A Study on Traceroute Potentiality in Revealing the Internet - - PowerPoint PPT Presentation

A Study on Traceroute Potentiality in Revealing the Internet AS-level Topology

• V. Luconi
• A. Faggiani, E. Gregori, A. Improta, L. Lenzini, L. Sani

IFIP Networking 2014 Conference - Trondheim - June 3rd, 2014

Outline

• The Internet AS-level topology discovery
• Methodology
• Traceroute infrastructures
• Results
• Conclusion

The Internet AS-level topology discovery

Introduction

• Knowing the Internet topology is important

✔ Planning business strategies ✔ Designing routing protocols ✔ Modelling its growth

• The AS-level captures the economic nature
• f the inter-domain routing of the Internet
• Measurement methods

✔ Passive: BGP ✔ Active: Traceroute

Problem: incompleteness

• The current knowledge of the Internet AS-level

topology is highly incomplete

✔ Inferring properties from an incomplete topology

could lead to biases

• Most works focused on BGP measurement

infrastructures

✔ Monitors are not optimally placed ✔ Methodology for quantifying the effectiveness of a

measurement infrastructure [1]

What about traceroute infrastructures?

[1] Gregori et al. On the Incompleteness of the AS-level graph: a Novel Methodology for BGP Route Collector Placement. IMC'12

Traceroute

Traceroute discovers the IP path from a source to a

• destination. IP-to-AS

mapping has to be done to discover the corresponding AS path.

Methodology

Economic relationships and export policies

• Provider-to-customer (p2c)
• Peer-to-peer (p2p)
• Stub ASes: ASes that do not provide connectivity

to other ASes

• Internet core: All non stub ASes

p2c-distance metric (I)

• E from A: 1
• E from C: 2
• C from F: undefined

p2c-distance of AS X from AS Y Minumum number of consecutive p2c links that connect X to Y

Gregori et al. IMC'12.

p2c-distance metric (I)

• E from A: 1
• E from C: 2
• C from F: undefined

p2c-distance of AS X from AS Y Minumum number of consecutive p2c links that connect X to Y

Gregori et al. IMC'12.

p2c-distance metric (I)

• E from A: 1
• E from C: 2
• C from F: undefined

p2c-distance of AS X from AS Y Minumum number of consecutive p2c links that connect X to Y

Gregori et al. IMC'12.

p2c-distance metric (I)

• E from A: 1
• E from C: 2
• C from F: undefined

p2c-distance of AS X from AS Y Minumum number of consecutive p2c links that connect X to Y

Gregori et al. IMC'12.

p2c-distance metric (II)

A necessary (not sufficient) condition for a BGP or traceroute monitor to reveal the full connectivity of an AS X is that the p2c-distance

• f X from that monitor is defined

It would be desirable to place monitors in the lower layers of the Internet Key concept Only a customer in a p2c relationship is able to reveal the full connectivity of his provider

Traceroute infrastructures

Traceroute infrastructures

We considered five infrastructures able to perform large-scale traceroute campaigns, found to be active in October 2013

# Probing ASes # Non stubs # Stubs Aqualab Dasu/Ono 2,442 1,398 (57.25%) 1,044 (42.75%) CAIDA Ark 76 60 (78.95%) 16 (21.05%) DIMES 251 145 (57.77%) 106 (42.23%) Portolan 360 246 (68.33%) 114 (31.67%) RIPE NCC Atlas 2,135 1,310 (61.36%) 825 (38.64%)

Traceroute infrastructures

We considered five infrastructures able to perform large-scale traceroute campaigns, found to be active in October 2013

# Probing ASes # Non stubs # Stubs Aqualab Dasu/Ono 2,442 1,398 (57.25%) 1,044 (42.75%) CAIDA Ark 76 60 (78.95%) 16 (21.05%) DIMES 251 145 (57.77%) 106 (42.23%) Portolan 360 246 (68.33%) 114 (31.67%) RIPE NCC Atlas 2,135 1,310 (61.36%) 825 (38.64%)

Large number of Probing ASes

Traceroute infrastructures

We considered five infrastructures able to perform large-scale traceroute campaigns, found to be active in October 2013

# Probing ASes # Non stubs # Stubs Aqualab Dasu/Ono 2,442 1,398 (57.25%) 1,044 (42.75%) CAIDA Ark 76 60 (78.95%) 16 (21.05%) DIMES 251 145 (57.77%) 106 (42.23%) Portolan 360 246 (68.33%) 114 (31.67%) RIPE NCC Atlas 2,135 1,310 (61.36%) 825 (38.64%)

Pervasiveness in the lowest layers of the Internet

Overlap

• Low overlapping between different

infrastructures

• Every infrastructure introduces new points of

view

Ark Atlas Dasu/Ono DIMES Portolan Ark

• 0.577

0.310 0.239 0.155 Atlas 0.019

• 0.296

0.049 0.088 Dasu/Ono 0.009 0.259

• 0.045

0.100 DIMES 0.068 0.414 0.438

• 0.243

Portolan 0.031 0.522 0.675 0.169

• Overlapcoefficient O( A, B)=∣A∩B∣

∣A∣

Overlap

• Low overlapping between different

infrastructures

• Every infrastructure introduces new points of

view

Ark Atlas Dasu/Ono DIMES Portolan Ark

• 0.577

0.310 0.239 0.155 Atlas 0.019

• 0.296

0.049 0.088 Dasu/Ono 0.009 0.259

• 0.045

0.100 DIMES 0.068 0.414 0.438

• 0.243

Portolan 0.031 0.522 0.675 0.169

• Overlapcoefficient O( A, B)=∣A∩B∣

∣A∣

1.9% of Atlas probing ASes are also found in Ark

A B

Overlap

• Low overlapping between different

infrastructures

• Every infrastructure introduces new points of

view

Ark Atlas Dasu/Ono DIMES Portolan Ark

• 0.577

0.310 0.239 0.155 Atlas 0.019

• 0.296

0.049 0.088 Dasu/Ono 0.009 0.259

• 0.045

0.100 DIMES 0.068 0.414 0.438

• 0.243

Portolan 0.031 0.522 0.675 0.169

• Overlapcoefficient O( A, B)=∣A∩B∣

∣A∣

57.7% of Ark probing ASes are also found in Atlas

A B

Overlap

• Low overlapping between different

infrastructures

• Every infrastructure introduces new points of

view

Ark Atlas Dasu/Ono DIMES Portolan Ark

• 0.577

0.310 0.239 0.155 Atlas 0.019

• 0.296

0.049 0.088 Dasu/Ono 0.009 0.259

• 0.045

0.100 DIMES 0.068 0.414 0.438

• 0.243

Portolan 0.031 0.522 0.675 0.169

• Overlapcoefficient O( A, B)=∣A∩B∣

∣A∣

Results

A deeper insight (I)

Coverage of the Internet core (8,181 non stub ASes in October 2013) by each project d = 1 d = 2 d = 3

Ark

361 (4.41%) 789 (9.64%) 1,117 (13.69%)

Atlas

2,367 (28.93%) 2,820 (34.47%) 2,949 (36.05%)

Dasu/Ono

2,465 (30.13%) 2,867 (35.04%) 2,981 (36.44%)

DIMES

517 (6.32%) 967 (12.06%) 1,332 (16.28%)

Portolan

700 (8.56%) 1,158 (14.16%) 1,458 (17.82%)

A deeper insight (I)

Coverage of the Internet core (8,181 non stub ASes in October 2013) by each project d = 1 d = 2 d = 3

Ark

361 (4.41%) 789 (9.64%) 1,117 (13.69%)

Atlas

2,367 (28.93%) 2,820 (34.47%) 2,949 (36.05%)

Dasu/Ono

2,465 (30.13%) 2,867 (35.04%) 2,981 (36.44%)

DIMES

517 (6.32%) 967 (12.06%) 1,332 (16.28%)

Portolan

700 (8.56%) 1,158 (14.16%) 1,458 (17.82%)

Number of non stubs with p2c-distance less than or equal to d from at least

• ne measurement monitor

A deeper insight (I)

Coverage of the Internet core (8,181 non stub ASes in October 2013) by each project d = 1 d = 2 d = 3

Ark

361 (4.41%) 789 (9.64%) 1,117 (13.69%)

Atlas

2,367 (28.93%) 2,820 (34.47%) 2,949 (36.05%)

Dasu/Ono

2,465 (30.13%) 2,867 (35.04%) 2,981 (36.44%)

DIMES

517 (6.32%) 967 (12.06%) 1,332 (16.28%)

Portolan

700 (8.56%) 1,158 (14.16%) 1,458 (17.82%)

A deeper insight (I)

Coverage of the Internet core (8,181 non stub ASes in October 2013) by each project d = 1 d = 2 d = 3

Ark

361 (4.41%) 789 (9.64%) 1,117 (13.69%)

Atlas

2,367 (28.93%) 2,820 (34.47%) 2,949 (36.05%)

Dasu/Ono

2,465 (30.13%) 2,867 (35.04%) 2,981 (36.44%)

DIMES

517 (6.32%) 967 (12.06%) 1,332 (16.28%)

Portolan

700 (8.56%) 1,158 (14.16%) 1,458 (17.82%)

A deeper insight (II)

Several probing ASes share a common set of providers and provide only redundant information

d = 2

A deeper insight (II)

Several probing ASes share a common set of providers and provide only redundant information

d = 2

p2c-overlap coefficient Fraction of the non stubs covered by one probing AS at distance d that are also covered by the other probing ASes of the same infrastructure

A deeper insight (II)

Several probing ASes share a common set of providers and provide only redundant information

d = 2

BGP and Traceroute together

Even putting all together the full coverage is still far from being achieved

Scenarios # VPs d = 1 d = 2 d = 3

• I. BGP only

166 648 (7.92%) 1,068 (13.05%) 1,301 (15.90%)

• II. BGP with Ark,

DIMES and Portolan 729 1,288 (15.74%) 1,728 (21.12%) 1,923 (23.50%)

• III. BGP with all

traceroute infrastructures 4,222 2,465 (44.22%) 2,867 (47.82%) 2,981 (48.48%)

BGP and Traceroute together

Even putting all together the full coverage is still far from being achieved

Scenarios # VPs d = 1 d = 2 d = 3

• I. BGP only

166 648 (7.92%) 1,068 (13.05%) 1,301 (15.90%)

• II. BGP with Ark,

DIMES and Portolan 729 1,288 (15.74%) 1,728 (21.12%) 1,923 (23.50%)

• III. BGP with all

traceroute infrastructures 4,222 2,465 (44.22%) 2,867 (47.82%) 2,981 (48.48%)

BGP and Traceroute together

Even putting all together the full coverage is still far from being achieved

Scenarios # VPs d = 1 d = 2 d = 3

• I. BGP only

166 648 (7.92%) 1,068 (13.05%) 1,301 (15.90%)

• II. BGP with Ark,

DIMES and Portolan 729 1,288 (15.74%) 1,728 (21.12%) 1,923 (23.50%)

• III. BGP with all

traceroute infrastructures 4,222 2,465 (44.22%) 2,867 (47.82%) 2,981 (48.48%)

Redundancy again (I)

Optimization problem for computing the degree of redundancy

Scenarios # VPs d = 1 d = 2 d = 3

• I. BGP only

166 648 (7.92%) 1,068 (13.05%) 1,301 (15.90%)

• II. BGP with Ark,

DIMES and Portolan 729 1,288 (15.74%) 1,728 (21.12%) 1,923 (23.50%)

• III. BGP with all

traceroute infrastructures 4,222 2,465 (44.22%) 2,867 (47.82%) 2,981 (48.48%)

Redundancy again (II)

How many VPs should we add to obtain the full coverage of the Internet core?

Scenarios # VPs d = 1 d = 2 d = 3

• I. BGP only

166 +4,593 +4,136 +4,075

• II. BGP with Ark,

DIMES and Portolan 729 +4,444 +4,027 +3,978

• III. BGP with all

traceroute infrastructures 4,222 +3,435 +3,199 +3,177

Redundancy again (III)

In this case how many VPs would provide

• nly redundant information?

Scenarios # VPs d = 1 d = 2 d = 3

• I. BGP only

166 104 (63.65%) 130 (78.31%) 147 (82.12%)

• II. BGP with Ark,

DIMES and Portolan 729 518 (71.05%) 584 (80.11%) 603 (82.72%)

• III. BGP with all

traceroute infrastructures 4,222 3,002 (71.10%) 3,249 (76.95%) 3,295 (78.04%)

Redundancy again (IV)

If we had the chance to ideally place monitors to obtain the same coverage

Scenarios # VPs d = 1 # VPs d = 2 # VPs d = 3 II 119 75 49 III 1,042 729 611

Redundancy again (V)

If we had the chance to ideally place monitors to obtain the same coverage

Scenarios # VPs d = 1 # VPs d = 2 # VPs d = 3 II 119 75 49 III 1,042 729 611

Number of monitors needed at p2c-distance d

Redundancy again (VI)

If we had the chance to ideally place the same amount of monitors

Scenarios

# VPs d = 1 d = 2 d = 3 II 729 3,017 (36.88%) 3,913 (47.83%) 4,228 (51.68%) III 4,222 7,612 (93.05%) 8,181 (100%) 8.181 (100%)

Redundancy again (VI)

If we had the chance to ideally place the same amount of monitors

Scenarios

# VPs d = 1 d = 2 d = 3 II 729 3,017 (36.88%) 3,913 (47.83%) 4,228 (51.68%) III 4,222 7,612 (93.05%) 8,181 (100%) 8.181 (100%)

Full coverage of the Internet core

Conclusion

• At first sight traceroute infrastructures

seem very appealing in an AS-level topology discovery perspective

• However, an analysis with the p2c-distance

reveals that they do not reach over 36% of coverage when considered separately

• Many probing ASes provide only redundant

information

Conclusion

• Traceroute infrastructures could be useful to

enhance the AS-level topology discovered by BGP RCs

• However, even putting all together many

vantage points provide only redundant information

• Atlas and Dasu/Ono would be very helpful if

they performed intensive campaigns oriented to AS-level topology discovery

The end

Thank you for your attention! Questions?

valerio.luconi@iet.unipi.it http://portolan.iet.unipi.it

Backup slides

Geographical distribution

Pervasiveness is mantained at a geographical scope

Ark Atlas Dasu/Ono DIMES Portolan Africa North America Latin America International Europe Asia Pacific

Geographical distribution

d = 3

Geographical distribution

d = 3

North America pulls down the whole coverage