[PPT] - What do Packet Dispersion Techniques Measure? Constantinos Dovrolis, PowerPoint Presentation

SLIDE 1

What do Packet Dispersion Techniques Measure?

Constantinos Dovrolis, Univ-Delaware Parmesh Ramanathan, Univ-Wisconsin David Moore, CAIDA

Constantinos Dovrolis - IEEE Infocom 2001 - April 24-26, 2001

1 of 28

SLIDE 2

Overview

Background: capacity and available bandwidth Dispersion of packet-pairs Dispersion of packet-trains A capacity estimation methodology: pathrate

Constantinos Dovrolis - IEEE Infocom 2001 - April 24-26, 2001

2 of 28

SLIDE 3

Part I

Background

Constantinos Dovrolis - IEEE Infocom 2001 - April 24-26, 2001

3 of 28

SLIDE 4

Definition of capacity

Maximum IP-layer throughput that a flow can get, without any cross traffic

A C Source Sink Link-1 Link-2 Link-3

C

i: capacity of link i (i

H)

Path capacity C is limited by narrow link n: C

min

i H fC i g

C

n

Constantinos Dovrolis - IEEE Infocom 2001 - April 24-26, 2001

4 of 28

SLIDE 5

Definition of available bandwidth

Maximum IP-layer throughput that a flow can get, given (stationary) cross traffic

A C Source Sink Link-1 Link-2 Link-3

u

i: utilization of link i Available bandwidth A limited by tight link t: A

min

i H C i

u

i

C

t

u

t

Constantinos Dovrolis - IEEE Infocom 2001 - April 24-26, 2001

5 of 28

SLIDE 6

Part II

Packet-pair dispersion

Constantinos Dovrolis - IEEE Infocom 2001 - April 24-26, 2001

6 of 28

SLIDE 7

Packet-pair: basic idea

Packet transmission time:

L

C Send two packets back-to-back Measure dispersion at receiver Estimate C as L

C

3C =L/C ∆ L/C L/3C

But.. cross traffic ‘noise’ can affect the packet dispersion

Constantinos Dovrolis - IEEE Infocom 2001 - April 24-26, 2001

7 of 28

SLIDE 8

Previous works on packet-pair dispersion

Largely considered cross traffic effects as ‘random noise’ Carter and Crovella (Infocom 1997), Lai and Baker (Infocom 1999): Statistical

techniques to extract most common measurement range (mode)

Paxson (Sigcomm 1997): observed multimodalities but did not relate them with

cross traffic

They suggest use of maximum-size packet-pair packets

Constantinos Dovrolis - IEEE Infocom 2001 - April 24-26, 2001

8 of 28

SLIDE 9

Multimodality of packet-pair estimates

Cross-traffic causes local modes below (SCDR) and above (PNCM) capacity

mode (CM)

10 20 30 40 50 60 70 80

Bandwidth (Mbps)

40 80 120 160 200 240 280 320 360 400

# of measurements

P={100,75,55,40,60,80}, L=Lc=1500B

u=20% Capacity Mode (CM) Post−Narrow Capacity Mode Sub−Capacity Dispersion Range (SCDR) (PNCM)

10 20 30 40 50 60 70 80

Bandwidth (Mbps)

20 40 60 80 100 120 140 160

# of measurements

P={100,75,55,40,60,80}, L=Lc=1500B

SCDR PNCM CM u=80%

Heavier cross traffic load makes CM weaker

Constantinos Dovrolis - IEEE Infocom 2001 - April 24-26, 2001

9 of 28

SLIDE 10

Creation of SCDR and PNCM modes

SCDR is caused by cross traffic interfering with packet-pair

i

L/40 L/40

t

L/60 L/60 L/60 CT 2 1

C =60Mbps

i-1

C =40Mbps t

1 2 L/40 + (2 L/60 - L/40) = L/30

PNCMs are caused by back-to-back packet-pairs after narrow link (first packet

is adequately delayed)

Constantinos Dovrolis - IEEE Infocom 2001 - April 24-26, 2001

10 of 28

SLIDE 11

Effect of cross traffic packet size

Distinct cross traffic packet sizes cause SCDR local modes Common Internet traffic packet sizes: 40B, 550B, 1500B

10 20 30 40 50 60 70 80

Bandwidth (Mbps)

20 40 60 80 100 120 140 160

# of measurements

P={100,75,55,40,60,80}, u=50%, L=1500B

Fixed CT packet size: Lc=1500B SCDR CM PNCMs

10 20 30 40 50 60 70 80

Bandwidth (Mbps)

10 20 30 40 50 60 70 80 90 100 110 120

# of measurements

P={100,75,55,40,60,80}, u=50%, L=770B

SCDR Lc uniform in [40,1500]B Variable CT packet size: CM PNCM

Constantinos Dovrolis - IEEE Infocom 2001 - April 24-26, 2001

11 of 28

SLIDE 12

Effect of packet-pair size

Previous work suggests use of maximum-sized packets But.. this is not optimal for uncovering capacity mode

10 20 30 40 50 60 70 80

Bandwidth (Mbps)

20 40 60 80 100 120 140 160 180 200 220

# of measurements

P={100,75,55,40,60,80}, u=50%

L=100B SCDR Lc : uniform in [40,1500]B CM PNCM

10 20 30 40 50 60 70 80

Bandwidth (Mbps)

10 20 30 40 50 60 70 80 90 100 110 120

# of measurements

P={100,75,55,40,60,80}, u=50%

L=1500B SCDR Lc : uniform in [40,1500]B CM

Constantinos Dovrolis - IEEE Infocom 2001 - April 24-26, 2001

12 of 28

SLIDE 13

Packet-pair dispersion: summary

Packet-pair technique: simple in unloaded paths Multimodal bandwidth distribution in loaded paths Most common measurement range (mode) is not always the capacity Capacity is normally a local mode (CM) Interfering cross traffic packets cause local modes or SCDR Loaded post-narrow links also cause local modes (PNCMs) Maximum packet size is not optimal for uncovering CM How can we tell which local mode is related to the capacity?

Constantinos Dovrolis - IEEE Infocom 2001 - April 24-26, 2001

13 of 28

SLIDE 14

Part III

Packet-train dispersion

Constantinos Dovrolis - IEEE Infocom 2001 - April 24-26, 2001

14 of 28

SLIDE 15

Packet-train dispersion

What do we measure with the dispersion of packet trains?

3 N 2 1 N 2 3 1 1 2 3 N ∆ (N) (N) (N) ∆ CT CT ∆ CT

R

R S C C C

1 2 3 2

Bandwidth estimate: N L N

What is the effect of length

N on bandwidth estimate? Carter & Crovella: packet-train dispersion estimates A

Constantinos Dovrolis - IEEE Infocom 2001 - April 24-26, 2001

15 of 28

SLIDE 16

Packet-train experiments

What happens as we increase the packet-train length N?

10 20 30 40 50 60 70 80

Bandwidth (Mbps)

20 40 60 80 100 120 140

# of measurements

P={100,75,55,40,60,80}, u=80%, L=Lc=1500B

N=2 CM

10 20 30 40 50 60 70 80

Bandwidth (Mbps)

20 40 60 80 100 120 140 160 180 200

# of measurements

P={100,75,55,40,60,80}, u=80%, L=Lc=1500B

CM N=3

Constantinos Dovrolis - IEEE Infocom 2001 - April 24-26, 2001

16 of 28

SLIDE 17

Packet-train experiments (cont’)

Range of measurements decreases and becomes unimodal Measurements tend to Asymptotic Dispersion Rate (ADR) (less than C)

10 20 30 40 50 60 70 80

Bandwidth (Mbps)

40 80 120 160 200 240

# of measurements

P={100,75,55,40,60,80}, u=80%, L=Lc=1500B

N=5

10 20 30 40 50 60 70 80

Bandwidth (Mbps)

40 80 120 160 200 240 280 320 360 400

# of measurements

P={100,75,55,40,60,80}, u=80%, L=Lc=1500B

N=10 Asymptotic Dispersion Rate R=15Mbps

Constantinos Dovrolis - IEEE Infocom 2001 - April 24-26, 2001

17 of 28

SLIDE 18

ADR in single-hop paths

For sufficiently large N: R

N
L
C
u
C
C
C
ADR is not the capacity

C ADR is not the available bandwidth A For single-hop paths, we can estimate A from R A

C
C
R
Constantinos Dovrolis - IEEE Infocom 2001 - April 24-26, 2001

18 of 28

SLIDE 19

Effect of cross traffic routing

Path-persistent and one-hop persistent cross traffic

1

C C C C Source

2 3 H

C Path Path

. . .

Sink Sources Sink Cross Traffic Cross Traffic H 3 2 1

H 3

C C C C C Path Source

2

Path

1

. . .

3 H 2 1

Sink Cross Traffic Sources Cross Traffic Sinks

Constantinos Dovrolis - IEEE Infocom 2001 - April 24-26, 2001

19 of 28

SLIDE 20

Packet-train dispersion in multi-hop paths

Can we derive the ADR for general cross traffic routing and paths? Cross traffic packets cause ‘bubbles’ between packet-train packets For one-hop path persistent cross traffic with C i

C:

C

P

H i u i

R
C
max

i H u i Lower bound: bubbles are never filled in Upper bound: bubbles are determined by tight link

Constantinos Dovrolis - IEEE Infocom 2001 - April 24-26, 2001

20 of 28

SLIDE 21

Packet-train dispersion: summary

Packet-trains: do not lead to more robust capacity estimation Packet-trains: do not lead to available bandwidth estimation As N increases, measurement range decreases and becomes unimodal As N increases, measurements tend to ADR ADR is always less than capacity Available bandwidth can be computed from ADR in single-hop paths

Constantinos Dovrolis - IEEE Infocom 2001 - April 24-26, 2001

21 of 28

SLIDE 22

Part IV

A capacity estimation methodology

Constantinos Dovrolis - IEEE Infocom 2001 - April 24-26, 2001

22 of 28

SLIDE 23

Pathrate: a capacity estimation methodology

Phase I:

Perform many (2000) packet-pair experiments to form distribution B Use packet size of about 800 bytes (maximum size: 1500 bytes) Determine local modes of distribution B Sequence of local modes in increasing order: M fm

m
m

M g

Constantinos Dovrolis - IEEE Infocom 2001 - April 24-26, 2001

23 of 28

SLIDE 24

Pathrate: a capacity estimation methodology (cont’)

Phase II:

Perform several packet-train experiments with certain N to get B N

If bandwidth distribution not unimodal, increase

N and repeat previous step Let

N be the minimum value of

N such that B N is unimodal Let

be range of the unique mode in

B N

Estimate capacity as:
C
m

k

min

fm i

M
m

i

g

Constantinos Dovrolis - IEEE Infocom 2001 - April 24-26, 2001

24 of 28

SLIDE 25

Example: Path

f100,70,60,40,55,80,65,90,40,75,90g Packet-pair modes M=f9,14,17,23,26,29,33,40,44,56,75,90g

10 20 30 40 50 60 70 80 90

Bandwidth (Mbps)

10 20 30 40 50 60 70

# of measurements

P={100,70,60,40,55,80,65,90,40,75,90}, u=30%

L=770B N=2 CM=40Mbps Lc: uniform in [40,1500]B

10 20 30 40 50 60 70 80 90

Bandwidth (Mbps)

10 20 30 40 50 60 70 80 90 100 110 120 130 140

# of measurements

P={100,70,60,40,55,80,65,90,40,75,90}, u=30%

L=770B N=8 ζ Lc: uniform in [40,1500]B =37Mbps

+

Constantinos Dovrolis - IEEE Infocom 2001 - April 24-26, 2001

25 of 28

SLIDE 26

From CAIDA (San Diego) to ETH (Zurich)

Packet-pair modes M=f9,11,13,15.5, 19.5, 27, 32, 43g

10 20 30 40 50 60

Bandwidth (Mbps)

10 20 30 40 50 60

# of measurements

jhana (CAIDA) to drwho (Zurich), L=1500B

N=2 CM=27Mbps

10 20 30 40 50 60

Bandwidth (Mbps)

10 20 30 40 50

# of measurements

jhana (CAIDA) to drwho (Zurich), L=1500B

N=12 ζ+ =24Mbps

Constantinos Dovrolis - IEEE Infocom 2001 - April 24-26, 2001

26 of 28

SLIDE 27

Pathrate verification

10 20 30 40 50 60 70 80

Actual capacity (Mbps)

10 20 30 40 50 60 70 80

Capacity estimate (Mbps) ϖ=1Mbps

10 20 30 40 50 60 70 80

Actual capacity (Mbps)

10 20 30 40 50 60 70 80

Capacity estimate (Mbps)

w=1Mbps if C<40Mbps, w=2Mbps otherwise

ϖ ϖ

=1Mbps =2Mbps

Higher capacity values require larger bandwidth resolution

Adaptive selection of bandwidth resolution?

Constantinos Dovrolis - IEEE Infocom 2001 - April 24-26, 2001

27 of 28

SLIDE 28

Summary and contributions

Capacity estimation in heavily-loaded paths is hard! Statistical filtering of packet-pair measurements does not work Use of maximum-sized packets is not optimal Packet-train dispersion does not estimate available bandwidth Available bandwidth can be computed from ADR in single-hop paths Pathrate takes into account effect of cross traffic on packet-trains

Constantinos Dovrolis - IEEE Infocom 2001 - April 24-26, 2001

28 of 28