A presentation by Collin Anderson given at the Internet Measurement session of APRICOT 2015 on Tuesday, 3rd March 2015

1. Internet Policy and Independent Network
Performance Measurement
APRICOT 2015
Collin Anderson

2. Overview of Measurement Lab

3. International Collaboration,
Modern Research
Interdisciplinary, Open, Reviewed

4. M-Lab’s Global Footprint
Servers on Every Continent,
Infrastructure Diversity

5. M-Lab’s Global Footprint
Sixteen Tests; Measurements
from Every Country

6. Throughput
Measurements
• M-Lab hosts two active
throughput measurements:
• Network Diagnostic Tool
(NDT)
• BISmark
• NDT is integrated with
numerous applications and
receives about 100,000 tests
per day.
• Nearly every country is well-
covered.

7. Network
Transparency
• Three active tests:
• Shaperprobe
• Glasnost
• Neubot
• Open Observatory of
Network Interference (OONI)
• Measure differential traffic
performance and traffic
classification regimes.
• Detect manipulation.

8. Path Information
Collect paris-traceroute
information for every
measurement.

9. How Does M-Lab Collect Its Data?

10. Measurement Clients
Performance measurement clients
for informing users of their network
capacity.

11. Software Integrations
Providing network information to
software to enable better user
experience.

12. Partnerships with
Regulatory Agencies
Monitoring the progress of
broadband deployment for policy-
making.

13. Consumer Measurement and Policy Case Study
Interconnection Disputes in the United States

14. The Regulator’s
Problem
• Utilization and performance
data is proprietary and hidden
from the public.
• User collected information is
often unreliable, incomparable,
methodologically inconsistent
and narrowly scoped.
• Longitudinal data is rare, so is
comparative measurement.
• Independent data is expensive,
collection efforts are
burdensome.

15. Network Diagnostic
Tool
• M-Lab’s most popular hosted
test, tens of thousands of
measurements daily.
• Simple test of bulk transfer
capacity.
• Multiple NDT implementations
available, across different
languages and for diversity of
purposes.
• All implementations share the
same methodology and are
inter compatible.

16. Measuring Performance Bulk Transfer Capacity
Interconnection

17. Transit ISP X Transit ISP Y
Access ISP A Access ISP B
Network
Interconnection
s
Transit ISP X Transit ISP Y
1
Methodology
Inferring the Source of Congestion

18. Access ISP A Access ISP B
Transit ISP X Transit ISP Y
Access ISP A Access ISP B
?
?
?
2
Methodology
Inferring the Source of Congestion

19. Transit ISP X Transit ISP Y
Access ISP A Access ISP B
work
ction
s
Transit ISP X Transit ISP Y
Access ISP A Access ISP B
✓
Transit ISP X Transit ISP Yransit ISP X Transit ISP Y
3
Methodology
Inferring the Source of Congestion

20. ccess ISP A Access ISP B Access ISP A Access ISP B
Transit ISP X Transit ISP Y
Access ISP A Access ISP B
✓
nsit ISP X Transit ISP Y
ess ISP A Access ISP B
4
Methodology
Inferring the Source of Congestion

21. Transit ISP X Transit ISP Y
Access ISP A Access ISP B
✓
Transit ISP X Transit ISP Y
Access ISP A Access ISP B
✓
Transit ISP X Transit ISP Y
Access ISP A Access ISP B
3 5
4
Methodology
Inferring the Source of Congestion

22. Sufficient Sample The United States Constitutes
the Largest Source of this Data

23. Inferring Sources of
Congestion in Practice
Using New York’s Comparison

24. Inferring Sources of
Congestion in Practice
Using New York’s Comparison

25. Inferring Sources of
Congestion in Practice
US Access ISPs and Cogent (2013-2014)

26. Inferring Sources of
Congestion in Practice
US Access ISPs and Cogent (2013-2014)

27. Internet Performance Varies Significantly
Throughout the Day and Across Interconnections

28. Diurnal Patterns Are
Instructive
Expectations of Normal Performance

29. Diurnal Patterns Are
Instructive
Expectations of Congested Performance

30. Diurnal Cycles In Practice Peak Hours, Peak Disruption

31. Not Limited to Throughput Latency Sensitive Applications Affected

32. Not Limited to One Transit
Provider
Level 3 and Verizon

33. Congestion is Continuing Congestion as of Q4 2014

34. Public Interest Data
• Interconnection Study results
were referred to in filings to
United States regulator agency
since the start of this year
alone.
• M-Lab data has been used by
public interest organizations,
private companies and
individual citizens representing
all sides of the debate.
• Cited in dozens of media
articles in interconnection.

35. Become Involved
• Performance information collected
from real users.
• Global perspective on performance.
• Comparative data across networks
and time.
• Full raw data and open
methodology.
• Little or no cost of entry.
• Proven platform and trusted data
used by international regulators.
Host an M-Lab Site

36. Collin Anderson
collin@measurementlab.net
Internet Policy and Independent Network
Performance Measurement