Tomography with Available Bandwidth with Available Bandwidth - - PowerPoint PPT Presentation

1 1

Tomography Tomography with Available Bandwidth with Available Bandwidth

Alok Alok Shriram Shriram Jasleen Jasleen Kaur Kaur

Department of Computer Science Department of Computer Science University of North Carolina at Chapel Hill University of North Carolina at Chapel Hill The The UNIVERSITY UNIVERSITY of

• f NORTH CAROLINA

NORTH CAROLINA at at CHAPEL HILL CHAPEL HILL

http:// http://www.cs.unc.edu/~jasleen/research www.cs.unc.edu/~jasleen/research/ /

2 2

Network Network tomography tomography

! ! Idea: Use end

Idea: Use end-

• to

to-

• end probes to estimate state of internal links

end probes to estimate state of internal links

− − Send simultaneous probes to destinations that share

Send simultaneous probes to destinations that share portions of their path portions of their path

− − Study correlations in end

Study correlations in end-

• to

to-

• end metrics

end metrics loss lossXY

XY = 1

= 1 – – (1 (1 -

• loss

lossXR

XR)(1

)(1 -

• loss

lossRY

RY)

) loss lossXZ

XZ = 1

= 1 – – (1 (1 -

• loss

lossXR

XR)(1

)(1 -

• loss

lossRZ

RZ)

) If If loss lossXY

XY =

= loss lossXZ

XZ = L,

= L, then then loss lossRY

RY =

= loss lossRZ

RZ = 0,

= 0, and and loss lossXR

XR = L

= L

! ! Past

Past tomographic tomographic work done mostly with work done mostly with delay delay and and loss loss Focus: Available Bandwidth Focus: Available Bandwidth Focus: Available Bandwidth

X R Z Y

3 3

Extending Extending tomography tomography to A.B. to A.B.

AB ABXY

XY = min(AB

= min(ABXR

XR, AB

, ABRY

RY)

) AB ABXZ

XZ = min(AB

= min(ABXR

XR, AB

, ABRZ

RZ)

) ⇒ ⇒ AB ABXR

XR ≥

≥ max( max(AB ABXY

XY, AB

, ABXZ

XZ)

) AB ABRY

RY ≥

≥ AB ABXY

XY

AB ABRZ

RZ ≥

≥ AB ABXZ

XZ

X R Z Y

Tomography with several sources and destinations may help identify multiple bottlenecks on end-to-end paths Tomography Tomography with several sources and destinations may help with several sources and destinations may help identify identify multiple multiple bottlenecks on end bottlenecks on end-

• to

to-

• end paths

end paths

4 4

! ! R R ! ! Rule 2 Rule 2: For every pair of 2 paths with : For every pair of 2 paths with equal equal end end-

• to

to-

• end A.B.

end A.B. − − Non Non-

• shared links are non

shared links are non-

• bottlenecks

bottlenecks Could lead to false negatives Could lead to false negatives

Identifying bottleneck links Identifying bottleneck links

Bottleneck identification rules: Bottleneck identification rules: ! ! Rule 1 Rule 1: For each path, : For each path, − − Links with the least A.B. are potential bottlenecks Links with the least A.B. are potential bottlenecks Could lead to false positives Could lead to false positives

W X R Z Y Bottleneck

5 5

Challenge 1: probing tool inconsistency Challenge 1: probing tool inconsistency

Inconsistency of the probing tool Inconsistency of the probing tool

W X R Z Y Bottleneck

Tool inconsistency limits the ability to distinguish between bottleneck links Tool inconsistency limits the ability to Tool inconsistency limits the ability to distinguish between bottleneck links distinguish between bottleneck links

6 6

Challenge 2: probe scheduling Challenge 2: probe scheduling

! ! Requirements:

Requirements:

− − Paths that share links should

Paths that share links should not not be probed concurrently be probed concurrently This scheduling problem is NP-hard! This scheduling problem is NP This scheduling problem is NP-

• hard!

hard!

X R Z Y

! ! Solution:

Solution:

− − Schedule link

Schedule link-

• sharing probes in separate steps

sharing probes in separate steps

− − Minimize the total number of steps used

Minimize the total number of steps used

− − Paths that share links should be probed concurrently

Paths that share links should be probed concurrently

7 7

Challenge 3: limit on topology Challenge 3: limit on topology

Available bandwidth dynamics Available bandwidth dynamics

Tool run-time limits the number of participating end-nodes Tool run Tool run-

• time limits the

time limits the number of participating end number of participating end-

• nodes

nodes

8 8

Hurdles in identifying bottleneck links Hurdles in identifying bottleneck links

! ! Tool inaccuracy

Tool inaccuracy − − Limits the accuracy of detecting bottlenecks

Limits the accuracy of detecting bottlenecks

! ! Tool run

Tool run-

• time and dynamics of available bandwidth

time and dynamics of available bandwidth − − Limit the number of participating end

Limit the number of participating end-

• nodes

nodes

− − Limit the reduction in false positives

Limit the reduction in false positives

PlanetLab measurements with Pathload: ~ 4 end-nodes PlanetLab PlanetLab measurements with measurements with Pathload Pathload: ~ 4 end : ~ 4 end-

• nodes

nodes

9 9

PlanetLab PlanetLab tomography tomography results results

! ! Results from 4 sets of 4

Results from 4 sets of 4-

• node

node PlanetLab PlanetLab topologies used topologies used

500 1000 1500 2000 2500 3000 3500 1 2 3 4 5 Rule 1 & 2 Rule 1

Number of bottlenecks per path Number of bottlenecks per path

500 1000 1500 2000 2500 1 2 3 4 5 6 7 Rule 1 & 2 Rule 1

Distance of bottlenecks from source Distance of bottlenecks from source

− − At least 1 bottleneck discovered on half the paths

At least 1 bottleneck discovered on half the paths

− − No more than 3 bottlenecks listed for most paths

No more than 3 bottlenecks listed for most paths

− − Most bottlenecks lie at 2

Most bottlenecks lie at 2-

• 3 hops from the source

3 hops from the source

10 10

Wish Wish-

• list for a probing tool

list for a probing tool

! ! High accuracy and consistency

High accuracy and consistency − − Within 1 Mbps?

Within 1 Mbps?

! ! High speed

High speed − − Within 1 sec?

Within 1 sec?

! ! Non

Non-

• interference

interference − − With cross

With cross-

• traffic

traffic

− − With concurrent probing tools?

With concurrent probing tools?