1. Studying Transnational Routing Detours through Surveillance States
Annie Edmundson, Roya Ensafi, Nick Feamster, Jennifer Rexford
Princeton University
RIPE 73
October 24th-28th, 2016

2. Characterizing and Avoiding Routing Detours
Characterizing detours
Which countries are Internet paths to popular destinations currently traversing?
Does local traffic leave the country? To where?
Nick:
Maybe a different title… it’s direct, but a bit boring.
Still could use a different title.
Graphic is a little bit choppy/bitmapped.
Avoiding detours
Can end-users avoid certain countries to popular destinations?
Can end-users keep more local traffic local?

3. Current State of Surveillance
Talking points:
-current countries are doing surveillance – at different levels of intensity
-countries are reacting to other countries conducting surveillance – i.e. Brazil
-we focused on these 5 countries because…
Surveillance States
Reactions against Surveillance
Studied Countries

4. Characterizing and Avoiding Routing Detours
Characterizing detours
Which countries are Internet paths to popular destinations currently traversing?
Does local traffic leave the country? To where?
The most common destination and transit country among all five countries studied is the United States.
Nick:
Maybe a different title… it’s direct, but a bit boring.
Still could use a different title.
Graphic is a little bit choppy/bitmapped.
Avoiding detours
Can end-users avoid certain countries to popular destinations?
Can end-users keep more local traffic local?

5. Measurement Study: Experiment
2. Extract 3rd party domains
Domains & 3rd Party Domains
Alexa Top 100 Domains
1. Connect to VPNs and curl
VPNs
3. DNS queries
Talking points:
-walk through the study procedure using the animations
Weird to have the flow chart go left to right and then right to left. Maybe make two rows, each going left to right, if you need more space.
RIPE Atlas
4. Collect DNS responses
Traceroutes
Domain: IPs
5. Traceroute to all IPs
RIPE Atlas

6. Where are popular domains hosted?
77.4% of paths that start in Brazil terminate in the United States
Talking points:
-first we looked at where paths were ending
-explain what each fraction represents
-discuss the United States row – about half (or more) paths originating in each country end in the US

7. Which countries are on the path to popular domains?
84.4% of paths that start in Brazil have the United States on the path
Talking points:
-then we looked at the whole path (not just endpoints)
-explain what each fraction represents
-discuss US row
-discuss Great Britain row

8. Characterizing and Avoiding Routing Detours
Characterizing detours
Which countries are Internet paths to popular destinations currently traversing?
Does local traffic leave the country? To where?
Despite having large IXPs, Brazil and Netherlands paths often trombones to the United States.
Nick:
Maybe a different title… it’s direct, but a bit boring.
Still could use a different title.
Graphic is a little bit choppy/bitmapped.
Avoiding detours
Can end-users avoid certain countries to popular destinations?
Can end-users keep more local traffic local?

9. Netherlands: Where is local traffic going?
Talking point:
-we looked at local traffic for the studied countries – explain tromboning
-point out US and Great Britain – both surveillance states
-show map of Europe with shaded countries to show/explain tromboning is due to geographic closeness (except for US)

10. Brazil: Where is local traffic going?
Talking points:
-local traffic and tromboning paths for brazil
-point out US
-show news article title about growth of IXPs in Brazil (and there is *still* tromboning to the US)

11. Kenya: Where is local traffic going?
Talking points:
-local traffic/tromboning Kenyan traffic
-point out GB and US
-show fiber map and proposals on building IXPs (to explain/motivate tromboning traffic)

12. Characterizing Routing Detours: Summary
Routing detours often transit surveillance states – especially the United States
Local traffic doesn’t always stay local
Talking points:
-summarize results from previous slides
-raise new question about avoidance
Is it possible to avoid certain countries by tunneling traffic through relays?

13. Characterizing and Avoiding Routing Detours
Characterizing detours
Which countries are Internet paths to popular destinations currently traversing?
Does local traffic leave the country? To where?
Nick:
Maybe a different title… it’s direct, but a bit boring.
Still could use a different title.
Graphic is a little bit choppy/bitmapped.
Yes, but it’s more difficult to avoid the United States than it is to avoid any other country.
Avoiding detours
Can end-users avoid certain countries to popular destinations?
Can end-users keep more local traffic local?

14. Country Avoidance
Country Avoidance = fraction of paths that do not pass through Country X
Talking points:
-explain why we need a metric for avoidance
-explain intuition behind metric
-walk through example/animation to compute country avoidance value
1/3
2/3
Relay
Path without Relay
Path with Relay

15. Avoidance Study: Experiment
Client to Relay Path:
1. Connect to VPNs
2. Traceroute to relay IPs
Traceroutes
VPNs
Relays
Relay to Server Path:
Talking points:
-walk through study procedure
1. ssh to relays
2. Traceroute to all IPs
Domain: IPs
Traceroutes
Relays

16. Can clients avoid countries more often?
Yes – many countries are almost completely avoidable for the top 100 domains
Talking points:
-discuss high level findings
-point out United States (no relays)
-point out United States (with relays) and emphasize the increase in avoidance, but also emphasize that the avoidance with relays is lower for US than for other countries
-point out Mauritius and South Africa – explain that it is due to fiber
-point out South Africa avoidance remains same because client->relays paths all have ZA on them
-point out increase in avoidance for Mauritius

17. Characterizing and Avoiding Routing Detours
Characterizing detours
Which countries are Internet paths to popular destinations currently traversing?
Does local traffic leave the country? To where?
Nick:
Maybe a different title… it’s direct, but a bit boring.
Still could use a different title.
Graphic is a little bit choppy/bitmapped.
Tromboning Brazilian paths decreased from 13.2% to 9.7%.
Avoiding detours
Can end-users avoid certain countries to popular destinations?
Can end-users keep more local traffic local?

18. System: Routing Around Nation-States
Developed an overlay network that:
Provides country avoidance
Is usable
Is scalable

19. System: Routing Around Nation-States
Relays act as web proxies + conduct measurements
Oracle triggers RIPE Atlas probes to conduct measurements
Clients use PAC file to select appropriate relay for avoiding a country

20. Future Work Connectivity within a country
Relationship between IXPs and nation state routing
Country avoidance based on IPv4 vs. IPv6 connectivity
Talking points:
-discuss each bullet

21. Conclusion
Paths commonly traverse known surveillance states – 84% of paths from Brazil traverse the United States
Relays can help prevent routing detours, but some of the more prominent surveillance states are the least avoidable
Tromboning Brazilian paths decreased from 13.2% to 9.7% with relays.
Talking points:
-discuss each bullet
-point out full write up on webpage
Full write-up and more data at:
ransom.cs.princeton.edu