department of communication technology

Post on 05-Jan-2016

28 Views

Category:

Documents

1 Downloads

Preview:

Click to see full reader

DESCRIPTION

Department of Communication Technology. Video Streaming over 802.11b LAN Wireless channel unreliability : managing the starvation phenomenon Mohamed Ali Ben Abid Monday, 28 June 2004. Supervisors Censors - PowerPoint PPT Presentation

TRANSCRIPT

Video Streaming over 802.11b LANWireless channel unreliability : managing the

starvation phenomenon

Mohamed Ali Ben Abid

Monday, 28 June 2004

Department of Communication Technology

Supervisors CensorsSupervisors Censors

Frank H.P. Fitzek Karsten ThygesenHans Peter Schwefel Thomas Toftegaard Nielsen

3

Actual Concept

802.11b LAN: mobility, high data speed

Video Streaming: more and more expanded in the wired network

Video Streaming over 802.11b LAN, a promising combination.

4

Project Presentation (1)

Goal : Optimizing the video client’s resources while maintaining a good video quality.

Means : Managing the Playout Buffer of the video. Estimating a buffer compensation for the

wireless channel unreliability.

5

Outline• Background

The 802.11b LAN Video Streaming

• The StudyProblem SettingScenarioMethodologyResultsConclusion

Project Presentation (2)

Background The 802.11b LANThe 802.11b LAN

Video StreamingVideo Streaming

7

802.11b LAN - Architecture

• different BSS, different MN

• 1 BSS controlled by 1 AP

The 802.11b LANThe 802.11b LAN

Access Mechanism

Layers

Errors

Architecture

Background

8

802.11 layers

• PHY layer : data transmission

• 802.11 MAC : fragmentation, Ack

• 802.2 : packets retransmission

The 802.11b LANThe 802.11b LAN

Access Mechanism

Layers

Errors

Architecture

Background

9

CSMA/CA Access Mechanism (1)

802.11b LAN802.11b LAN

Access Mechanism

Layers

Architecture

Background

Errors

IFS

SIFS : separate transmissions, 28 μs

DIFS : station to start transmission, 128 μs

Positive Acknowledgement

Virtual Carrier Sense

• hidden node problem

• RTS/CTS

10

CSMA/CA Access Mechanism (2)

The 802.11b LANThe 802.11b LAN

Access Mechanism

Layers

Architecture

Background

Errors

The access method is Distributed Coordination Function (DCF)

11

CSMA/CA Access Mechanism (3)

The 802.11b LANThe 802.11b LAN

Access Mechanism

Layers

Architecture

Background

Errors

• The Backoff algorithm :

• Contention window from CW_min (16) to CW_max (1024).

• m = maximum transmissions times.

12

Errors in the channel

The 802.11b LANThe 802.11b LAN

Access Mechanism

Layers

Errors

Architecture

Background

Main Types of errors : frame loss / erroneous frames.

Causes of errors due to the channel :

Shadowing

Multipath fading

PHY layer adjusting the sending rate.

Detection/Correction Mechanisms :

if CRC failed, frame discarded

each MAC frame ACKnowledged (unicast)

ARQ (Send and Wait)

FEC (adds redundant bits)

BackgroundThe 802.11b LANThe 802.11b LAN

Video StreamingVideo Streaming

14

Video Structure

Video StreamingVideo Streaming

Real-time Requirements

Streaming principle

Video structure

Background

Protocol Stack

def:

Video frame = Picture

• e.g. QCIF compression format : 1 picture = 176*144 pixels

• with YUV representation, 1 pixel : 3Bytes

Gives frame size (Byte)

15

Streaming principle (1)

Video StreamingVideo Streaming

Real-time Requirements

Streaming principle

Video Structure

Background

Protocol Stack

Why is frame size variable ?

16

Streaming principle (2)

Video StreamingVideo Streaming

Real-time Requirements

Streaming principle

Video Structure

Background

Protocol Stack

• Example of frame size PDF (Friends 2x16)

here, the total number of frames is 32455

17

Video Requirements

• Burstiness of video + wireless channel unreliability Packet losses & delays

Video StreamingVideo Streaming

Real-time Requirements

Streaming principle

Video Structure

Background

Protocol Stack

Tradeoff : number of Data Link retransmission Nr / delay introduced.

FER < 8/100

Nr_max = 4 (unicast)

= 0 (multicast)

UDP traffic (no layer 4

retransmission)

18

Protocol Stack

Video StreamingVideo Streaming

Real - time Requirements

Streaming principle

Video Structure

Background

Protocol Stack

The Study Problem Setting

ScenarioScenario

MethodologyMethodology

ResultsResults

ConclusionConclusion

20

Problem Setting (1)

Main ProblemMain Problem

PBO constraints

definitions

ε dependences

PBO/IBO

The Study

Playout Buffer Occupancy (PBO) :

Intitial Buffer Occupancy (IBO) =

T_start(display) – T_start(buffer filling)

21

Problem Setting (2)

Main ProblemMain Problem

PBO constraints

definitions

ε dependences

PBO/IBO

The Study

• θ ?

• M ? Overflow ?

• T0, T’ ?

• Starvation, interruption ?

Playout Buffer Occupancy

(PBO)free in an error free channel

22

Problem Setting (3)

• P9, P10…still not in buffer

• e.g. if F4 = P8, F4 displayed, buffer empty : starvation

. Then, e.g. if F5 = (P9,P10)

& if P9, P10 did not arrive

interruption in display

Main ProblemMain Problem

PBO constraints

definitions

ε dependences

PBO/IBO

The Study

23

Problem Setting (4)• θ = Initial buffer occupancy (error free

channel)• ε = Buffer compensation to the

wireless channel unreliability• Initial_Buffer = θ + ε

0 <(a) PBO = PBOfree + ε < M+ ε <(b)S (a) = no interruption (b) = no buffer overflow

Main ProblemMain Problem

PBO constraints

Variables definition

ε dependences

PBO/IBO

The Study

Project focus : (a)

given wireless scenario/ given video

Chose an appropriate ε

24

Problem Setting (5)

ε depends on the following parameters :

Wireless conditions• N = number of MNs• Distance(s) laptop(s)/AP• Competing traffic(s)• FER (must be < 8%)• NLoS• Interference (neglected)• Handovers (not here)

Video Features• Θ, T’A priori estimation : ε < 5%* Θ Main ProblemMain Problem

PBO constraints

Variables definition

ε dependences

PBO/IBO

The Study

The StudyProblem SettingProblem Setting

ScenarioScenario

MethodologyMethodology

ResultsResults

ConclusionConclusion

26

Scenario (1)

• Server : desktop, P3-800MHz, 256MB RAM, 100Mbps Ethernet Card, 10/100 BaseT cable

•AP is Nokia A032 and cards are Nokia C110

•MN = 1 laptop P4-2.2GHz, 256MB RAM, WinXP

ScenarioScenario

4 scenariii

Main features

Experiment Scheme

The Study

27

Scenario (2)

layer 3 fragmentation threshold :

1475 B No L3 fragmentation

layer 2 fragmentation threshold :

2346 B No L2 fragmentation

• UDP datagram size = 1460 B

ScenarioScenario

4 scenariii

Main features

Experiment Scheme

The Study

28

Scenario (3)

• Video modelized by the traffic (Friends 2x16)

duration :1300 s mean rate : 759486 bit/s

Iperf generated traffic is UDP traffic sent with a rate of 759486 bit/s for 1300s.

ScenarioScenario

4 scenariii

Main features

Experiment Scheme

The Study

29

Scenario (4)

ScenarioScenario

4 scenariii

Main features

Experiment Scheme

The Study

• Unicast / Multicast

30

Scenario (5)

• Channel : Non overlapping conditions

Automatically choosed channel is number 10, but experiments made again with channel 1, 7, 13 (no difference / no interference problem)

ScenarioScenario

4 scenariii

Main features

Experiment Scheme

The Study

31

Scenario (6)

• 4 scenarii :

ScenarioScenario

4 scenariii

Main features

Experiment Scheme

The Study

(*) UDP traffic sent at 759486 bps from time 0s to 1300s.

& competing TCP traffic sent at 4.38 Mbps from time 360s to 960s.

The Study Problem SettingProblem Setting

ScenarioScenario

MethodologyMethodology

ResultsResults

ConclusionConclusion

33

Methodology (1)

Data is sent by the server with the CBR : λArrival Times delivered by Ethereal

cumulative data volume V(t) can be plotted:

MethodologyMethodology

Deducing ε

Plotting the margin

Definitions

The Study

34

Methodology (2)

• The Cumulative (receiving) throughput,

Λ(t) = V(t)/t < λ ; (t>0)

• The margin function μ(t) :

μ(t) = [ λ - Λ(t) ]*t

= λ*t – V(t) > 0 ; (t>0)MethodologyMethodology

Deducing ε

Plotting the margin

Definitions

The Study

35

Methodology (3)

the difference gives μ(t)

MethodologyMethodology

Deducing ε

Plotting the margin

Definitions

The Study

36

Methodology (4) – deducing ε

then, plotting :

the Probability Density Function (PDF)

of the margin μ the Cumulative Distribution Function

(CDF) of the margin μ

MethodologyMethodology

Deducing ε

Plotting the margin

Definitions

The Study

37

Methodology (5) – deducing ε

• Also, using the PBO of the video (during the time T’

MethodologyMethodology

Deducing ε

Plotting the margin

Definitions

The Study

38

Methodology (6) – deducing ε

MethodologyMethodology

Deducing ε

Plotting the margin

Definitions

The Study

39

Methodology (7) – deducing ε

• Choosing an appropriate ε ?Simple method : (e.g) ε = μ / CDF(μ) =0.9More judicuous:

Pstarvation = (Pr (B + < x) . fμ (x). dx < 10-4

 where, B = PBOfree and x from to infinity

(FB (x - ) . fμ (x). dx < 10-4

  ( CDF [PBOfree(x - )] *

PDF [(x)]. dx < 10-4

 

MethodologyMethodology

Deducing ε

Plotting the margin

Definitions

The Study

40

The Study

Problem SettingProblem Setting

ScenarioScenario

MethodologyMethodology

ResultsResults

ConclusionConclusion

41

Remembering Scenarii

42

Results (1)

• For Friends 2x16, θ = 6.79 Mbyte 5 % * θ ~ 0.3 MByte• Using the simple method:

Scenario 1 : ε = 0.25 MByte Scenario 2 : ε = 0.30 MByte Scenario 3 : ε = 2.75 Mbyte !!! (need to use the second method found 1.4 Mbyte with method 2)Scenario 4 : ε = 0.31 MByte

ResultsResults

Problems Managing

SEQuence number

Found ε /scenario

The Study

43

Results (2)

• Ethereal : IP ID field SEQ numbers of missing packets

ResultsResults

Problems Managing

SEQuence number

Found ε /scenario

The Study

44

Results (3)

ResultsResults

Problems Managing

SEQuence number

Found ε /scenario

The Study

45

Results (4)

ResultsResults

Problems Managing

SEQuence number

Found ε /scenario

The Study

46

Results (5)

ResultsResults

Problems Managing

SEQuence number

Found ε /scenario

The Study

47

Results (6)

• Pb 1 : μ(t) sometimes negative ?!?

μ(t) = = λ*t – V(t) > 0 ; (t>0)

e.g : scenario 2

ResultsResults

Problems Managing

SEQuence number

Found ε /scenario

The Study

48

Results (7)

Choice of origin !!

ResultsResults

Problems Managing

SEQuence number

Found ε /scenario

The Study

49

Results (8)

• Pb2 : Why cumulative loss data is different from the maximum value of μ ?

e.g. (scenario 2) respectively 0.17 Mbit & 2.4 Mbit

AP adjusting the sending rate :

AP sends with λAP < λ

& λAP is variable (VBR)ResultsResults

Problems Managing

SEQuence number

Found ε /scenario

The Study

50

Results (9)

• Future possible corrections: study λAP (Sniffer near AP)

Suppress the time in the wired network

• Ter (wired) = Temission-reception

Temission = 1460*8/10*106 (10Mbps) =1.17ms

Tpropag = 5*2/200000 = 0.085 ms (neglected)

T traitment , Tqueues (negleted)

ResultsResults

Problems Managing

SEQuence number

Found ε /scenario

The Study

51

Results (10)

• Ter (wired) ~ 1.17 ms

• mean IAT = 1460*8/ λ = 15 ms

ResultsResults

Problems Managing

SEQuence number

Found ε /scenario

The Study

52

The Study

Problem SettingProblem Setting

ScenarioScenario

MethodologyMethodology

ResultsResults

ConclusionConclusion

53

Conclusion (1)

• Tradeoff between wireless channel unreliability and Video Streaming stringent QoS requirements

• ε defined as buffer compensation

to manage the starvation phenomenom

• ε depends both on the wireless conditions and the video features

54

Conclusion (2)

• Video features :PBOfree , T’ and θ• Wireless parameters : distance

AP/laptop, Mode, traffic duration, datagram lengths, mean rate, competing traffic, NloS…

• CBR λ, volume V(t)• Margin function defined :μ(t) = λ*t – V(t) > 0 ; (t>0)

55

Conclusion (3)

• ε is deduced from the PBOfree (video features) and μ (wireless conditions)

e.g : ε / ( CDF [PBOfree(x - )] * PDF [(x)]. dx < 10-4

ε ~ 5% θ (unicast)

ε ~ 20% θ !! (multicast) to be reviewed

56

Conclusion (4)

• Future work : solve origin problems consider λAP instead of λ

(use of sniffer in air interface)mobility/handoversDifferent laptops with different traffics at

different starting times

57

THANK YOU

Mohamed Ali Ben Abid

Supervisors

Frank H. P. Fitzek

Hans Peter Schwefel

Censors

Karsten Thygesen

Thomas Toftegaard Nielsen

top related