A Structural Approach to Latency Prediction

1
A Structural Approach to Latency Prediction

• IMC 2006
• Harsha V. Madhyastha Thomas Anderson Arvind
  Krishnamurthy
• University of Washington University of Washington
  University of Washington
• Neil Spring Arun Venkataramani
• University of Maryland University of
  Massachusetts Amherst
• Presenter Chen Chih-Ming_at_NSLAB

2
Outline

• State of Art

3
State of Art

• Landmark Approach
• IDMap
• King
• Dimensionality Reduction Approach
• GNP
• Vivaldi

4
Compare to Vivaldi

• Fail to Blackbox
• Bottleneck capacity
• Loss
• Jitter
• Detour

5

• Hypothesis
• Overlap path

6
Overview

• Build an atlas
• Traceroute one ip per routable prefix
• BGP information from Route Views

IP Prefix1
V
IP Prefix1
IP Prefix1
IP Prefix1
IP Prefix1
7

• Client integration
• Traceroute
• Probe multiple path by random ip

V2
8

• Path prediction
• Find intersection from altas
• Approximate actual path

9

• Latency estimation
• SIRTTSI/2
• ID(RTTVD-RTTVI)/2
• Lantency EX(SIiIiD)

10
Clustering Routers

• Cluster routers that belong to the same AS
• Alias resolution
• clustering interface to routers
• by Mercaor
• Intra-AS clustering
• Lenth of reverse path for each vantage

11
V1
D
S
12
Overhead

• Effort in generating the atlas
• Daily basis
• 4Kbps for each vantage

13
Evalution

• Build Internet atlas
• Traceroutes from 158 PlanetLab nodes to each BGP
  prefix
• Public traceroute
• Validation set of paths
• Traceroutes from 582 public traceroute servers to
  destinations in random prefixes
• Exclude validation set in altas

14
Conclusion

• Altas has real AS path
• Similarity metric from RSIM metric
• SPAnPP/PA?PP
• Prediction is policy-compiant
• Latencies is more accurate
• Detour prediction
• 43

15
(No Transcript)
16
(No Transcript)
17
(No Transcript)
18
(No Transcript)
19
Comment

• Structural model may be better
• Traceroute