Imperfect of

1
Imperfect of End-to-End Routing Behavior in the
Internet

• Yao Zhao

2
Bad Resolution of Traceroute

• In D1, mean interval of 1-2 days
• Each NDP make a traceroute measurement once every
  two hours
• In D2, 60 with mean interval of 2 hours and 40
  with mean interval of 2.75 days
• Underestimate the prevalence of network
  connectivity problem

3
How Representative Are the Observations?

• 37 hosts of known 50,000 hosts in NSFNET
• Do routers in the same AS have similar
  characteristics?
• Different roles of routers
• Different relationship to other ASes
• 8 of 1000 routers involved in the measurement

4
Not Smart Experiment Design

• Centralized design
• Underestimation of network problems
• Can be improved by using batch jobs
• Simple traceroute
• Do not work smartly, only do traceroute as
  requested
• Cannot repeat traceroute or adjust interval based
  on the current situation

5
Underestimate Routing Asymmetry

• Paxon studied asymmetry in the level of city and
  AS
• Why city?
• Why not routers?

6
Conclusions are Out of Date

• Paxons data were collected in 1994 and 1995
• Now it is year 2006
• Internet grows 20 times larger. More than 20,000
  ASes so far.
• Do the conclusion drawn in 1996 still hold today?
• De the routing problems disappear or become
  severer?

7
(No Transcript)