Avoiding zigzag quality switching in real contentadaptive video streaming

Wassim Ramadan, Eugen Dedu and Julien Bourgeois

Laboratoire d’Informatique de l’Universite de Franche-ComteMontbeliard, France

DICTAP21–23 June 2011

Wassim Ramadan, Eugen Dedu and Julien Bourgeois Avoiding zigzag quality switching 1 / 15

Introduction Zigzag avoidance Experiments Conclusions and perspectives

Goals of zigzag avoiding

Improve user video experience by minimizing unnecessaryquality switching

Optimise network resource usage by minimizing packet lossrate

Wassim Ramadan, Eugen Dedu and Julien Bourgeois Avoiding zigzag quality switching 2 / 15

Introduction Zigzag avoidance Experiments Conclusions and perspectives

Video adaptation and zigzag quality switching

Dynamic bandwidth vs static bitrates

0

0.5

1

1.5

2

2.5

3

3.5

0 20 40 60 80 100 120 140 160 180

Bitr

ate

(Mbi

t/s)

Time (s)

available bandwidthQ0

Q1Q2

Q3

Ideal bitrate

If bitrate<bw:non-optimised

If bitrate>bw: lost packets

Ideal: bitrate=bw

many zigzags occur

0

0.5

1

1.5

2

2.5

3

0 20 40 60 80 100 120 140 160 180

0

20

40

60

80

100

120

Bitr

ate

of tr

ansm

itted

vid

eo (

Mb/

s)

Time (s)

adapted video bitrate

Zigzag problem

Adaptation leads to:

many zigzags

poor quality

Wassim Ramadan, Eugen Dedu and Julien Bourgeois Avoiding zigzag quality switching 3 / 15

Introduction Zigzag avoidance Experiments Conclusions and perspectives

Video adaptation and zigzag quality switching

Dynamic bandwidth vs static bitrates

0

0.5

1

1.5

2

2.5

3

3.5

0 20 40 60 80 100 120 140 160 180

Bitr

ate

(Mbi

t/s)

Time (s)

available bandwidthQ0

Q1Q2

Q3

Ideal bitrate

If bitrate<bw:non-optimised

If bitrate>bw: lost packets

Ideal: bitrate=bw

many zigzags occur

0

0.5

1

1.5

2

2.5

3

0 20 40 60 80 100 120 140 160 180

0

20

40

60

80

100

120

Bitr

ate

of tr

ansm

itted

vid

eo (

Mb/

s)

Time (s)

adapted video bitrate

Zigzag problem

Adaptation leads to:

many zigzags

poor quality

Wassim Ramadan, Eugen Dedu and Julien Bourgeois Avoiding zigzag quality switching 3 / 15

Introduction Zigzag avoidance Experiments Conclusions and perspectives

Outline

1 Zigzag avoidance

2 Experiments

3 Conclusions and perspectives

Wassim Ramadan, Eugen Dedu and Julien Bourgeois Avoiding zigzag quality switching 4 / 15

Introduction Zigzag avoidance Experiments Conclusions and perspectives

Outline

1 Zigzag avoidanceZAAL (Zigzag Avoidance ALgorithm)

Wassim Ramadan, Eugen Dedu and Julien Bourgeois Avoiding zigzag quality switching 5 / 15

Introduction Zigzag avoidance Experiments Conclusions and perspectives

Overview of ZAAL

ZAAL Zigzag Avoidance ALgorithm

ZAAL maintains a successfulness value for each bitrate which:

is an arithmetic real value between 0 and 1

indicates the successfulness of the bitrate if used

uses a historic of the application last attempts to use thebitrate

is calculated at the end of an adaptation period

is obtained by applying an EWMAa algorithm on the bitratesuccessfulness historic

aEWMA: Exponential Weighted Moving Average

Wassim Ramadan, Eugen Dedu and Julien Bourgeois Avoiding zigzag quality switching 6 / 15

Introduction Zigzag avoidance Experiments Conclusions and perspectives

Details of ZAAL

ZAAL works as follows

at the beginning of a video transmission, all successfulnessvalues Si

a are set to 1

the application switches to a higher quality i + 1 only ifSi > β b

ZAAL uses the following general formula:Si = (1 − α/d)Si + s(α/d) c de

ai : bitrate indexbβ = 0.7 switching thresholdcif the bitrate succeeds s = 1 else s = 0dα = 0.3 degree of EWMA weighting increase/decreaseeand d = 1, 2 or 4 division factor

Wassim Ramadan, Eugen Dedu and Julien Bourgeois Avoiding zigzag quality switching 7 / 15

Introduction Zigzag avoidance Experiments Conclusions and perspectives

Formula of ZAAL

Successfulness value can change in three cases

Application increasesthe bitrate

Application maintainsthe bitrate

Application decreasesthe bitrate

s=1d=1

S j : ji=1−α S jα S i=1−α S i

s=0d=1

S j : ji=1−α S jα

s=1d=1

s=1d=4

s=1d=2

S i=1−α /2 S iα /2

S i1=1−α / 4S jα / 4

(a) (b)Wassim Ramadan, Eugen Dedu and Julien Bourgeois Avoiding zigzag quality switching 8 / 15

Introduction Zigzag avoidance Experiments Conclusions and perspectives

Outline

2 ExperimentsNetwork topologyZigzag avoidance checking

Wassim Ramadan, Eugen Dedu and Julien Bourgeois Avoiding zigzag quality switching 9 / 15

Introduction Zigzag avoidance Experiments Conclusions and perspectives

Network used in experiments

Network parameters

Parameter name Parameter value

Experiments place In buildingPacket size 1024 bytesPC1 (sender): Wired card, 100Mb/sPC2 (shaper machine): Wireless card, 802.11b/g 54Mb/sPC2 (shaper machine): Wired card, 100Mb/sPC3 (receiver): Wired card, 100Mb/sPC1,2&3 OS Linux (Ubuntu 64bits)PC1,2&3 OS kernel 2.6.35 genericDCCP Included in the kernelDistance (AP ↔ PC2) 50cm

Wassim Ramadan, Eugen Dedu and Julien Bourgeois Avoiding zigzag quality switching 10 / 15

Introduction Zigzag avoidance Experiments Conclusions and perspectives

Zigzag avoidance, ex. of one flow in (case of) traffic shaping

0

0.5

1

1.5

2

2.5

3

0 20 40 60 80 100 120 140 160 180

0

20

40

60

80

100

120

Bitr

ate

of tr

ansm

itted

vid

eo (

Mb/

s)

Time (s)

adapted video bitrate

(c) without ZAAL: many zigzags occur

0

0.5

1

1.5

2

2.5

3

0 20 40 60 80 100 120 140 160 180

0

20

40

60

80

100

120

Bitr

ate

of tr

ansm

itted

vid

eo (

Mb/

s)

Time (s)

adapted video bitrate

(d) with ZAAL: few zigzag occur

Wassim Ramadan, Eugen Dedu and Julien Bourgeois Avoiding zigzag quality switching 11 / 15

Introduction Zigzag avoidance Experiments Conclusions and perspectives

Performance comparisonTwo experiments: one flow and ten concurrent flows

Traffic shaping 10 concurrent flowsMethod Sent pkts Received pkts Lost pkts Sent pkts Received pkts Lost pktsWithout ZAAL 47795 42043 5752 (12%) 41191 32307 8884 (21%)With ZAAL 43548 39865 3683 ( 8%) 36713 32477 4236 (11%)

Number of sent and received packets (average of all flows) with and without ZAAL

in the first experiment, ZAAL has less sent and receivedpackets than without ZAAL, but using ZAAL is more usefulbecause it avoids the zigzag and leads to 30% fewer packetlosses

in the second experiment, ZAAL is better in terms of sent andreceived packets, avoiding the zigzag in the same time

Wassim Ramadan, Eugen Dedu and Julien Bourgeois Avoiding zigzag quality switching 12 / 15

Introduction Zigzag avoidance Experiments Conclusions and perspectives

Outline

3 Conclusions and perspectives

Wassim Ramadan, Eugen Dedu and Julien Bourgeois Avoiding zigzag quality switching 13 / 15

Introduction Zigzag avoidance Experiments Conclusions and perspectives

Conclusions

ZAAL is a general solution, since it can be integrated to anyadaptation method

ZAAL avoids the undesirable constant switching in quality(the zigzag problem)

Perspectives

working on a hybrid solution, which uses a bandwidthestimation method together with ZAAL

Wassim Ramadan, Eugen Dedu and Julien Bourgeois Avoiding zigzag quality switching 14 / 15

Introduction Zigzag avoidance Experiments Conclusions and perspectives

Conclusions

ZAAL is a general solution, since it can be integrated to anyadaptation method

ZAAL avoids the undesirable constant switching in quality(the zigzag problem)

Perspectives

working on a hybrid solution, which uses a bandwidthestimation method together with ZAAL

Wassim Ramadan, Eugen Dedu and Julien Bourgeois Avoiding zigzag quality switching 14 / 15

Introduction Zigzag avoidance Experiments Conclusions and perspectives

Questions?

Wassim Ramadan, Eugen Dedu and Julien Bourgeois Avoiding zigzag quality switching 15 / 15