●

●●●●

●●●●

●●●

●●●●

●●●●

●●●

●●● ●●●

0 10 20 30 40 50d (km)

P(d)

network●

FrancePortugalSpain

Page 30: Decentralized routing in social networks

Let’s route!“City”=province (20 Portugal, 96 France, 52 Spain)

We try to deliver the message to the city

Experiment conditions:• 60K Random pairs• Error if not delivered in 100 hops

Page 31: Decentralized routing in social networks

Let’s route!“City”=province (20 Portugal, 96 France, 52 Spain)

We try to deliver the message to the city

Experiment conditions:• 60K Random pairs• Error if not delivered in 100 hops

Page 32: Decentralized routing in social networks

Let’s route!Proposed methods

DFS: we don’t send to others who already got the message, in case everybody got, send back to the first person who sent us the message

GEO: send to the friend geographically closest to target

DEG: send to the best connected friend

Page 33: Decentralized routing in social networks

Let’s route!

●●●●

●●

●●

●●

●●

●●

●●

●●

●●●

●●●

●●●

●●●

●●●

●●●

●●●

●●●

●●

●●●

●●●

●●●

●●

●●

●●●

●●●

●●●

●●

●●●

●●●●

●●●

●●

●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●

●●●●

●●

●●

●●

●●

●●

●●●

●●●

●●●

●●●

●●●

●●●

●●●●●●●

●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●

●●●●

●●

●●

●●

●●

●●●

●●●

●●●

●●●●

●●●●

●●●●●●●●●●

●●●

●●●●

●●●

●●●●

●●●

●●●●

●●●●

●●●

●●●●

●●●

●●●●

●●●●

●●●

●●●●

●●●●

●●●●

●●●●

●●●●

●●●

●●●

●●●

●●●

●●●●

●●●●●●●●

●●●●

●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●

france portugal spain

0.00

0.05

0.10

0.15

0.20

10 100 1000 10 100 1000 10 100 1000l − 1

freq

exp●

routing−dfsrouting−dfs−degrouting−dfs−georouting−dfs−geo−deg

Page 34: Decentralized routing in social networks

Let’s route!

10

15

20

25

30

●●

1e−04 0.001 0.01 0.1 1Error Rate

<l>

network● france

portugalspain

algorithm●

routing−dfsrouting−dfs−georouting−dfs−degrouting−dfs−geo−deg

Page 35: Decentralized routing in social networks

Let’s route!

10

20

30

●●●

●● ●

●●

●●●

● ●●●

●●●

●●●

●●

●●

● ●

●●

●●●

● ●●●●

●●●● ●

●●

●●●

●●

●●

●●

●●

● ●

●●

●●

●●

●●

●●●

● ●

●●

0.001 0.010 0.100Relative Province Size

<l>

users population●

●●

0.10.20.30.40.50.60.7

network●

FrancePortugalSpain

Page 36: Decentralized routing in social networks

Let’s route!“City”=municipality (305 Portugal, 3520 France, 8410 Spain)

We try to deliver the message to the city

Experiment conditions:• 60K Random pairs• Error if not delivered in 100 hops

Page 37: Decentralized routing in social networks

Let’s route!

10

20

30

40

●●

1e−04 0.001 0.01 0.1 1Error Rate

<l>

network● france

portugalspain

algorithm●

routing−dfsrouting−dfs−degrouting−dfs−georouting−dfs−geo−deg

Page 38: Decentralized routing in social networks

Let’s route!

10

20

30

40

●●

1e−04 0.001 0.01 0.1 1Error Rate

<l>

network● france

portugalspain

algorithm●

routing−dfsrouting−dfs−degrouting−dfs−georouting−dfs−geo−deg

Page 39: Decentralized routing in social networks

Let’s route!

10

20

30

40

●●

1e−04 0.001 0.01 0.1 1Error Rate

<l>

network● france

portugalspain

algorithm●

routing−dfsrouting−dfs−degrouting−dfs−georouting−dfs−geo−deg

Page 40: Decentralized routing in social networks

Let’s route!

10

20

30

40

●●

1e−04 0.001 0.01 0.1 1Error Rate

<l>

network● france

portugalspain

algorithm●

routing−dfsrouting−dfs−degrouting−dfs−georouting−dfs−geo−deg

Page 41: Decentralized routing in social networks

Let’s route!

10

20

30

40

●●

1e−04 0.001 0.01 0.1 1Error Rate

<l>

network● france

portugalspain

algorithm●

routing−dfsrouting−dfs−degrouting−dfs−georouting−dfs−geo−deg

Page 42: Decentralized routing in social networks

Let’s route!

0

20

40

60

80

100

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●●

●●

●●

●●

●●

●●

●●

●●

●●

● ●

●●

●●

●●●

●●

●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●●

●●●

●●

●●

●●●

●●

●●●

●●

●●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●●●

●●

●●

●●

●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●●

●●

●●

●●

●●

●●

●●

●●

●●●

●●●

●●

●●

●●

●●

●●

●●

●●

●●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●●

●●●

●●

●●

●●

●●

●●

●●

●●

●●●

●●

●●

●●

●●

●●

●●

●●

●●●

●●

●●●●

●●

●●

●●

●●●

●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●●

●●

●●

●●

●●●

●●

●●

●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●●

●●

●●

●●

●●●

●●

●●

●●●

●●

0.001 0.100Relative Province Size

<l>

network●

FrancePortugalSpain

Relative City Size

Page 43: Decentralized routing in social networks

Let’s route! IntracityMore difficult, success is reaching the target

Location is not so important

Add new criteria

COM: community vector, try to forward to the one in the same community than target (communities obtained via Louvain Method)

Page 44: Decentralized routing in social networks

Paris Lisbon Madrid

0.000

0.002

0.004

0.006

0.008

0.010

0 20 40 60 80 1000 20 40 60 80 1000 20 40 60 80 100l

P(l)

Algortihmrouting−dfsrouting−com−degrouting−dfs−degrouting−dfs−comrouting−dfs−com−degrouting−dfs−com−geo−deg

Let’s route! Intracity

Page 45: Decentralized routing in social networks

Let’s route! Intracity

●●

Paris Lisbon Madrid

30

40

50

0.5 0.6 0.7 0.8 0.9 1.00.5 0.6 0.7 0.8 0.9 1.00.5 0.6 0.7 0.8 0.9 1.0Error Rate

<l>

Algorithm●

routing−dfsrouting−com−degrouting−dfs−degrouting−dfs−comrouting−dfs−com−degrouting−dfs−com−geo−deg

Page 46: Decentralized routing in social networks

Let’s route! Intracity

●●

Paris Lisbon Madrid

30

40

50

0.5 0.6 0.7 0.8 0.9 1.00.5 0.6 0.7 0.8 0.9 1.00.5 0.6 0.7 0.8 0.9 1.0Error Rate

<l>

Algorithm●

routing−dfsrouting−com−degrouting−dfs−degrouting−dfs−comrouting−dfs−com−degrouting−dfs−com−geo−deg

Page 47: Decentralized routing in social networks

Let’s route! Intracity Porto Lisboa Setúbal Braga Aveiro

Santarém Viseu Vila Real Bragança Coimbra

Viana do Castelo Leiria Évora Faro Portalegre

Castelo Branco Beja Guarda Madeira Azores

102030405060

102030405060

102030405060

102030405060

0.0 0.2 0.4 0.6 0.8 1.00.0 0.2 0.4 0.6 0.8 1.00.0 0.2 0.4 0.6 0.8 1.00.0 0.2 0.4 0.6 0.8 1.00.0 0.2 0.4 0.6 0.8 1.0Error Rate

<l>

Algorithmrouting−com−degrouting−com−geo−degrouting−dfsrouting−dfs−comrouting−dfs−com−degrouting−dfs−com−georouting−dfs−com−geo−degrouting−dfs−degrouting−dfs−geo−deg

Page 48: Decentralized routing in social networks

Let’s route! Intracity

0.0

0.2

0.4

0.6

0.8

● ●● ●

● ●● ●● ● ●●

●● ●●● ●●●● ●

● ● ●● ●● ●●● ●● ●● ●● ● ●●●●●●

● ●● ●●

●●● ●● ●● ●●

● ● ●● ●● ●● ● ●● ●●

●● ●● ● ●● ● ● ●●● ● ●●

●●● ● ●●● ● ●●● ●● ●● ●●● ●● ●● ●● ●● ●●● ●●

● ● ●●● ●● ● ●●● ●● ● ●●● ●●● ●● ● ● ●● ●● ●●● ●● ● ●●●●● ●● ●●● ●● ● ●

● ●●● ●● ●● ●●● ●● ●●● ● ●●● ●● ●● ●● ●●●

● ● ●● ●● ●● ●●●●● ●●● ●● ●●●● ● ●● ●●● ● ●●●● ●● ●●●● ● ● ●●

●● ●●● ●●●●● ●● ●● ●● ●

● ●●●

● ●●●● ●● ● ●● ●

●● ● ● ●● ● ●

●●● ●

● ● ● ●●

● ●

●●●

●●● ●

20 30 40 50avg

err

network●

franceportugalspain

#Nodes●

2e+054e+056e+058e+05

level● municipality

province

Page 49: Decentralized routing in social networks

Let’s route! Intracity

0.2

0.4

0.6

0.8

0 500 1000 1500 2000 2500 3000

# Nodes u 103

E dfs<c

om<d

eg networkFrancePortugalSpain

Page 50: Decentralized routing in social networks

Let’s route! Intracity

0.0

0.2

0.4

0.6

0.8

●●●●

●●● ●● ●● ●●● ● ●● ●●● ●●● ● ●● ● ● ●● ● ●●● ● ●●●● ●●●

●●● ●●●● ●●● ● ●●● ● ●●● ● ●●●●● ● ●●●● ● ● ●● ● ●● ●● ● ●●● ●● ● ●●●●●● ● ● ●● ● ●● ●●● ● ● ● ● ●●●●● ● ● ●●● ● ●●●● ● ●●●● ●●● ● ●●●● ●●● ●● ●●● ●●● ● ●●● ●●● ● ●● ●●● ●● ● ●●● ●

●●●● ● ●●● ●●●● ● ●● ● ●●● ●●● ● ●●●● ●●● ●● ● ●● ●●● ● ● ●●● ● ●● ● ●●● ●● ●● ● ● ●●● ●●● ●●● ●● ●●●● ●●● ● ●● ●●● ●● ●● ● ●●●●● ●● ●

●● ● ● ● ● ●● ●● ● ●● ●● ●●● ●●● ●●

●●● ●● ●●

●●● ●

●●● ●

1000 10000 1e+05 1e+06#Nodes

E dfs−c

om−d

eg

network●

franceportugalspain

level● municipality

province

Page 51: Decentralized routing in social networks

Let’s route! Intracity

0.0

0.2

0.4

0.6

0.8

●● ●●

●● ● ●● ●●●● ● ● ●● ●● ● ●●● ● ●●● ● ●● ● ●●● ● ●●●● ●● ●

●● ●● ●●● ●●● ● ●●● ● ● ●● ●● ●●●● ● ●●●● ●●●● ● ●●●● ● ●●● ●● ● ●● ●●● ●●● ●● ● ●● ●● ●●● ● ● ●●● ●● ●● ●●● ●●●●● ● ●●● ● ●● ●● ●●●●● ●● ●● ●● ●●● ●● ●● ● ●●● ● ●● ●●● ● ●● ● ●●●

●●●● ● ●●● ●●●●●● ● ●● ●● ●●●● ●●●● ●●● ●● ● ●● ●●● ● ● ●●● ● ●●●●●● ●● ●●●● ●●● ●●● ●●● ●● ●●●● ●●●● ●●●● ● ●●● ●● ●●

●●● ●● ●●● ● ● ●● ●● ●●● ●

●●● ● ●● ●● ● ●●●●●●

● ●●

●● ●●

●●● ●

4 6 8<k>

E dfs−com

−deg

network●

franceportugalspain

level● municipality

province

Page 52: Decentralized routing in social networks

Let’s route! Intracity

0.0

0.1

0.2

0.3

3 4 5 6 7 8<k>

Population

#Nodes

networkFrancePortugalSpain

Page 53: Decentralized routing in social networks

Let’s route! Intracity

2

1

10

100

1000

10000

1000 10000 1e+05 1e+06 1e+07# Nodes

1−

E dfs−c

om−d

eg

1−

E dfs

network●

franceportugalspain

levelmunicipalityprovince

Figure 2. Relation between our proposal and regular DFS across 468 urbannetworks

B. Generalizing results

An identical procedure has been performed for the top 100cities in each country, and additionally in all the provinces2.The reason for including provinces as well that is in bigcities usually urban areas go beyond the municipality in all3 countries. In all these 468 urban networks, previous resultsare consistent: routing-dfs-com-deg is still the better optionwith error rate 0.41 versus 0.51 in the case of includinggeographical information.

From now on, we will refer only to the routing-dfs-com-

deg. For these algorithm, results in error rate are stronglycorrelated with average length of arrived messages (0.92, seesuplementary information II-A), so we will focus on the errorrate, given the information provided by the average path lengthturns out to be redundant.

We have identified two factors with influence in the errorrate: the number of nodes size and the average degree. Thenumber of nodes positive correlated with the logarithm ofnetwork size, and the average degree dramatically increasesthe error rate if the average degree goes below 4 (see Suple-mentary Information II-B).

Since both efects overlap, we will measure the perfomanceof the algorithm compared to the perfomance of routing-dfs

which is the best one can do if no additional informationabout neighbouring nodes is provided. Figure 2 presents thevariation of the performance with network size. An interestingconclusion can be extracted: even if absolute error rate valuestend to be higher with network size, they grow slow O(log V )

while a DFS performs O(V ). This means is actually in bigurban networks where there is a bigger advantage on usingadditional information (in this case, communities and degree)to guide the network search.

2By province we will denote provincias in Spain and Portugal, and depart-

ments in France. Numbers for these divisions are 20, 52 and 96 respectively.Regarding geographical scope, the biggest distance is 120 km.

II. SUPLEMENTARY INFORMATION

A. Correlation < l > and Edfs�com�deg

0.0

0.2

0.4

0.6

0.8

● ●● ●

● ●● ●● ● ●●

●● ●●● ●●●● ●

● ● ●● ●● ●●● ●● ●● ●● ● ●●●●●●

● ●● ●●

●●● ●● ●● ●●

● ● ●● ●● ●● ● ●● ●●

●● ●● ● ●● ● ● ●●● ● ●●

●●● ● ●●● ● ●●● ●● ●● ●●● ●● ●● ●● ●● ●●● ●●

● ● ●●● ●● ● ●●● ●● ● ●●● ●●● ●● ● ● ●● ●● ●●● ●● ● ●●●●● ●● ●●● ●● ● ●

● ●●● ●● ●● ●●● ●● ●●● ● ●●● ●● ●● ●● ●●●

● ● ●● ●● ●● ●●●●● ●●● ●● ●●●● ● ●● ●●● ● ●●●● ●● ●●●● ● ● ●●

●● ●●● ●●●●● ●● ●● ●● ●

● ●●●

● ●●●● ●● ● ●● ●

●● ● ● ●● ● ●

●●● ●

● ● ● ●●

● ●

●●●

●●● ●

20 30 40 50avg

err

network●

franceportugalspain

#Nodes●

2e+054e+056e+058e+05

level● municipality

province

B. Parameters influencing Edfs�com�deg

Model R2

Edfs�com�deg

= ↵1 + ↵2V 0.18E

dfs�com�deg

= ↵1 + ↵2 log V 0.39E

dfs�com�deg

= ↵1 + ↵2 log V � ↵3hki 0.70

0.0

0.2

0.4

0.6

0.8

●●●●

●●● ●● ●● ●●● ● ●● ●●● ●●● ● ●● ● ● ●● ● ●●● ● ●●●● ●●●

●●● ●●●● ●●● ● ●●● ● ●●● ● ●●●●● ● ●●●● ● ● ●● ● ●● ●● ● ●●● ●● ● ●●●●●● ● ● ●● ● ●● ●●● ● ● ● ● ●●●●● ● ● ●●● ● ●●●● ● ●●●● ●●● ● ●●●● ●●● ●● ●●● ●●● ● ●●● ●●● ● ●● ●●● ●● ● ●●● ●

●●●● ● ●●● ●●●● ● ●● ● ●●● ●●● ● ●●●● ●●● ●● ● ●● ●●● ● ● ●●● ● ●● ● ●●● ●● ●● ● ● ●●● ●●● ●●● ●● ●●●● ●●● ● ●● ●●● ●● ●● ● ●●●●● ●● ●

●● ● ● ● ● ●● ●● ● ●● ●● ●●● ●●● ●●

●●● ●● ●●

●●● ●

●●● ●

1000 10000 1e+05 1e+06#Nodes

E dfs−c

om−d

eg

network●

franceportugalspain

level● municipality

province

0.0

0.2

0.4

0.6

0.8

●● ●●

●● ● ●● ●●●● ● ● ●● ●● ● ●●● ● ●●● ● ●● ● ●●● ● ●●●● ●● ●

●● ●● ●●● ●●● ● ●●● ● ● ●● ●● ●●●● ● ●●●● ●●●● ● ●●●● ● ●●● ●● ● ●● ●●● ●●● ●● ● ●● ●● ●●● ● ● ●●● ●● ●● ●●● ●●●●● ● ●●● ● ●● ●● ●●●●● ●● ●● ●● ●●● ●● ●● ● ●●● ● ●● ●●● ● ●● ● ●●●

●●●● ● ●●● ●●●●●● ● ●● ●● ●●●● ●●●● ●●● ●● ● ●● ●●● ● ● ●●● ● ●●●●●● ●● ●●●● ●●● ●●● ●●● ●● ●●●● ●●●● ●●●● ● ●●● ●● ●●

●●● ●● ●●● ● ● ●● ●● ●●● ●

●●● ● ●● ●● ● ●●●●●●

● ●●

●● ●●

●●● ●

4 6 8<k>

E dfs−com

−deg

network●

franceportugalspain

level● municipality

province

Page 54: Decentralized routing in social networks

Let’s route! Intracity

1

10

100

1000

10000

1000 10000 1e+05 1e+06 1e+07# Nodes

1−E d

fs−co

m−de

g

1−E d

fs

network●

franceportugalspain

levelmunicipalityprovince

0.92