1. Peering at the Internet’s Frontier: A First Look at ISP Interconnectivity in Africa
Arpit Gupta, Matt Calder, Nick Feamster, Marshini Chetty, Enrico Calandro, Ethan Katz-Bassett
In this talk we will talk about why we observe degraded Internet performance in Africa and discuss few possible solutions

2. BISmark: A platform to study home networks
BISmark Gateway
ISP Network
Internet
Home
Network
Modem
Access link
Custom OpenWrt firmware
Netgear gateways – 650 MHz processor, 128 MB RAM
Active and passive measurements in and out of home network

3. 200+ gateways in 20+ countries (Dec 2014)
Deployment
200+ gateways in 20+ countries (Dec 2014)

4. Motivation: Peering Introduces Significant Latency
Marshini Chetty et al. ”Measuring Broadband Performance in South Africa” ACM Symposium on Computing for Development (DEV). Cape Town, South Africa. December 2013.

5. Better understanding of existing connectivity required
Why Study Africa?
High latency routing paths
Broadband access is expensive
Better understanding of existing connectivity required

6. Outline ISP interconnectivity within Africa
Reasons for poor interconnectivity
Possible solutions

7. Uncover ISP Interconnectivity
Construct view of Internet paths within Africa
Circuitous Routing Paths
Identify crps and quantify their prevalence

8. Circuitous Routing Paths
AS2
Europe/ US
Africa
AS1
AS3

9. How to Identify Circuitous Routing Paths?
AS2
Europe/ US
Africa
AS1
AS3
Use traceroute probes

10. Use BISmark Routers to Run Traceroutes
Represents performance behavior for end users
Deployed in 17 South African homes
Placed across 7 major regional ISPs

11. Destinations for Traceroute Probes
M-Lab Servers
Google Cache Nodes

12. Identifying Circuitous Routing Paths
M-Lab J’Burg
M-Lab Nairobi …
7, , 7.793, South Africa, AS16637
8, , 8.338, South Africa, Cape Town IXP
9, , , South Africa, AS36937
14, , , South Africa, AS2018 …
8, , , South Africa, AS16637
9, , , United Kingdom, London IXP (LINX)
10, , , South Africa, AS36944
14, , , Kenya, AS36914
Packets leaving the African Continent

13. Latency Inflation Inflated Latencies M-Lab J’Burg M-Lab Nairobi …
7, , 7.793, South Africa, AS16637
8, , 8.338, South Africa, Cape Town IXP
9, , , South Africa, AS36937
14, , , South Africa, AS2018 …
8, , , South Africa, AS16637
9, , , United Kingdom, London IXP (LINX)
10, , , South Africa, AS36944
14, , , Kenya, AS36914
Inflated Latencies

14. Outline ISP interconnectivity within Africa
Reasons for poor interconnectivity
Possible solutions

15. Reasons for Poor Interconnectivity
Regional IXPs fails to keep local traffic local
IXP Prevalence
Failure because either:
Local ISPs not present at regional IXPs, or
If present, IXP participants don’t peer with each other

16. Fewer Participants at Regional IXPs
ISPs prioritize connecting to European IXPs
Fewer incentives to connect at regional ones
KENET
South Africa
Kenya
Liquid Telecom
J’Burg IXP
Kenya IXP

17. Missing Peering Links at Regional IXPs
Most content not available locally
Less incentives to peer with local ISPs
South Africa
Kenya
KENET
MTN
MTN IS IS
J’Burg IXP
Kenya IXP

18. Outline ISP interconnectivity within Africa
Reasons for poor interconnectivity
Possible Solutions

19. Improving Latency Deploying additional local cache nodes
Adding more peering links at regional IXPs

20. Adding More Cache Nodes
Traceroute Probes between:
BISmark Routers (eyeball)
Google Cache Node in Uganda (content)
Cache Node hosted by MTN
Emulates scenario where content is placed in a nearby country
Latency Improvements
Inside Customer Cone
BISmark Routers
Addition of cache nodes provides limited benefits with missing peering links

21. Adding More Peering Links at Regional IXPs
Simulated peering between all the participants at:
J’burg IXP (JINX)
Kenya IXP (KIXP)
Emulates scenario where more ISPs connect and peer at regional IXPs
Latency improvement up to 250 ms
Δ ~ 250ms
Latency Improvement
Adding cache nodes improves latency performance and adding more peering links ensures that this benefit trickles down to more number of users
Additional peering links  Significant latency improvements

22. Cache Nodes with Additional Peering Links
Cache Nodes in Kenya (1-5)
Two scenarios:
Existing peering links
Additional peering links
Minimum latency between BISmark routers & cache nodes considered
No Further Latency Improvements
Additional peering links  Lesser number of cache nodes

23. Mismatch of Incentives
Add More Cache Nodes
Less Local Content
Poor Interconnectivity
Add More Peering Links

24. Users Not Getting What They Pay For
Marshini Chetty et al. ”Measuring Broadband Performance in South Africa” ACM Symposium on Computing for Development (DEV). Cape Town, South Africa. December 2013.

25. Mobile Broadband: Higher Throughput
Mobile vs fixed
MyBroadband

26. Mobile Latencies Can Be More Variable
BISmark

27. Summary Circuitous routing paths within Africa
Adding more peering links and local cache nodes improves latency performances upto ~250 ms
Better incentive mechanisms required

29. High Latency Paths within Africa
Geographical Distance
Latency Penalties:
Ratio of observed and speed of light propagation latencies
Darker blocks imply higher latency penalties
High Penalties for routes to M-Lab server in Nairobi
High Latency Penalties
M-Lab Servers

30. IXP Prevalence Quantifies presence of IXPs for routing paths
Similar to routing path prevalence
Lower IXP prevalence observed for circuitous routing paths

31. Role of Regional IXPs M-Lab J’Burg M-Lab Nairobi
High Regional IXP Prevalence
No Regional IXPs
M-Lab J’Burg
M-Lab Nairobi