End to End Routing Behavior in the Internet

1
End to End Routing Behavior in the Internet

• Vern Paxson
• Network Research Group
• Lawrence Berkeley National Laboratory
• University of California, Berkeley

Presented by Shreeram Sahasrabudhe (sas4)
2
Goal

• Trying to answer questions like
• What sort of pathologies and failures occur in
  the Internet routing?
• Do routes remain Stable over time or change
  frequently?
• Do routes from A and B tend to be symmetric as
  routes from B to A?
• Framework
• 40000 end-to-end route measurements using
  repeated
• traceroutes between 37 Internet sites in 1994
  1995

3
Routing in the Internet

• Internet
• Autonomous Systems (AS) collection of routers
  and hosts running a single Interior Gateway
  Protocol (IGP)
• Administrative Domain Routers and hosts are
  under a single administrative authority and a set
  of IGPs.
• Border Gateway Protocol (BGP) allows arbitrary
  interconnection topologies between AS.

4
Methodology

• Apparatus
• Network Probe Daemon (NPD) run on a number of
  sites.
• These NPDs were contacted at exponentially
  distributed intervals by a control program
  npd_control, to measure route to another NPD
  site using traceroute.
• N2 scaling if N sites then it can measure O(N2)
  internet paths between sites.
• Two sets of measurements D1 (Dec 1994
  measurement time interval 1-2 days) and D2 (Dec
  1995 60 mean inter-measurement interval of 2
  hours, 40 mean interval of about 2.75 days,
  paired measurements A?B and B?A )
• Why Exponential Sampling?
• Corresponds to additive random sampling which
  samples all instantaneous signal values with
  equal probability.
• Measurement times form a poisson process. Hence
  can apply Wolffs PASTA principle.
• How much representative of the real Internet?
• Experiment routes cover more than half the AS by
  weight (prominence in internet routing).
• Significant cross-section of internet behavior.

5
Routing Pathologies

• Routing Loops
• Forwarding loop Information loop
• traceroute loop (0.13 in D1, 0.16 D2)
• Erroneous Routing
• Found 1 example CT?Israel?UK!
• Connectivity altered mid-stream
• 0.16 D1 and 0.44 D2
• Fluttering
• Rapidly oscillating routing
• Infrastructure Failures
• An inside router gives host unreachable message
• Unreachable due to many hops
• 6 from D2 measurements went beyond the
  traceroutes limit of 30
• Mean path length D1(15.6) and D2(16.2)

6
Outages

• Temporary outages
• Sequence of consecutive traceroute packets are
  lost
• 1 to 5 losses D1(44), D2(55)
• 6 or more D1(0.96), D2(2.2)
• For more than 6 losses (30 sec outage 5sec X
  6losses)
• Time-of-day patterns
• Temporary outage at least 30sec 8 at 3-4pm
• Infrastructure failure 9.3 at 3-4pm and 7.6
  at 6-7pm

7
Routing Stability

• Routing Prevalence
• Prevalence A likelihood of currently observed
  router r
• being observed again in future
• Internet paths are strongly dominated by a single
  route, but
• there is also a significant site-to-site
  variation

8

• Routing Persistence
• Persistence Given that we observed a router r at
  time t, how long before that route is likely to
  have changed.
• More difficult to measure
• Classification
• Rapid route alternation Observed measurements
  made lt 60 sec apart and lt 10min apart
• Medium scale route alternation Observed
  measurements lt 1 hour
• Large-scale route alternation lt 6 hrs
• Long Lived Routes Remaining (gt 6hours)
• What is the chance of observing a route for at
  least a week 90

9
Routing Symmetry

• Affects
• One way propagation estimates (RTT/2)
• Asymmetry complicates network measurement,
  trouble shooting, accounting and routers flow
  calculation
• Asymmetry complicates network troubleshooting
• Analysis of routing symmetry
• 49 of D2 measurements observed an asymmetric
  path between sites.
• 30 of the paired measurements observed different
  autonomous systems in the virtual paths two
  directions.

10
Conclusions

• The likelihood of encountering a major routing
  pathology more than doubled in a year from end of
  94 to 95.
• Internet paths are heavily dominated by a single
  prevalent route, but the persistence varies from
  seconds to days
• There is not typical internet path. Different
  sites or pairs of sites encounter very different
  routing characteristics.
• Possible improvements to the methodology could
  be
• NPDs could be given more variety of measurement
  requests.
• Use a more sophisticated tool than traceroute.