Modelling IPTV Services

Modelling IPTV Services Fernando Ramos*

[email protected]

*Together with Jon Crowcroft, Ian White, Richard Gibbens, Fei Song, P. Rodriguez

Outline

  Introduction: what IPTV is not, and why do we care

  Motivation to model IPTV services

  The IPTV traffic model, in some detail (W.I.P.)

  Conclusions

9 July 2009 Fernando Ramos, Cosener's Multi-Service Networks

What IPTV is not

live TV

web TV

p2p TV

on-demand video service cable-like TV service

IPTV is a cable-like TV service offered on top of an IP network

X X X

9 July 2009 Fernando Ramos, Cosener's Multi-Service Networks

Why do we care with IPTV?

  One of the fastest growing television services in the world [1]   2005: 2 million users   2007: 14 million users   ...and growing

  High bandwidth and strict QoS requirements   Big impact in the IP network

[1]ParksAssociates.TvservicesinEurope:UpdateandOutlook,2008

9 July 2009 Fernando Ramos, Cosener's Multi-Service Networks

Overview of an IPTV network

STB

PC

TV

DSLAM

Customer Premises

Internet IP Network

TV Head End

.

.

.

9 July 2009 Fernando Ramos, Cosener's Multi-Service Networks

Motivation – Why do we need a realistic IPTV Traffic Model?

  Brand new service on top of an IP network   User behaviour very different from other IP-based

applications

9 July 2009 Fernando Ramos, Cosener's Multi-Service Networks

Motivation – Why do we need a realistic IPTV Traffic Model?

  To evaluate different delivery systems for IPTV

  To evaluate different network architectures for IPTV

9 July 2009 Fernando Ramos, Cosener's Multi-Service Networks

The dataset   We have analysed real IPTV data from one of the largest

IPTV service providers   ~ 6 months worth of data   ~ 250,000 customers   ~ 620 DSLAMs   ~ 150 TV channels

  NB: We consider a user is zapping if he switches between 2 TV channels in less than 1 minute.

9 July 2009 Fernando Ramos, Cosener's Multi-Service Networks

IPTV Traffic Model   Workload characteristics

  Zapping blocks containing a random number of switching events (zapping period)

  Separated by watching/away periods of random length

9 July 2009 Fernando Ramos, Cosener's Multi-Service Networks

0

5

10

15

20

25

30

06:00 07:12 08:24 09:36 10:48 12:00 13:12 14:24 15:36

Swit

chin

g E

vent

Num

ber

Time of the day

Zap blocks

Inter-zap block periods

IPTV Traffic Model

9 July 2009 Fernando Ramos, Cosener's Multi-Service Networks

WATCHING MODE

ZAPPING MODE

IPTV Traffic Model - Detailed

9 July 2009 Fernando Ramos, Cosener's Multi-Service Networks

Findings: Empirical data fits with 2 gamma and 1 exponential (consistent across regions) To do: Check consistency for different channels Check consistency for period of the day

0.1

1

1 10 100 1000

Pro

port

ion

(CD

F)

Inter zap block interval (minutes) [100 DSLAMs, 40k users]

Empirical data

Gamma CDF

Exponential CDF

Gamma 2 CDF

IPTV Traffic Model - Detailed

9 July 2009 Fernando Ramos, Cosener's Multi-Service Networks

Findings: Empirical data fits with gamma distribution (consistent across regions) To do: Check consistency for period of the day

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

1 10

Pro

port

ion

(CD

F)

Zap block size (log) [100 DSLAMs, 40k users]

Empirical data Gamma CDF Geommetric CDF Poisson CDF

IPTV Traffic Model - Detailed

9 July 2009 Fernando Ramos, Cosener's Multi-Service Networks

Findings: Popularity is a) Zipf-like for top channels, b) decays abruptly for non-popular ones. To do: Add dependency of previous channel.

1.E-06

1.E-05

1.E-04

1.E-03

1.E-02

1.E-01

1.E+00

1 10 100

Nor

mal

ised

cha

nnel

acc

ess

(log

)

Channel index sorted by popularity (log) [49 DSLAMs, 16k users]

Normalised watching counter

Zipf-like distribution

Conclusions   Preliminary results of an IPTV Workload model were

presented   Some of the main findings:

  Workload characteristics: Burst (zapping) periods separated by watch/way periods

  Popularity: a) Zipf-like for top channels, b) decays fast for non-popular ones

  Watching period empirical data fits with 2 gamma and 1 exponential distributions

  Number of channels in a zap period fits with gamma distribution

  See you at the SIGCOMM Poster Session!

9 July 2009 Fernando Ramos, Cosener's Multi-Service Networks

[email protected]

9 July 2009 Fernando Ramos, Cosener's Multi-Service Networks

THANK YOU!