Efficient Algorithms for Large-Scale Topology Discovery

1
Efficient Algorithms for Large-ScaleTopology
Discovery

• Benoit Donnet
• joint work with Philippe Raoult, Timur Friedman
  and Mark Crovella
• Sigmetrics 2005 Banff (Canada)

2
Context

• Network measurement
• Internet topology discovery using distributed
  traceroute monitors
• IP interface level
• Existing tools
• Skitter (CAIDA)
• TTM (RIPE NCC)
• AMP (NLANR)
• DIMES (Tel Aviv U.)

3
Scaling Problem

• More monitors means more load on
• network resources
• destinations
• Classical approaches either
• stay small (skitter, TTM, AMP)
• trace slowly (DIMES)
• Can we trace more efficiently?

4
Contributions

• Quantification of scaling problems
• Intra-monitor redundancy
• Inter-monitor redundancy
• Efficient cooperative topology discovery
  algorithm
• Doubletree

5
Intra-monitor Redundancy (1)
6
Intra-monitor Redundancy (2)
7
Inter-monitor Redundancy (1)
8
Inter-monitor Redundancy (2)
9
Doubletree Tree-like Structure of Routes

• Both redundancy (i.e. inter and intra) suggest
  two different probing schemes
• They are based on the tree-like structure of
  routes
• Intra-monitor
• monitor-rooted tree (first suggested by Govindan
  et al.)
• Inter-monitor
• destination-rooted tree

10
Doubletree Monitor-rooted Tree
11
Doubletree Destination-rooted Tree
12
Doubletree Reconciliation

• Backward and forward probing are opposite schemes
• How can we reconciliate them?
• Starts probing at some hop h
• First, performing forward probing from h
• Second, performing backward probing from h-1

13
Doubletree Stop Sets

• Not necessary to maintain the whole tree
  structure.
• Each monitor uses stop sets (interface, root)
• Local Stop Set B interface
• Backward probing
• Global Stop Set F (interface, destination)
• Forward probing
• Shared between monitors

14
Doubletree Results (1)Intra-Monitor
skitter
Doubletree
15
Doubletree Results (2)Inter-Monitor
skitter
Doubletree
16
Conclusion

• We point out redundancy in classical topology
  discovery approaches using two metrics
• Intra-monitor redundancy
• Inter-monitor redundancy
• Based on these metrics, we define the Doubletree
  algorithm
• Measurement load reduction up to 76
• Interface and link coverage above 90