Seminar 37-310 / Summer Semester 2000

1

Internet Connectivity

• Seminar 37-310 / Summer Semester 2000

Christian A. Plattner, 25.04.2000
2
Internet Connectivity

• On Power-Law Relationships of the Internet
  Topology
• End-to-End Routing Behaviour in the Internet

3
Power-Laws of the Internet

• What does the Internet look like?
• Are there any properties that dont change in
  time?
• How will it look like a year from now?
• How can I generate Internet-like graphs for my
  simulations?

4
Benefits from Understanding the Topology of the
Internet

• Protocol design
• More accurate artificial models
• Estimates for topological parameters

5
Internet Topology

• Two different viewsrouter level vs. interdomain
  level
• Partitioning the network into ASs
• Intra- vs. Inter-AS routing

6
Router- vs. Interdomain-level
7
AS Example UUnet
8
Power-Laws of the Internet

• Power-Law 1 dv ? rvR
• Power-Law 2 fd ? dO
• Hop-plot exponent P(h) ? hH
• Edges E N/(2R 2) (1-1/NR1)
• Diameter ? (N2 / N 2E) 1/ H

9
Practical Uses of the Power-Laws

• Describing graphs
• Protocol performance
• Graph generation and selection
• Predictions and extrapolations

Year 1999 2000 2001 2002
Nodes 4389 5763 7137 8511
Edges 8256 12639 15301 18394
d 4.26 4.39 4.61 4.78
10
End-to-End Routing Behaviour in the Internet

• What sorts of pathologies and failures occur in
  the Internet?
• Do routes remain stable over time or change
  frequently?
• Do routes from A to B tend to be symmetric as
  routes from B to A?

11
But we know the protocols very well...

• Distinction between protocols and behaviour is
  important!
• Its not obvious how the backbone dynamics
  translate into the routing dynamics seen by an
  end user

12
Methodology

• Measurement of a large sample of Internet routes
• Measurements were done in two different periods
• Measuring instrument traceroute

13
The Design of the Experiment

• Network probe daemon
• Npd_control
• Key property N2 scaling

14
How representative can the Observations be ?

• About 30 Internet hosts participated
• 1995 6.6 million Internet hosts!
• About 1000 active ASs in 1995

15
The Experiment

• D1 November 8 through December 24, 1994
• D2 November 3 through December 21, 1995
• Main difference D2 consists of paired traceroutes

16
Routing Pathologies

• Routing loops
• Erroneous routing
• Connectivity altered mid-stream
• Fluttering
• Infrastructure failures
• Unreachable due to many hops
• Temporary outages

17
Summary of Routing Pathologies
Pathology Probability Trend
Persistent loops 0.13-0.16
Erroneous routing 0.004-0.004
Mid-stream change 0.16 ? 0.44 worse
Infrastructure failure 0.21 ? 0.48 worse
Outage gt 30 secs 0.96 ? 2.2 worse
Total pathologies 1.5 ? 3.3 worse
18
End-to-End Routing Stability

• Two definitions of stability
• Routing prevalence
• Routing persistence

19
Routing (A)symmetry

• One-way propagation time
• Anticipatory flow state
• Network troubleshooting
• Causes of routing asymmetries?

20
Summary

• Overwiew of the Internet topology
• Power-Laws of the Internet and their practical
  use
• End-to-end routing behaviour