The Impact of Internet Policy and Topology on Delayed Routing convergence

1
The Impact of Internet Policy and Topology on
Delayed Routing convergence

• C. Labovitz, A. Ahuja, R. Wattenhofer and S.
  Venkatachary.
• IEEE INFOCOM 2001.

2
Roadmap

• The Problem
• The Contribution
• Motivation
• Background
• Testbed
• Experimental Results
• Conclusions
• Future Work

3
The Problem

• The internet lacks inter-domain path fail over.
• How soon can convergence be expected when path(s)
  fail.
• What is the effect of path failures on different
  ISPs, large and small.

4
Roadmap

• The Problem
• The Contribution
• Motivation
• Background
• Testbed
• Experimental Results
• Conclusions
• Future Work

5
The Contribution

• Demonstrate that the time for end-to-end internet
  convergence depends on the length of the length
  of the longest possible back-up AS path between
  source and destination node.
• Show clearly, larger ISPs have faster convergence
  latencies.

6
The Contribution..more specifically

• Time complexity of for internet fail-over
  convergence is upper bounded by 30n sec.
• Show that, routes from customers of larger ISPs
  exhibit faster convergence.
• Presence of vagabond paths, s/w errors.
• Most default free Internet routes exhibit
  multiple alternate secondary paths.

7
Roadmap

• The Problem
• The Contribution
• Motivation
• Background
• Testbed
• Experimental Results
• Conclusions
• Future Work

8
Motivation

• There is a relative lack of reliability in the
  internet vis-à-vis the PSTN network.
• Multi-homed internet sites often experience long
  connectivity latencies after faults(s) occur.

9
Roadmap

• The Problem
• The Contribution
• Motivation
• Background
• Testbed
• Experimental Results
• Conclusions
• Future Work

10
Background

• BGP, a path vector protocol, uses an ASPath
  sequence of intermediate ASs.
• Uses this ASPath for loop detection and policy
  decisions.
• Most vendor implementations of BGP fall back on
  the best path selection based on ASPath length
  after fault.

11
Background..contd.

• BGP includes a MinRouteAdver timer.
• This provides a buffer between BGP updates and
  also allows for aggregation of routes containing
  common information.
• T. Griffin and G. Wilfong, An analysis of BGP
  convergence properties, ACM SIGCOM, 1999.
  possible to implement mutuall unsatisfiable
  policies in BGP
• L. Gao and J. Rexford,Stable Internet Routing
  without Global Coordination, ACM SIGMETRICS,
  2000.adherence to common policies guarantees
  convergence

12
Roadmap

• The Problem
• The Contribution
• Motivation
• Background
• Testbed
• Experimental Results
• Conclusions
• Future Work

13
TestBed

• Unix, probe machineswith peering sessions with
  more than 20 ISPs (geographically and
  topologically diverse).
• Inject faults into more than 10 ISPs.
• Software from MRT IPMA projects, running on
  FreeBSD systems.
• Observation spread over 6 months.

14
Routing Policy
Most ISPs simply resort to trusting their peers.
! ! ! !
15
Roadmap

• The Problem
• The Contribution
• Motivation
• Background
• Testbed
• Experimental Results
• Conclusions
• Future Work

16
Experimental Results

• Effect of topology on Convergence
• Analyzed routing topologies between more than
  200 pair of ISPs.
• Focus on MinRouteAdver Timers and length of
  ASPaths on convergence.

17
Experimental Results
Routes from ISP1
18
Experimental Results
Routes from ISP2
19
Experimental Results
Routes from ISP3
20
Experimental Results
21
Experimental Results
22
Roadmap

• The Problem
• The Contribution
• Motivation
• Background
• Testbed
• Experimental Results
• Conclusions
• Future Work

23
Conclusions

• Shown conclusively that the internet currently
  lacks high levels of reliability and fault
  tolerance.
• Current internet convergence latencies, upto 15
  min. is very high.
• Complexity of multi homed internet path fail over
  scales with the longest possible back up path for
  that route.
• Customers sensitive to fail overs, should multi
  home to larger ISPs.
• There remains a chasm in formulating nechanisms
  to handle vagabond paths.

24
Roadmap

• The Problem
• The Contribution
• Motivation
• Background
• Testbed
• Experimental Results
• Conclusions
• Future Work

25
Future Work

• Evolve mechanism to handle vagabond paths.
• Implement necessary changes to BGP in order to
  reduce convergence latencies.

26
Thats all folks.
Comments anirban_at_cs.ucr.edu