Model-based Identification of Dominant Congested Links

1
Model-based Identification of Dominant Congested
Links

• Wei Wei, Bing Wang, Don Towsley, Jim Kurose
• weiwei, bing, towsley, kurose_at_cs.umass.edu

2
Outline

• Motivation
• Virtual probe, virtual queuing delay
• Dominant congested links
• Identifying dominant congested links
• Validation
• Conclusions, future work

3
Motivation

• Dominant congested link (informally) link with
  most losses and significant delays on end-end
  path
• Applications
• traffic engineering
• understand dynamics of network
• Direct measurement of an individual link
  difficult
• commercial reasons
• existence of multiple ISPs along path

4
Virtual Probe, Virtual Queuing Delay

• Virtual Probe infinitesimally small packet
• does not disturb real traffic, never dropped
• queuing delay due to queue occupancy
• If queue full, mark as lost, experience maximum
  queuing delay, go to next link
• Virtual Queuing Delay W
• End-end queuing delay of virtual probes with loss
  marks
• Two important questions about W
• Most loss marks at one link?
• Major part of W due to experiencing maximum
  queuing delay?

5
Virtual Probe, Virtual Queuing Delay cont.
6
Strongly Dominant Congested Link (SDCL)

• Link k is a strongly dominant congested link in
  t1,t2) iff for any virtual probe sent at any
  time t in t1,t2) satisfies,
• all losses occur only at link k
• If experience max queuing delay on link k, this
  max queuing delay is at least sum of queuing
  delays it experiences on other links

7
Weakly Dominant Congested Link (WDCL)

• Link k is a weakly dominant congested link with
  parameter ? and in t1, t2, iff a virtual
  probe sent at t satisfies

where 0 ? ? lt0.5, 0 ? ? 1,
8
SDCL Illustration
Qk
Qk
W
Qk maximum queuing delay W virtual queuing delay
9
Property of SDCL
Example
Hypothesis H0 A SDCL exists. Find D minwFW(w)
gt 0,Check FW(2D). If FW(2D) lt 1, reject.
Otherwise, accept.
10
Property of WDCL
Example
Hypothesis H0 A WDCL exists. Find D minwFW(w)
gt ?,Check FW(2D). If FW(2D) lt (1- ?)(1-f),
reject. Otherwise, accept.
11
An Example Test of SDCL
gt

• H0 rejected

12
Inferring Virtual Queuing Delay Distribution FW(w)

• Use virtual queuing delay distribution to test if
  DCL exist
• Infer FW(w)
• Linear Interpolation
• Hidden Markov model
• Markov model with a hidden dimension

13
Markov Model with a Hidden Dimension

• Model components
• State (Xt, Yt), Yt delay, Xt hidden state
• N of hidden states
• M of delay bins
• p(i,j) initial distribution
• P(i,j)(k,l) transition matrix
• s(j) P(lossdelay j)
• When N1, a Markov model

14
Packet Probes and Model Inference

• One-way End-end Periodic probes
• Delay Yt, t1, 2, , T.
• Yt if probe t is lost
• Parameter inference algorithm
• Forward-backward inference
• Iterative approach
• After algorithm converges
• s(j)P(lossdelayj), j1,2, , M.

15
Obtain Virtual Queuing Delay Distribution FW(w)
from s(w)

• Obtain virtual queuing delay distribution from
  model and trace

16
Evaluation

• Ns simulation
• Controlled environment
• Global knowledge
• Validation of methodology
• Internet experiment
• Applying methodology in real world
• Probe duration needed to obtain correct
  identification

17
Simulation Setup
p1
p2
p3
18
Validation via Simulation

• (p1,p2,p3) (0, .002, .038)

WDCL(.07, .1)?
D4 FW(8) 1 gt (1-.07)(1-.1) YES
19
Internet Experiments
Residence House USC Loss prob.
0.04 WDCL(.1,.1)?
D1, FW(2D)lt(1-.1)(1-.1) No
20
Conclusions and Future Work

• Existence of DCL
• Introduce virtual queuing delay
• Model-based approach from one-way end-end
  measurement
• Only minutes of probes needed
• Future work
• Controlled test-bed experiments and more/richer
  Internet experiments
• Scenarios where wireless network is present

21

• Thank you!