On the Fixpoint Problem of QoE ChargingQoE Charging

Peter Reichl

RBUCE WEST I t ti l R h Ch i NICE“RBUCE WEST International Research Chair „NICE“(Network-based Information and Communication Ecosystems)

Université Européenne de Bretagne / Télécom Bretagnep g / g

IRISA Rennes, 31.05.2011

Agenda

I t d ti d F k Introduction and Framework

Starting Point: Charging for Quality Starting Point: Charging for Quality

F Q lit f S i t Q lit f E i From Quality-of-Service to Quality-of-Experience

Charging for QoE

Conclusions & Outlook

Towards Future Communication Ecosystems Starting point: telecommunications > communication engineering

Idea: holistic and interdisciplinary approach Idea: holistic and interdisciplinary approach

New framework: future communication ecosystems

Dimensions: techno-economics vs socio-economics

Towards Future Communication Ecosystems Starting point: telecommunications > communication engineering

Idea: holistic and interdisciplinary approach Idea: holistic and interdisciplinary approach

New framework: future communication ecosystems

Dimensions: techno-economics vs socio-economics

11 religion11 religion10 politics9 legal8 commercial7 application4 transport3 network2 link

C l i 1 Shif h f f f i d d i

2 link1 physical

© Bob Briscoe

Main focus no longer on technology, but what we can do with itConclusion 1: Shift the focus of future services and system design towards the user and her real needs.

Example: Kilkki‘s Quality Ecosystem Model

Idea: describe ecosystem for Quality of Experience (QoE) based on the diverse roles of the end user

QoE MOSQUEQGE ?

QoE MOSQUEQGE ?

UserUserPersonPersonGroupmemberGroup

member ApplicationApplication

QUEQGE ?UserUserPersonPersonGroup

memberGroup

member ApplicationApplication

QUEQGE ?

membermember

QoSmembermember

QoS

CustomerCustomer NetworkNetworkQCECustomerCustomer NetworkNetworkQCEQoSQoEQUEQCEQGEMOS

Quality-of-ServiceQuality-of-ExperienceQuality-of-User ExperienceQuality-of-Customer ExperienceQuality-of-Group-member ExperienceM O i i S

QoSQoEQUEQCEQGEMOS

Quality-of-ServiceQuality-of-ExperienceQuality-of-User ExperienceQuality-of-Customer ExperienceQuality-of-Group-member ExperienceM O i i SMOS Mean Opinion ScoreMOS Mean Opinion Score

Note: each subconcept to be modeled and evaluated

[Kilkki 2008]

Note: each subconcept to be modeled and evaluatedindependently in order to arrive at global QoE

An Interdisciplinary „Triangle of Forces“

Summary: research in telecommunications no longer restricted to the engineering partg g p

Hence: need for truly interdisciplinary research

U Basic framework: “NUT Trilemma”

= triangle of forces between

TN

= triangle of forces between- Network efficiency ( economics)- User acceptance ( usability)

[R., Hausheer, Stiller 2003]User acceptance ( usability)

- Technical feasibility ( technology)

Challenge: putting an interdisciplinary approach into practice

Conclusion 2: Start from solid technological grounds and integrate Conclusion 2: Start from solid technological grounds and integrate economic and user aspects into a holistic framework.

Charging for Quality

Quality as Key Differentiator

Global („exponential“) explosion of traffic volume Intensified competition among Telco‘s and ISPs Intensified competition among Telco s and ISPs Question: how to keep customers happy (while at the same

time staying in business)?

Quality as key differentiator Tight link quality↔charging

Global Consumer Internet Traffic Volume (Forecast).Source: Cisco VNI 2011.

Pricing for Differentiated Services

Some years ago in the subway of Paris:

- users are offered choice of travelling in 1st or 2nd class carriages

diff b t l - no difference between classes (same seats, same space, same service, same timetable) except for the price

Odlyzko‘s proposal: apply this scheme to packet-based networks „Paris Metro Pricing“ [Odlyzko 1998]g

Basic idea: network-level feedback cycle

[Od y o 998]

Charging priceQoS

demand

Pricing for Differentiated Services (cont‘d)

Question: How should service differentiation look like from an economic perspective? And will this work for the case of (e.g. t ) ti ISP ? two) competing ISPs?

Cases: - Social Planner maximize welfare- Monopoly/Duopoly maximimize profit for provider

Assumption: small number of QoS classes (e.g. two only)

Model [Gibbens/Mason/Steinberg 2000]Model [Gibbens/Mason/Steinberg 2000]- Social planner + competing profit-maximizing networks,

each potentially offering several service classes- Classes = (sub-)networks with different congestion levels- Congestion of network determined by number of users on

network and network capacitynetwork and network capacity

The Standard Model

User joining network i receives utility with V = positive benefit for user

ii pKViU ),(p

Ki = congestion-dependent dis-benefit = preference for lack of congestion

i it tipi = price per unit time

Further simplification: ii C

Qi

iK networkofcapacity

network on users of mass

Main assumptions of the model:

iCinetwork ofcapacity

- congestion function linear, utility function separable

- each user contributes identically to congestion

- prices subscription-based (not usage-based)

- preference drawn from uniform [0;1]-distribution (elastic traffic small , inelastic traffic large ))(elastic traffic small , inelastic traffic large ))

- each network maximizes individual profit: iii Qp

Game-theoretic Formulation

Provider has to make three decisions: number of networks – capacities – prices

Example: 2 networks (capacities CA, CB), pA pB

mass of users joining A is 1-* (uniform distribution of )

Network B Network A0 10 1

Nash equilibrium indifference relation for „marginal user“ *:

BA pVpVBUAU

1 *

**

***

BABABABB

BB

AA

ppCCCCCC

pC

VpC

VBUAU

4

,,

2*

BA CC

2

Example: Social Planner (SP)

Consider one network of total capacity 2CSP i i i t t l lf lit t k i t t l SP maximizing total welfare splits network into two equal „subnetworks“ and allocates user appropriately

Total user surplus equals Total user surplus equals

*1* 1*

dC

VdC

VVSP

1***221 23

*0

CV

CC

Choose * to maximize VSP:

2C

...6076.06

71*01*2*6*

2

d

dVSP

Optimal division of market share: 61% vs. 39%

From Quality-of-Service to Quality-of-Experience

Towards a Holistic Quality Framework Starting point: an interdisciplinary view on service quality

Interesting observation: original definitions of QoS strongly refer to user perception (e.g. ITU-T E.800: “effect of service performance which determines the degree of user satisfaction“)

However: since then research has focused more on QoS However: since then, research has focused more on QoS parameters than service quality itself

Recent reaction: Quality-of-Experience (QoE)Recent reaction: Quality of Experience (QoE)

Idea: holistic service quality frameworkQoS Access QoS AccessQoS Access

QoS BackboneQoD QoDQoS Quality-of-ServiceQoD Quality of DesignQoD Quality-of-DesignQoE Quality-of-Experience

User-to-User QoE[R. 2007]

Quality of Experience

Definition 1 [ITU-T 2008]: “Overall acceptability of a service or application, as perceived subjectively by the end-user”pp , p j y y

Definition 2 [Dagstuhl 2009]: “Degree of delight of the user of a service, influenced by content, network, device, application, user

i d l d f ”expectations and goals, and context of use”

Basic approaches: asking the user ( Mean Opinion Score, MOS)vs educated guess about probable answer (PESQ PSQA )vs. educated guess about probable answer (PESQ, PSQA,…)

Typical MOS Scale

MOS Value Quality5 Excellent

Typical MOS Scale

4 Good3 Fair2 Poor2 Poor1 Bad

Fundamental Laws for Quality of Experience

Basic question: what is the „value“ of a resource/service

elastictraffic

non-elastic traffic

y„ /for the end customer?

Formal answer (microeconomics): bandwidth

utili

ty

o a a s e ( c oeco o cs)ui(x) := utility function for customer i to receive service x

Usual assumptions: monotonically increasing concave Usual assumptions: monotonically increasing, concave, …

Typical candidate: logarithm functionth ti ll f ibl- mathematically feasible

- many nice properties, e.g. proportional fairness (Kelly et al.)

But: isn‘t there a better justification??

Inspiration: recent results from QoE evaluations

Example 1: VoIP Quality under PSQA

Rubino et al.: Pseudo-Subjective Quality Assessment (PSQA)l i t l f Q E f lti di li ti- learning tool for QoE of multimedia applications

- basic approach: Random Neural Networks

Scenario: Speex codec, bitrates varying from 2.4 to 24.8 kbps

[Rubino et al 2007][Rubino et al. 2007]

Example 2: QoE for Mobile Broadband

FTW Project ACE: Advancing the Customer Experience

Goal: predict user satisfaction with a service based on traffic data from a passive network monitoring tool

File download scenario: users download single MP3 and ZIP files at different network speeds (256 – 4096 kbps)

Result: logarithmic dependencies between bandwidth and MOS

[R., Tuffin, Schatz 2011]

The Weber-Fechner Law

Once upon a time (in fact 1834): E. Weber, G Fechner and the birth of psychophysicsG. Fechner and the birth of psychophysics

Idea: operation of human sensory systembased on just noticeable differences“based on „just noticeable differences

Formally: differential perception dPproportional to relative change dS/S of physical stimulus

log SkPdSkdP

Well-known principle for human vision, hearing, smelling,

0log

SkP

SkdP

Well known principle for human vision, hearing, smelling, touching, even numerical cognition…

Question: valid also in ICT context ?Question: valid also in ICT context ?

[R. et al. 2010, 2011]

„Time is Bandwidth“

Starting point: experimental setup suggests measurement of plain user waiting timeplain user waiting time

„WQL hypothesis“: relationship between Waiting time and its QoE evaluation on a linear ACR scale is LogarithmicQ g

Scenarios: connection setup time / page load times

[Egger et al. 2011]

„Time is Bandwidth“ – but not always

However: reality more complex

Example: web browsing

Issues: - Stimuli vs impairment- Time vs bandwidth- Perceived vs application

page load time

Conclusions:- pure waiting task (= simple web usage scenarios) follow the

WFL given that waiting time is considered as stimulusWFL, given that waiting time is considered as stimulus- existing web browsing QoE models that reduce interactive

browsing to a simple request-response transaction with a given waiting time considered not sufficient

[Egger et al. 2011]

Charging for Quality-of-Experience

Charging for QoE

Question: What does this mean for charging? Is there a difference between charging for QoS and charging for QoE?

Note: double role of charging under QoE- essential contribution to user expectation (context)- essential contribution to user expectation (context)- result of quality evaluation

Hence: extended fixed point problem (user-level feedback)

QoE

Charging

demand

priceQoS

demand

The Standard Model Revisited

Remember:

G li d i

ii pKViU ),(

ViU )()( Generalized version:

with V = positive benefit for user

iii ppqViU );,(),(

qi = offered network-level QoS = preference for QoEpi = price pi price

Questions: - Is there a general form for the utility function?

- Is the utility function still separable? Depending on the application?

First step: user trials

M3I Trials: Overview

General idea: quantitative and qualitative investigation of user attitudes and behaviour in a dynamic/differential pricing environment for Internet servicesenvironment for Internet services

User trials: conducted at BT in 2000/2001G l i i i f li d i bili- Goal: quantitative ratings of quality and service acceptability

- Test material: two 30 sec audio-video clips- Manipulated parameters:- Manipulated parameters:

- video quality level (1/5/10/15/20 frames per second) - number and degree of QoS changes (0/10/20 changes g Q g ( / / g

per minute)- Test method:

- apriori user classification: gold – silver – bronze - 5-grade MOS scale for quality rating

binary decision for acceptability - binary decision for acceptability - willingness-to-pay response panel

M3I User Trials: Results

Quality rating [MOS]: 4

5Bronze

4

5Silver

4

5Gold

1

2

3

1 5 10 15 25

1

2

3

1 5 10 15 25F R

1

2

3

1 5 10 15 25F R t

Acceptabilityrating [%]: 80

100

80

100

80

100

1 5 10 15 25Frame Rate Frame Rate Frame Rate

rating [%]:

20

40

60

20

40

60

0

20

40

60

Willingness-to-pay

01 5 10 15 25

Frame Rate

01 5 10 15 25

Frame Rate

01 5 10 15 25

Frame Rate

25 25 25

[pence per min]:

Constant QoS 10

15

20

5

10

15

20

5

10

15

20

Q10 changes/minute20 changes/minute

0

5

1 5 10 15 25Frame Rate

0

5

1 5 10 15 25Frame Rate

0

5

1 5 10 15 25Frame Rate

ETICS Trials

Lessons learned from M3I: - Realistic test situation + flexible on-demand scenarios- simplicity of interaction (including acceptance and revision

of choices)

Basic setting of ETICS trials: - Real-time transmission interconnection scenarios (VoD)( )- QoS modeled in terms of different packet loss rates- User willingness-to-pay in terms of real money- Each movie lasts 20 min and starts with free trial phase- Three more „quality selection phases“ during runtime

(upgrade possible downgrade not allowed)(upgrade possible, downgrade not allowed)- Remaining deposit paid to the user after end of trial- In addition, smooth increase of loss rate possible without , p

notification of the user

Technical Setup

Basic setup: - extensive video library hosted on a content servery- VLC server streams selected video over Ethernet to thin client

connected to 40 inch LCD TV- after selection of movie, user may modify packet loss rate

and charge via Web interface on tablet PC

[Sackl et al. 2012]

Video Quality Selection

Four offered quality classes:

[Sackl et al. 2012]

Results

43 participants, 129 videos, quality assessment after each video

Interaction behaviour:

[Sackl et al. 2012]

Results

The price of quality:

User classification:

[Sackl et al. 2012]

S d C l iSummary and Conclusions

What have we learned

Introduction: fundamental paradigm change towards information and communication ecosystems

Key example: Quality-of-Service / Quality-of-Experience

Starting point: charging for QoS

- Game-theoretic model for service differentiation

Recent step: from QoS to QoE

Question: charging for QoE

- Extended fix point problem

Initial steps: M3I and ETICS trials

C l i Conclusions

References and Further Reading S. Egger, P. Reichl, T. Hossfeld, R. Schatz: "Time is Bandwidth"? Narrowing the Gap between Subjective Time Perception and Quality

of Experience. Proc. IEEE International Conference on Communications (ICC’12) – Communication QoS, Reliability and ModelingSymposium, Ottawa, Canada, June 2012.

M. Fiedler, K. Kilkki, P. Reichl (eds.): From Quality of Service to Quality of Experience. Dagstuhl Seminar Proceedings 09192, Schloss Dagstuhl Germany 2009 URL: http://drops dagstuhl de/portals/index php?semnr=09192/Dagstuhl, Germany, 2009. URL: http://drops.dagstuhl.de/portals/index.php?semnr=09192/

R. Gibbens, R. Mason, R. Steinberg: Internet Service Classes under Competition. IEEE JSAC, vol. 18, no. 12, Dec 2000.

K. Kilkki: Quality of experience in communications ecosystem. Journal of Universal Computer Science, vol.14, pp. 615–624, 2008.

A. Odlyzko: Paris Metro Pricing for the Internet. Proc. ACM Conference on Electronic Commerce 1998.

P. Reichl: From ‘Quality-of-Service’ and ‘Quality-of-Design’ to ‘Quality-of-Experience’: A Holistic View on Future Interactive Telecommunication Services. Invited Paper, 15th IEEE SoftCOM’07, Split, Croatia, Sept. 2007.

P. Reichl: From Charging for Quality-of-Service to Charging for Quality-of-Experience. Annals of Telecommunications, 65 (3) pp. 189–199, 2010.

P Reichl S Egger R Schatz A D’Alconzo: The Logarithmic Nature of QoE and the Role of the Weber Fechner Law in QoE P. Reichl, S. Egger, R. Schatz, A. D’Alconzo: The Logarithmic Nature of QoE and the Role of the Weber-Fechner Law in QoEAssessment. Proc. IEEE ICC, Cape Town, South Africa, May 2010.

P. Reichl, D. Hausheer, B. Stiller: The Cumulus Pricing Model as an Adaptive Frame-work for Feasible, Efficient and User-friendly Tariffing of Internet Services. Journal of Computer Networks, vol. 43 (1), pp. 3 – 24, Elsevier, Sept. 2003.

P. Reichl, B. Tuffin, R. Schatz: Logarithmic Laws in Service Quality Perception: Where Microeconomics Meets Psychophysics andP. Reichl, B. Tuffin, R. Schatz: Logarithmic Laws in Service Quality Perception: Where Microeconomics Meets Psychophysics and Quality of Experience. Telecommunication Systems Journal (Springer) vol. 55 no. 1, Jan. 2014. Published online 18 June 2011.

G. Rubino: Quantifying the Quality of Audio and Video Transmissions over the Internet: the PSQA Approach. In: J. Barria (ed.), Design and Operations of Communication Networks: A Review of Wired and Wireless Modelling and Management Challenges. Imperial College Press 2005.

A S kl S E P Z i kl P R i hl Q E Al h i T i Q lit i t M E i ith R fi d M th d l f th A. Sackl, S. Egger, P. Zwickl, P. Reichl: QoE Alchemiy: Turning Quality into Money. Experiences with a Refined Methodology for the Evaluation of Willingness-to-Pay. 4th International Workshop on Quality of Multimedia Experience (QoMEX’12), Yarra Valley, Australia, July 2012.

A. Sackl, P. Zwickl, P. Reichl: From Quality of Experience to Willingness to Pay for Interconnection Service Quality. Proc. ETICS Workshop @ IFIP NETWORKING’12, Prague, Czech Republic, May 2012. Springer LNCS.

B. Stiller, P. Reichl, J. Gerke, P. Flury: A Generic and Modular Internet Charging System for the Cumulus Pricing Scheme. Journal of Network and Systems Management 3 (9), pp. 293 – 325, Sept. 2001.

The End

Prof. Dr. Peter Reichl

RBUCE WEST International Research Chair“Network-based Information and CommunicationEcosystems” (NICE)

é é d / éléUniversité Européenne de Bretagne / Télécom Bretagne

Key Researcher“User-Centered Interaction & Communication Economics”FTW Forschungszentrum Telekommunikation WienFTW Forschungszentrum Telekommunikation Wien

peter.reichl@telecom-bretagne.eureichl@ftw.at

Thanks for your attention!Any feedback welcome!Any feedback welcome!