You are on page 1of 54

Context

Modeling
Comparison through experimentation
General conclusion and perspectives

Towards User-aware Peer-to-Peer Live


Video Streaming Systems

Presented by

Ihsan Ullah
Supervised by
Dr. Guillaume Doyen & Prof. Dominique Gaïti

Environnements de Réseaux Autonomes


Institut Charles Delaunay – UMR 6279
Université de Technologie de Troyes
France

November 3, 2011
Ihsan Ullah User-aware P2P live streaming systems 1/50
Context
Modeling
Comparison through experimentation
General conclusion and perspectives

Outline
1 Context
Introduction
Performance issues
Related work
2 Modeling
Non-contextual approach
Contextual approach
3 Comparison through experimentation
Optimization for improved quality
Experimental evaluation
4 General conclusion and perspectives

Ihsan Ullah User-aware P2P live streaming systems 2/50


Context
Introduction
Modeling
Performance issues
Comparison through experimentation
Related work
General conclusion and perspectives

Outline
1 Context
Introduction
Performance issues
Related work
2 Modeling
Non-contextual approach
Contextual approach
3 Comparison through experimentation
Optimization for improved quality
Experimental evaluation
4 General conclusion and perspectives

Ihsan Ullah User-aware P2P live streaming systems 3/50


Context
Introduction
Modeling
Performance issues
Comparison through experimentation
Related work
General conclusion and perspectives

Video traffic trends


Due to the availability of broadband technologies and
more powerful personal computers, video traffic over
the Internet has enormously increased

40% 50% 61%


Source: Cisco Visual Networking Index (VNI)

Ihsan Ullah User-aware P2P live streaming systems 4/50


Context
Introduction
Modeling
Performance issues
Comparison through experimentation
Related work
General conclusion and perspectives

Video streaming
Streaming is the transfer of media such as audio or video
over a network as a steady and continuous stream [HB05]
Video-on-Demand (VoD)
Receiver-driven
Users can request for any video, any time
Extended buffering

Live video streaming


Source-driven
Broadcasts the newly generated content
Limited buffering
Ihsan Ullah User-aware P2P live streaming systems 5/50
Context
Introduction
Modeling
Performance issues
Comparison through experimentation
Related work
General conclusion and perspectives

Video streaming architectures


Video streaming

With router support


With Without
IP multicast [DC90] router support router support

Scalability, security and


lack of a business model Without
end-systems support
With
end-systems support

Telco-managed IPTV Telco-managed


IPTV

uses IP multicast Client/Server CDN-based P2P


Limited to private Decentralization level

networks and expensive [LRLZ08]

Without router support


Without end-systems support (Centralized C/S, CDN)
Scalability and cost (Servers and upload bandwidth)
With end-systems support (P2P)

Ihsan Ullah User-aware P2P live streaming systems 6/50


Context
Introduction
Modeling
Performance issues
Comparison through experimentation
Related work
General conclusion and perspectives

P2P live video streaming

End-hosts (peers) form a


self-organizing network
Peers share their upload
bandwidth through relaying
content to each other
Uses the existing IP
infrastructure
Easy to deploy with low cost
Programmable end-hosts
Potentially scalable

Ihsan Ullah User-aware P2P live streaming systems 7/50


Context
Introduction
Modeling
Performance issues
Comparison through experimentation
Related work
General conclusion and perspectives

P2P live video delivery methods


Push-based Pull-based Hybrid push/pull

Push Pull

Ihsan Ullah User-aware P2P live streaming systems 8/50


Context
Introduction
Modeling
Performance issues
Comparison through experimentation
Related work
General conclusion and perspectives

Problem statement
Impact of user behavior
Peers relay content, which are controlled by users
Users join and leave freely (churn), which result into:
Stream disruption;
Streaming quality degradation and
Buffer underflow

Churn-resilient structures and large buffers


Mitigate churn but lead to longer startup, switching
and playback delays
Streaming quality perceived by users is not
comparable to that of television
Ihsan Ullah User-aware P2P live streaming systems 9/50
Context
Introduction
Modeling
Performance issues
Comparison through experimentation
Related work
General conclusion and perspectives

Global models

Current approaches consider user behavior globally


at the systems scale
Older-stable principle [WLX08, TSLZ06, WXL07]
Observed distributions (Lognormal, Exponential,
Weibull, Pareto)
Capture system-level behavior patterns but cannot
enable to anticipate the behavior of an individual user

Ihsan Ullah User-aware P2P live streaming systems 10/50


Context
Introduction
Modeling
Performance issues
Comparison through experimentation
Related work
General conclusion and perspectives

Incomplete models

Current models of user behavior metrics do not


consider all of their influential factors
Works [WLX08, TSLZ06, WXL07] consider elapsed
time as the only influential factor of stability
[VAM+ 06] specify the arrival process of users as a
Poisson distribution
Give general insights but cannot enable an accurate
estimation of these metrics

Ihsan Ullah User-aware P2P live streaming systems 11/50


Context
Modeling Non-contextual approach
Comparison through experimentation Contextual approach
General conclusion and perspectives

Outline
1 Context
Introduction
Performance issues
Related work
2 Modeling
Non-contextual approach
Contextual approach
3 Comparison through experimentation
Optimization for improved quality
Experimental evaluation
4 General conclusion and perspectives

Ihsan Ullah User-aware P2P live streaming systems 12/50


Context
Modeling Non-contextual approach
Comparison through experimentation Contextual approach
General conclusion and perspectives

Hypotheses

Current session duration can be estimated from a


history of past ones
Based on surveys over viewing habits of regular
television [WSvM09, RG97, BEG82], "A past viewing
behavior is a strong predictor of the current one"
Each peer estimates its own session and provides
this information to others on request
We assume a pure push-based single tree system
since it is highly churn-sensitive

Ihsan Ullah User-aware P2P live streaming systems 13/50


Context
Modeling Non-contextual approach
Comparison through experimentation Contextual approach
General conclusion and perspectives

Exponential Moving Average (EMA)

Averages previous sessions by giving exponentially


decreasing weights to older ones
The length of the current session ESt can be
estimated through (1)

ESt = α × St−1 + (1 − α) × ESt−1 (1)

Aimed at the anticipation of user departures


To favor successful anticipations, we reduce each
estimated session by 20% (EMA20 )
Chosen through testing over different values

Ihsan Ullah User-aware P2P live streaming systems 14/50


Context
Modeling Non-contextual approach
Comparison through experimentation Contextual approach
General conclusion and perspectives

Bayes’ Rule (BR)


Discretization of T into {t1 , t2 ...tk } such that
t1 < t2 ...tk − 1 < tk
Estimates P(Scur > ti ) from observations of past
sessions for each ti
αj + |oj | + 1
f (φj |O, O ′ ) = (2)
|O| + |O ′ | + 2
The estimated current session duration is the
maximum time slot for which the probability is greater
than or equal to threshold Pth ,
t ∗ = max(f (φj ) > Pth ) (3)
tj

Ihsan Ullah User-aware P2P live streaming systems 15/50


Context
Modeling Non-contextual approach
Comparison through experimentation Contextual approach
General conclusion and perspectives

A receiver-driven approach
Start

A
Join/Schedule
move

Continue

B F
When the scheduled time is up C

Make a list of stable


neighbors

E D G
Attempt a move to NO
the most stable

Remove from List is


success? No
the list empty?

Yes H I
Leave the Yes
previous

End

Ihsan Ullah User-aware P2P live streaming systems 16/50


Context
Modeling Non-contextual approach
Comparison through experimentation Contextual approach
General conclusion and perspectives

Experimental framework
Simulated behaviors
Dynamic: Uncorrelated consecutive sessions
(Pearson correlation coefficient ρ = 0.002)
Persistent: Strongly correlated consecutive sessions
(Pearson correlation coefficient ρ = 0.8)

12 computers in a LAN environment

Scribe: Push-based system

Bitrate: 256 Kbps

Arrival process: Poisson distribution (modified)


[YZZZ06]
Ihsan Ullah User-aware P2P live streaming systems 17/50
Context
Modeling Non-contextual approach
Comparison through experimentation Contextual approach
General conclusion and perspectives

Measured parameters
Frame losses
Video frames lost due to abrupt departures
Shows the efficiency of the approach in terms of
reducing the stream disruption
Attempted moves
A move is attempted to switch from the current parent
to a new one
Shows the communication cost
Successful moves
The actual switching of a peer from its current parent
to a new one
Shows the efficacy of the optimization strategy

Ihsan Ullah User-aware P2P live streaming systems 18/50


Context
Modeling Non-contextual approach
Comparison through experimentation Contextual approach
General conclusion and perspectives

Experimental results
Dynamic case
x 10
5 Frame loss Attempted moves Successful moves
500 500
Total number of lost frames

4 EMA

Total successful moves


Total attempted moves
20
400 400 Bayes
3
300 300

2
200 200
Scribe
1 EMA20 100 EMA 100
20
Bayes Bayes
0 0 0
0 1000 2000 3000 0 1000 2000 3000 0 1000 2000 3000
Time in Seconds Time in Seconds Time in Seconds

Persistent case
x 10
5 Frame loss Attempted moves Successful moves
500 500
Total number of lost frames

Total successful moves


4 EMA
20
400
Attempted moves

400 Bayes
3
300 300

2
200 200
Scribe
1 EMA 100 EMA 100
20 20
Bayes Bayes
0 0 0
0 1000 2000 3000 0 1000 2000 3000 0 1000 2000 3000
Time in Seconds Time in Seconds Time in Seconds

Ihsan Ullah User-aware P2P live streaming systems 19/50


Context
Modeling Non-contextual approach
Comparison through experimentation Contextual approach
General conclusion and perspectives

Conclusion and limits

Conclusion
Both approaches reduce frame losses which improve
the quality of streaming
Communication cost of the receiver-driven strategy is
small

Limits
A large number of attempted moves have not taken
place due to the limitation of receiver-driven strategy
Only two kinds of user behavior have been
considered over which estimators are not consistent

Ihsan Ullah User-aware P2P live streaming systems 20/50


Context
Modeling Non-contextual approach
Comparison through experimentation Contextual approach
General conclusion and perspectives

Outline
1 Context
Introduction
Performance issues
Related work
2 Modeling
Non-contextual approach
Contextual approach
3 Comparison through experimentation
Optimization for improved quality
Experimental evaluation
4 General conclusion and perspectives

Ihsan Ullah User-aware P2P live streaming systems 21/50


Context
Modeling Non-contextual approach
Comparison through experimentation Contextual approach
General conclusion and perspectives

Overview of measurements
Metrics
Reference Type System Period Method
OD SD CP A/D D/U
[SGGS09] Feb. to Nov. 2008 X X X
[VGLN07] Apr. to Dec. 2006 Crawler X X
[WLX08] PPLive Nov. 2006 (about 28 hours) X X
[HLL+ 07] 2006 − 07 Passive/ X X X X
Crawler
[SFB+ 09] P2P Jun. 2006 Passive X X
[JLC07] PPStream unknown Crawler X X X
[SMC+ 09] Zattoo Mar. 2008 (2 weaks) X X
[ZLL05] Mar. 2005 (4 days) X X X
Cool-
[XKL07] Oct. 2006 (1 day) X X X
Streaming
[LXK+ 07] Sep. 2006 (1 day) X X X X
[Aga07] unknown 2006 (11 hours) X X
[TSZ+ 07] GridMedia Jan. 2006 (4.5 hours) X
[LWLZ09] UUSee May to Jun. 2008 (5 days) Logs X X X
[TSLZ06] Feb. 2005/Jan. 2006 X
CCTV
[TSLZ06] C/S Oct. to Jan. 2004 − 05 X
[SMZ04] CDN Akamai Oct. to Jan. 2003 − 04 X X X
[VAM+ 06] C/S unknown 2002 (90 days) X X X
[QGL+ 09a] Apr. 2008 (6 days) X X X X
Telco-
[QGL+ 09b] IPTV
Managed Jun. 2008 (1 month) X
[CRC+ 08] May to Oct. 2007 X X X

OD: Online Duration; SD: Session Duration; CP: Channel Popularity; A/D: Arrival/Departure rates; D/U:
Download/Upload traffic

Ihsan Ullah User-aware P2P live streaming systems 22/50


Context
Modeling Non-contextual approach
Comparison through experimentation Contextual approach
General conclusion and perspectives

Characterization
Popularity Session duration
0
10 1

0.9
Users’ requests distribution

CDF of session durations


−1
0.8
10
0.7

0.6
−2
10 0.5

0.4

0.3
−3
10 Silverstone et al.
Qiu et al. 0.2 Vu et al.
Sripanidkulchai et al. Tang et al.
0.1
Qiu et al. Veloso et al.
−4
10 0
0 1 2 0 1 2 3 4
10 10 10 10 10 10 10 10
Rank (from higher to lower) Session durations in minutes

Measurements characterize user behavior patterns


Provide insights at the system scale

Ihsan Ullah User-aware P2P live streaming systems 23/50


Context
Modeling Non-contextual approach
Comparison through experimentation Contextual approach
General conclusion and perspectives

Measurements’ synthesis

Environmental impacting Environment

factors: Variables in the Time-of-Day Day-of-Week Content Type

context that impact user


behavior metrics
Surfing Session Arrival Departure
Probability Duration rate rate
User behavior metrics: User

Variables that represent Elapsed


Time
Popularity
Bandwidth
Contribution
Failure
Rate
Ratio
user activities

Network performance
parameters: Variables Partners discovery Streaming Quality Delay

having relationships with Network

user behavior metrics


Ihsan Ullah User-aware P2P live streaming systems 24/50
Context
Modeling Non-contextual approach
Comparison through experimentation Contextual approach
General conclusion and perspectives

Information domain

Environment

Time-of-Day Day-of-Week Content Type


Several variables with
dependency relationships
Surfing Session Arrival
Estimation of any variable Probability Duration rate
Departure
rate

can be required Elapsed Bandwidth Failure


User

Popularity Contribution Rate


Time
Ratio

Modeling these variables


and their relationships to
enable probabilistic Partners discovery Streaming Quality Delay

estimations Network

Ihsan Ullah User-aware P2P live streaming systems 25/50


Context
Modeling Non-contextual approach
Comparison through experimentation Contextual approach
General conclusion and perspectives

Bayesian networks
Bayesian networks are graphical models that encode
probabilistic relationships among variables of interest
[Hec95]
Simplify the joint distribution through conditional
independence

Y
n
p(x) = p(xi |pai ) (4)
i=1
Naturally model the relationships of cause and effect
Allow multiway inference
Support mixed variables
Graphically represent the real world problem domain
Ihsan Ullah User-aware P2P live streaming systems 26/50
Context
Modeling Non-contextual approach
Comparison through experimentation Contextual approach
General conclusion and perspectives

Model derivation from the causal graph

Relationships with Environment

Time-of-Day Day-of-Week Content Type


contradictory findings are
omitted (Day-of-week)
Architecture-dependent Surfing Session Arrival Departure
Probability Duration rate
variables are omitted rate

User

(partners discovery) Elapsed


Time
Popularity
Bandwidth
Contribution
Failure
Rate
Ratio

Similar variables that can


be represented by one
variable are merged Partners discovery Streaming Quality Delay

(surfing probability, failure


Network
rates, session duration)

Ihsan Ullah User-aware P2P live streaming systems 27/50


Context
Modeling Non-contextual approach
Comparison through experimentation Contextual approach
General conclusion and perspectives

Our Bayesian network model

A mixed model of discrete Time-of-day Content type Elapsed time

and continuous variables


All continuous variables
are gaussian distributed Arrival
rate Session
Delay
duration

Content type has 3 states


(reality, fiction and sport)
Elapsed time has 2 states Departure
rate
Popularity
Bandwidth
contribution

(surfing, viewing) ratio

Time-of-day has 24 states


(chosen through Streaming
quality

simulations)

Ihsan Ullah User-aware P2P live streaming systems 28/50


Context
Modeling Non-contextual approach
Comparison through experimentation Contextual approach
General conclusion and perspectives

Simulation scenarios
GlobalBN

Global scenario:
One network for the whole
community Peer Peer

Trained over combined traces


Centralized estimations Peer Peer

Local scenario:
A dedicated network for each Peer
LocalBN
Peer
LocalBN
peer
Trained over individual traces
Local estimations LocalBN
Peer Peer
LocalBN
Ihsan Ullah User-aware P2P live streaming systems 29/50
Context
Modeling Non-contextual approach
Comparison through experimentation Contextual approach
General conclusion and perspectives

Simulation parameters

Bayes Net Toolbox (BNT) for Matlab

Traces generation for six kinds of fictional characters


(Johnatan, Emma, Stephan, Anna, Peter and ELlen)
through a semi-Markovian model [BUD+ 11]

Total population of 1000 users

Simulated duration in the global scenario: 40 days

Simulated duration in the local scenario: 200 days

Content duration: 2 hours

Ihsan Ullah User-aware P2P live streaming systems 30/50


Context
Modeling Non-contextual approach
Comparison through experimentation Contextual approach
General conclusion and perspectives

Results in the global scenario


25 50

45

Estimated session durations


20 40
RMSE over a day

35

15 30

25

10 20

15

5 10

0 0
5 10 15 20 25 30 35 40 0 20 40 60 80 100
Number of days Actual session durations

Turns into a global model


Due to mixed traces an average behavior is learned
The overall accuracy is not good
Ihsan Ullah User-aware P2P live streaming systems 31/50
Context
Modeling Non-contextual approach
Comparison through experimentation Contextual approach
General conclusion and perspectives

Results in the local scenario


70

Estimated session durations


RMSE over ten days periord

40 J
E 60
35
S
30 A 50
P
25 L 40

20
30
15
20
10
10
5

0 0
20 40 60 80 100 120 140 160 180 0 10 20 30 40 50 60
Time in days Actual session durations

Better performance
Consistent over different behaviors
Shows good accuracy

Long learning time


The network takes about 40 days to learn the behavior accurately

Ihsan Ullah User-aware P2P live streaming systems 32/50


Context
Modeling Non-contextual approach
Comparison through experimentation Contextual approach
General conclusion and perspectives

Learning time minimization

Training of a separate BN Traces

for each class of users Datasets

A classifier of users DS 1 , DS 2 , . . ., DS k

according to their
behaviors Training classifier Training BNs

Assigning of the Trained BN 1 ,BN 2 , . . ., BN


Classifier k
corresponding BN to each
user
Using the assigned BN for Classification Assigning BN

estimations User

Ihsan Ullah User-aware P2P live streaming systems 33/50


Context
Modeling Non-contextual approach
Comparison through experimentation Contextual approach
General conclusion and perspectives

User classifier

A Bayesian network Time-of-day Popularity

classifier
Based on the viewing
durations of users
Session
User Class
Includes all the related Duration

variables
Estimation of the user
class variable gives the Streaming
Content Type
Quality
required class

Ihsan Ullah User-aware P2P live streaming systems 34/50


Context
Modeling Non-contextual approach
Comparison through experimentation Contextual approach
General conclusion and perspectives

Classification error
4
x 10
15
Incorrect
Correct

Number of classifications
10

0
J E S A P L
User classes

For one half of the personas the error remains


between 8% and 9%
For the remaining half of the personas the varies
between 15% and 24%
Ihsan Ullah User-aware P2P live streaming systems 35/50
Context
Modeling Non-contextual approach
Comparison through experimentation Contextual approach
General conclusion and perspectives

Considering the magnitude of probability


14 60

Unknown classifications (in %)


12 50

10
Percent error

40
8
30
6
20
4

2 10

0 0
0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
Threshold Threshold

Cannot classify all users


Can only be used when the classification of each user is
not mandatory
Ihsan Ullah User-aware P2P live streaming systems 36/50
Context
Modeling Non-contextual approach
Comparison through experimentation Contextual approach
General conclusion and perspectives

Considering the classification history


2
10

1
10
Percent error

0
10

−1
10

−2
10
1 5 10 15 20 25
History size

Takes into account the previous classifications


With about 5 observations the error reduces up to a
negligible level
Ihsan Ullah User-aware P2P live streaming systems 37/50
Context
Modeling Non-contextual approach
Comparison through experimentation Contextual approach
General conclusion and perspectives

Conclusion

Our Bayesian network model is suitable for the


learning and estimation of peer-level variables in the
local scenario

Global scenario is useful for the estimation of


system-level variables

User classifier reduces this time significantly to one


or a few observations

With the classifier both scenarios can be used

Ihsan Ullah User-aware P2P live streaming systems 38/50


Context
Modeling Optimization for improved quality
Comparison through experimentation Experimental evaluation
General conclusion and perspectives

Outline
1 Context
Introduction
Performance issues
Related work
2 Modeling
Non-contextual approach
Contextual approach
3 Comparison through experimentation
Optimization for improved quality
Experimental evaluation
4 General conclusion and perspectives

Ihsan Ullah User-aware P2P live streaming systems 39/50


Context
Modeling Optimization for improved quality
Comparison through experimentation Experimental evaluation
General conclusion and perspectives

Sender-driven stabilization strategy

Issue with the receiver-driven strategy


Peers struggle to find stable neighbors
A large number of unsuccessful moves

New strategy
Instead of moving the receiver, the sender is moved
Pushes the unstable peers towards the outskirts of
the tree
Works in a decentralized way

Ihsan Ullah User-aware P2P live streaming systems 40/50


Context
Modeling Optimization for improved quality
Comparison through experimentation Experimental evaluation
General conclusion and perspectives

Swap
Start

A A
Join/Schedule
swap

Continue
B C E
B
When the scheduled time is up

Choose stable +list

D E C
D
inform other
children (list)

Swap with stable


child

accepted Remove from List is Request the


No No
as a child? the list empty? next peer

Yes
Yes
Add parent

Random join

End

Ihsan Ullah User-aware P2P live streaming systems 41/50


Context
Modeling Optimization for improved quality
Comparison through experimentation Experimental evaluation
General conclusion and perspectives

Experimental environment

PlanetLab: A globally distributed testbed for


networking and distributed systems’ research
Scribe: Push-based single tree system
BNs and classifier trained over 60 days of traces
2 hours traces used for experimentation from day 61
Local scenario
60 PlanetLab nodes
Bitrate: 64 Kbps

Ihsan Ullah User-aware P2P live streaming systems 42/50


Context
Modeling Optimization for improved quality
Comparison through experimentation Experimental evaluation
General conclusion and perspectives

Peers’ population throughout each


experiment
100
90
80
Number of online peers

70
60
50
40
30
WS
20 BN
EMA
10 20
BR
0
0 20 40 60 80 100 120
Time in minutes

Population shows similar pattern during all


experiments showing similar experimental conditions
Ihsan Ullah User-aware P2P live streaming systems 43/50
Context
Modeling Optimization for improved quality
Comparison through experimentation Experimental evaluation
General conclusion and perspectives

Frame losses
5
x 10
70 12
WS WS
BN BN

Cumulative missed packets


60
Missed packets ratio (%)

EMA 10 EMA
20 20
BR BR
50
8
40
6
30
4
20

10 2

0 0
0 20 40 60 80 100 120 0 20 40 60 80 100 120
Time in minutes Time in minutes

Consideration of user behavior reduces frame losses


Contextual approach shows better results than non
contextual one

Ihsan Ullah User-aware P2P live streaming systems 44/50


Context
Modeling Optimization for improved quality
Comparison through experimentation Experimental evaluation
General conclusion and perspectives

Communication cost
350 2000
BN BN
EMA 1800 EMA

Cumulative control messages


Number of control messages

300 BR BR
1600
250 1400
1200
200
1000
150
800

100 600
400
50
200
0 0
0 20 40 60 80 100 120 0 20 40 60 80 100 120
Time in minutes Time in minutes

BN produces a larger number of messages than


other approaches
Overall it incurs a small overhead (average 0.223
messages per minute by a peer)
Ihsan Ullah User-aware P2P live streaming systems 45/50
Context
Modeling Closing remarks
Comparison through experimentation Perspectives
General conclusion and perspectives

Outline
1 Context
Introduction
Performance issues
Related work
2 Modeling
Non-contextual approach
Contextual approach
3 Comparison through experimentation
Optimization for improved quality
Experimental evaluation
4 General conclusion and perspectives

Ihsan Ullah User-aware P2P live streaming systems 46/50


Context
Modeling Closing remarks
Comparison through experimentation Perspectives
General conclusion and perspectives

Summary of the work


Synthesis of user behavior measurements and
extraction of the causal graph

Two kinds of models (contextual and non-contextual)

Users classifier to minimize the learning time of the


contextual model

Two stabilization strategies (receiver-driven,


sender-driven)

Comparison of the two approaches through


simulations and experimentations
Ihsan Ullah User-aware P2P live streaming systems 47/50
Context
Modeling Closing remarks
Comparison through experimentation Perspectives
General conclusion and perspectives

Conclusion and limits


Conclusion
User-awareness reduces stream disruption
Other performance metrics (delay, streaming quality)
are also improved
Contextual approach appears to be more accurate
than non-contextual one
Sender-driven (swap) stabilization strategy is more
effective

Limits
No support for dynamic behaviors
No consideration of malicious behavior
Ihsan Ullah User-aware P2P live streaming systems 48/50
Context
Modeling Closing remarks
Comparison through experimentation Perspectives
General conclusion and perspectives

Future directions
Short term future work
Inclusion of other metrics with stability
Modification to predict next channel
Feedback mechanism for dynamic behaviors
Extension of the semi-Markovian model

Long term future work


Investigation of user-oriented topologies
Consideration of malicious behavior
Application to resource allocation in C/S and CDN
systems
Ihsan Ullah User-aware P2P live streaming systems 49/50
References

Thesis publications
International Journals
I. Ullah, G. Doyen, G. Bonnet and D. Gaïti, “A Survey and Synthesis of User Behavior Measurements in
Video Streaming Systems”, IEEE Communications Surveys & Tutorials, 2011.

I. Ullah, G. Doyen, G. Bonnet and D. Gaïti, “A Bayesian approach for user aware peer-to-peer video
streaming systems”, Elsevier Journal of Signal Processing: Image Communication (to appear), 2011.

International Conferences
I. Ullah, G. Doyen, G. Bonnet and D. Gaïti, “User behavior anticipation in P2P live video streaming systems
through a Bayesian network”, in: 12th IFIP/IEEE International Symposium on Integrated Network
Management (IM). IEEE, 2011, pp. 337-344.

G. Bonnet, I. Ullah, G. Doyen, L. Fillatre, D. Gaïti and I. Nikiforov, “A Semi-Markovian individual model of
users for P2P video streaming applications”, in: 4th IFIP International Conference on New Technologies,
Mobility and Security (NTMS). IEEE, 2011, pp. 1-5.

I. Ullah, G. Doyen, G. Bonnet and D. Gaïti, “Modeling User Behavior in P2P Live Video Streaming Systems
through a Bayesian Network”, in: 4th IFIP International Conference on Autonomous Infrastructure,
Management and Security, (AIMS), Springer, 2-13, 2010.

I. Ullah, G. Doyen, G. Bonnet and D. Gaïti, “Improving Performance of ALM Systems with Bayesian
Estimation of Peers Dynamics”, in: 12th IFIP/IEEE International Conference on Management of Multimedia
and Mobile Networks and Services, (MMNS). Springer, 2009, pp. 157-169.

Ihsan Ullah User-aware P2P live streaming systems 50/50


References

Sachin Agarwal.
A case study of large scale P2P video multicast.
In International Conference on IP Multimedia Subsystem Architecture and Applications (IMSAA), pages 1–5.
IEEE, 2007.

T. P. Barwise, A. S. C. Ehrenberg, and G. J. Goodhardt.


Glued to the Box?: Patterns of TV Repeat-Viewing.
Journal of Communication, 32(4):22–29, 1982.

Grégory Bonnet, Ihsan Ullah, Guillaume Doyen, Lionel Fillatre, Dominique Gaïti, and Igor Nikiforov.
A Semi-Markovian individual model of users for P2P video streaming applications.
In 4th IFIP International Conference on New Technologies, Mobility and Security (NTMS), pages 1–5. IEEE,
2011.
Meeyoung Cha, Pablo Rodriguez, Jon Crowcroft, Sue Moon, and Xavier Amatriain.
Watching television over an IP network.
In Internet Measurement Conference (IMC), pages 71–84. ACM, 2008.

Stephen E. Deering and David R. Cheriton.


Multicast routing in datagram internetworks and extended LANs.
ACM Trans. Comput. Syst., 8(2):85–110, 1990.

Markus Hofmann and Leland R. Beaumont.


Content Networking: Architecture, Protocols, and Practice.
Morgan Kaufmann Publishers Inc., San Francisco, CA, USA, 2005.

David Heckerman.
A tutorial on learning with Bayesian networks.
Technical report, Learning in Graphical Models, 1995.

Ihsan Ullah User-aware P2P live streaming systems 50/50


References

Xiaojun Hei, Chao Liang, Jian Liang, Yong Liu, and Keith W. Ross.
A measurement study of a large-scale P2P IPTV system.
IEEE Transactions on Multimedia, 9(8):1672–1687, 2007.

Jinkang Jia, Chunxi Li, and Changjia Chen.


Characterizing PPStream across Internet.
In IFIP International Conference on Network and Parallel Computing Workshops, pages 413–418, 2007.

Jiangchuan Liu, Sanjay G. Rao, Bo Li, and Hui Zhang.


Opportunities and Challenges of Peer-to-Peer Internet Video Broadcast.
Proceedings of the IEEE, 96(1):11–24, 2008.

Zimu Liu, Chuan Wu, Baochun Li, and Shuqiao Zhao.


Distilling superior peers in large-scale P2P streaming systems.
In IEEE INFOCOM, pages 82–90, 2009.

Bo Li, Susu Xie, Gabriel Yik Keung, Jiangchuan Liu, Ion Stoica, Hui Zhang, and Xinyan Zhang.
An empirical study of the Coolstreaming+ system.
IEEE Journal on Selected Areas in Communications, 25(9):1627–1639, 2007.

Tongqing Qiu, Zihui Ge, Seungjoon Lee, Jia Wang, Jun Xu, and Qi Zhao.
Modeling user activities in a large IPTV system.
In Internet Measurement Conference (IMC), pages 430–441. ACM, 2009.

Tongqing Qiu, Zihui Ge, Seungjoon Lee, Jia Wang, Qi Zhao, and Jun Xu.
Modeling channel popularity dynamics in a large IPTV system.
In 11th international joint conference on Measurement and Modeling of Computer Systems (SIGMETRICS),
pages 275–286, 2009.

Aviva W. Rosenstein and August E. Grant.

Ihsan Ullah User-aware P2P live streaming systems 50/50


References

Reconceptualizing the role of habit: A new model of television audience activity.


Journal of Broadcasting & Electronic Media, 41:324–344, 1997.

Thomas Silverston, Olivier Fourmaux, Alessio Botta, Alberto Dainotti, Antonio Pescapè, Giorgio Ventre, and
Kavé Salamatian.
Traffic analysis of Peer-to-Peer IPTV communities.
Computer Networks, 53(4):470–484, 2009.

Salvatore Spoto, Rossano Gaeta, Marco Grangetto, and Matteo Sereno.


Analysis of PPLive through active and passive measurements.
In IEEE International Symposium on Parallel & Distributed Processing, pages 1–7, 2009.

Khaldoon Shami, Damien Magoni, Hyunseok Chang, Wenjie Wang, and Sugih Jamin.
Impacts of peer characteristics on P2PTV networks scalability.
In IEEE INFOCOM, pages 2736–2740, 2009.

Kunwadee Sripanidkulchai, Bruce Maggs, and Hui Zhang.


An analysis of live streaming workloads on the Internet.
In Internet Measurement Conference (IMC), pages 41–54. ACM, 2004.

Yun Tang, Lifeng Sun, Jian-Guang Luo, and Yuzhuo Zhong.


Characterizing user behavior to improve quality of streaming service over P2P networks.
In Advances in Multimedia Information Processing (PCM), pages 175–184, 2006.

Yun Tang, Lifeng Sun, Kaiyun Zhang, Shi-Qiang Yang, and Yuzhuo Zhong.
Longer, better: On extending user online duration to improve quality of streaming service in P2P networks.
In IEEE International Conference on Multimedia and Expo (ICME), pages 2158–2161, 2007.

Eveline Veloso, Virgílio Almeida, Wagner Jr. Meira, Azer Bestavros, and Shudong Jin.
A hierarchical characterization of a live streaming media workload.

Ihsan Ullah User-aware P2P live streaming systems 50/50


References

IEEE/ACM Transactions on Networking, 14(1):133–146, 2006.

Long Vu, Indranil Gupta, Jin Liang, and Klara Nahrstedt.


Measurement and modeling of a large-scale overlay for multimedia streaming.
In The Fourth International Conference on Heterogeneous Networking for Quality, Reliability, Security and
Robustness & Workshops (QSHINE), pages 1–7, 2007.

Feng Wang, Jiangchuan Liu, and Yongqiang Xiong.


Stable peers: Existence, importance, and application in Peer-to-Peer live video streaming.
In IEEE INFOCOM, pages 1364–1372, 2008.

Anke Wonneberge, Klaus Schoenbach, and Lex van Meurs.


Dynamics of individual television viewing behavior: Models, empirical evidence, and a research program.
Communication Studies, 60(3):235 – 252, 2009.

Feng Wang, Yongqiang Xiong, and Jiangchuan Liu.


mTreebone: A hybrid tree/mesh overlay for application-layer live video multicast.
In 27th International Conference on Distributed Computing Systems, page 49, 2007.

Susu Xie, Gabriel Y. Keung, and Bo Li.


A measurement of a large-scale Peer-to-Peer live video streaming system.
In International Conference on Parallel Processing Workshops (ICPP), pages 57–62, 2007.

Hongliang Yu, Dongdong Zheng, Ben Y. Zhao, and Weimin Zheng.


Understanding user behavior in large-scale video-on-demand systems.
ACM SIGOPS Operating Systems Review, 40(4):333–344, 2006.

Xinyan Zhang, Jiangchuan Liu, and Bo Li.


On large scale Peer-to-Peer live video distribution: Coolstreaming and its prelimianry experimental results.
In IEEE International Workshop on Multimedia Signal Processing, 2005.

Ihsan Ullah User-aware P2P live streaming systems 50/50

You might also like