Hossfeld qc man2015_context_monitoring_web

25
Prof. Dr. Tobias Hoßfeld Chair of Modeling of Adaptive Systems (MAS) Institute for Computer Science and Business Information Systems (ICB) University of Duisburg-Essen Can context monitoring improve QoE? A case study of video flash crowds in the Internet of Services Hossfeld, Tobias; Skorin- Kapov, Lea; Haddad, Yoram; Pocta, Peter; Siris, Vasilios A.; Zgank, Andrej; Melvin,

Transcript of Hossfeld qc man2015_context_monitoring_web

Page 1: Hossfeld qc man2015_context_monitoring_web

Prof. Dr. Tobias Hoßfeld

Chair of Modeling of Adaptive Systems (MAS)Institute for Computer Science and Business Information Systems (ICB)University of Duisburg-Essen

www.mas.wiwi.uni-due.de

Can context monitoring improve QoE? A case study of video flash crowds in the Internet of Services

Hossfeld, Tobias; Skorin-Kapov, Lea; Haddad, Yoram; Pocta, Peter; Siris, Vasilios A.; Zgank, Andrej; Melvin, Hugh

Page 2: Hossfeld qc man2015_context_monitoring_web

Definition of Context and Context Influence Factors

• Context is any information that assists in determining a situation(s) related to a user, network or device.[A.K. Dey and G.D. Abowd. Toward a better understanding of context and context-awareness, Technical Report Georgia Institute of Technology]

• Context refers to anything that can be used to specify or clarify the meaning of an event. [P. Reichl et al, Towards a comprehensive framework for QoE and user behavior modelling, QoMEX 2015]

• Context influence factors are factors that embrace any situational property to describe the user’s environment in terms of physical, temporal, social, economic, task, and technical characteristics. [U. Reiter et al, Factors influencing quality of experience. In Quality of Experience, pp. 55-72. Springer International Publishing, 2014.]

or system.

or system‘s

Page 3: Hossfeld qc man2015_context_monitoring_web

Context Monitoring and QoE Monitoring

mas.wiwi.uni-due.de 3

Context monitoring

QoEmonitoring

e.g. devicecapabilities e.g. video

buffer status

e.g. user expectations

e.g. predicted traffic demands

e.g. available resources

e.g. QoS

utilization of data

Is context monitoring more relevant than QoE monitoring

for managing QoE?

Page 4: Hossfeld qc man2015_context_monitoring_web

Context Factors

• Physical environment in which services and devices are used.– home, office, commuting, and other places, – indoors vs outdoors.

• Social environment– service consumption e.g. alone, with an important person, with a group of friends, or in a

public place (consider gaming, watching video), – popularity of contents.

• Economic context– price for service consumption, tariff model: time, volume, flat– costs

• System context– load of system– system offloading possible, e.g. wifi offloading

• Usage context– Goal, task of service consumption, e.g. information retrieval vs. time killing– background vs. foreground application

mas.wiwi.uni-due.de 4

Examples• Follow the moon:

temporal and economic context

• Video streaming: physical context

• Video flash crowds: social context

Page 5: Hossfeld qc man2015_context_monitoring_web

Agenda

• Context monitoring and QoE monitoring

• Example use case: video flash crowds

• QoE model for HTTP adaptive streaming

• Numerical results

• Open issues: realization

Page 6: Hossfeld qc man2015_context_monitoring_web

SIMULATION MODEL

Page 7: Hossfeld qc man2015_context_monitoring_web

HTTP Adaptive Streaming

mas.wiwi.uni-due.de 7

Page 8: Hossfeld qc man2015_context_monitoring_web

mas.wiwi.uni-due.de 9

Content Delivery with a CDN

Core network

Accessnetwork

Contentserver

ClientsCDNserver

Page 9: Hossfeld qc man2015_context_monitoring_web

Edge Content Delivery Network

mas.wiwi.uni-due.de 10

$$$

Global CDN Backbone$$$

Access Provider

Access Provider

TransitProvider

Point of PresencePoint of

Presence

Point of Presence

Point of Presence

$$$

Edge Cache

Page 10: Hossfeld qc man2015_context_monitoring_web

Simulation Scenario: Video Flash Crowd

• Video player– playout threshold of 6s – video stalls for empty buffer

• Video contents– Segment size of 2s– Two quality layers

• Flash crowd arrivals– users arrive– Exponential distributed interarrival times with rate – P(T<90s) = 99.27%,

• HAS algorithm• CDN load balancing

mas.wiwi.uni-due.de 11

CDN 1

CDN 2

Flash Crowd

ISP

bottleneck

Page 11: Hossfeld qc man2015_context_monitoring_web

HAS Algorithm and CDN Load Balancing

• CDN load balancing strategies1. CDN directs the first users to CDN 1, subsequent users

are assigned to CDN 2. Second, the CDN.2. Context monitoring based on information about the

flash crowd from a third party. Users are assigned to the CDN with the lowest number of users.

• HTTP adaptation strategy1. Actual buffer and throughput of last segment to

determine quality level of next segment2. Additional context information on number of users and

capacity per CDN3. Non-adaptive streaming algorithm: high quality level

mas.wiwi.uni-due.de 12

Bit rate

Time

TCP throughput

Requested chunks

Page 12: Hossfeld qc man2015_context_monitoring_web

SIMPLE QOE MODEL FOR HTTP ADAPTIVE STREAMING

Page 13: Hossfeld qc man2015_context_monitoring_web

What is the influence of stalling on Video QoE?

IQX-Hypothesis

ExcellentGoodFairPoorBad

54321

ImperceptiblePerceptible

Slightly annoyingAnnoying

Very annoying

• Small number of interruptions strongly affect YouTube QoE

Provider (i.e. content and network provider) must avoid stalling

0 1 2 3 4 5 61

2

3

4

5

number of stallings

MO

S

crowdsourcinglaboratory

QoE ( x )=α e−βx+γ

mas.wiwi.uni-due.de 14

Page 14: Hossfeld qc man2015_context_monitoring_web

Survey: Subjective Studies on HAS QoE

• Seufert, M.; Egger, S.; Slanina, M.; Zinner, T.; Hoßfeld, T.; Tran-Gia, P., "A Survey on Quality of Experience of HTTP Adaptive Streaming," Communications Surveys & Tutorials, IEEE , vol.17, no.1, pp.469,492, 2015doi: 10.1109/COMST.2014.2360940

mas.wiwi.uni-due.de 15

HTTP Adaptive

Streaming

Video Quality

Human Computer Interaction

Networking

etc.

Page 15: Hossfeld qc man2015_context_monitoring_web

Switching Frequency vs. Time on Layer

• In several works, switching frequency is reported to influence QoE

• Often parameters „number/frequency of switches“ and „time on layer“ are correlated and change simultaneously

• Keeping „time on layer“ constant no influence of switching frequency could be found

mas.wiwi.uni-due.de 16

Page 16: Hossfeld qc man2015_context_monitoring_web

Simple QoE Model for Two Quality Layers

• Simple QoE model based on two key influence factors• IQX provides a very good fit to the data points (R²=0.98)

mas.wiwi.uni-due.de 17

IQX-Hypothesis

Page 17: Hossfeld qc man2015_context_monitoring_web

Combined QoE Model

• Quality Adaptation Model– Based on time t on high layer– following IQX hypothesis

• Stalling Model– Based on number of stalls– following IQX hypothesis

• HTTP Adaptive Streaming Model

– Model still follows IQX hypothesis

mas.wiwi.uni-due.de 18

IQX-Hypothesis

Page 18: Hossfeld qc man2015_context_monitoring_web

NUMERICAL RESULTS:VIDEO FLASH CROWDS

Page 19: Hossfeld qc man2015_context_monitoring_web

Simulation Results

• No context information is used – CDN load balancing strategy: K=13– HAS quality

adaptation mechanism.

• CDN1 can serve13 / 35 users in high / low quality

• CDN2 can serve10 / 26 users inhigh / low quality

• Reaction too slow

mas.wiwi.uni-due.de 20

Page 20: Hossfeld qc man2015_context_monitoring_web

CDN Load Balancing Strategy

• Static assignment cannot achieve optimum

• Reactive approachbased on contextinformation improvesQoE for all users

mas.wiwi.uni-due.de 21

Page 21: Hossfeld qc man2015_context_monitoring_web

Summary of Results: CDN and HAS

mas.wiwi.uni-due.de 22

Bit rate

Time

TCP throughput

Requested chunks

Page 22: Hossfeld qc man2015_context_monitoring_web

Conclusions

• Context monitoring complements QoE monitoring– Utilization of additional information– Different types of context may be monitored

• Example of video flash crowds– Performance and QoE gain significantly improves– Technical realization needs to be developed

• Realization of context monitoring

mas.wiwi.uni-due.de 23

Page 23: Hossfeld qc man2015_context_monitoring_web

www.mas.wiwi.uni-due.de 24

Realization of Context Monitoring using Social Data

• Accessing data from third party:Internet of Services

• Social data has to be monitored– Scale (single user, selected users,

all users)– Period (every hour, once a day, …)– Source (Online Social Networks OSNs,

Services, Service Providers, ISPs,…)

online social network

http://www.smartenit.eu

Content ProviderISPs

CDNs

$$$

$$$ $$$

$$$

AdsData

analysis

Page 24: Hossfeld qc man2015_context_monitoring_web

www.mas.wiwi.uni-due.de 25

Reseach Qestions: Social Data Monitoring

• Example: How to access data from OSNs?

•Design questions– Identification of relevant social data– Access method– Sampling strategy (scale, period, source,…)– Incentives (if necessary)

Method Information PredictionOSN collaboration All information Global, DetailedEnd user grants access to his data

Private information about end user and shared information about friends

Local, Detailed

Crawling/Sampling Public information Global, Vague

Page 25: Hossfeld qc man2015_context_monitoring_web

THANKS