Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.
Watching Television  Over an IP Network Meeyoung Cha MPI-SWS Pablo Rodriguez Telefonica Research Jon Crowcroft U. of Cambr...
Internet TV (IPTV) <ul><li>Delivering television channels over an IP network </li></ul><ul><li>20 M  subscribers worldwide...
Why study TV viewing patterns? <ul><li>Understanding of  human viewing behaviors </li></ul><ul><ul><li>Identify social and...
Challenges in traditional TV research <ul><li>Nielsen TV rating </li></ul><ul><ul><li>Select representative samples </li><...
A first study on Telco’s IPTV workloads <ul><li>Collected raw data of everybody watching TV </li></ul><ul><ul><li>A quarte...
Telco’s IPTV service architecture Set-top  box TV DSLAM  customer premise IP backbone TV head end All 150 channels 1-2 cha...
<ul><li>User’s channel change input </li></ul><ul><ul><li>IGMP messages collected  across all 700 DSLAMs </li></ul></ul><u...
Part1.  IPTV overview and dataset Part2.  Analysis of  viewing patterns Part3.  Channel change probability
<ul><li>60% channel changes happen within 10 seconds </li></ul><ul><li>Infrastructure must support fast channel changes </...
Assumptions about user modes <ul><li>Difficulty in inferring user  away  mode </li></ul><ul><ul><li>TV is OFF; or left ON ...
<ul><li>Each user in one of the three states at any given time </li></ul><ul><li>Active session: consecutive time spent on...
<ul><li>Durations </li></ul><ul><ul><li>An average household watched  2.54 hours  of TV and 6.3 channels   (distinct) a da...
<ul><li>Viewing hours across users highly correlated  </li></ul><ul><li>Two peaks at lunch (3PM) and dinner (10PM) times <...
<ul><li>Applied 2-hour thresholds for certain genres (movies, documentaries, sports, etc) </li></ul>Diurnal pattern with l...
<ul><li>90% of concurrent viewers watch 20% of channels </li></ul><ul><li>Follow the Pareto principal  </li></ul>Channel p...
<ul><li>Viewer share of top channels higher at peak times </li></ul><ul><li>Popularity of top channels reinforced at peak ...
Implications of viewing patterns <ul><li>60% of channel changes within 10 seconds ( surfing ) </li></ul><ul><li>=> Challen...
Part1.  IPTV overview and dataset Part2.  Analysis of  viewing patterns Part3.  Channel change probability
Channel change patterns <ul><li>Our goal is to understand </li></ul><ul><ul><li>How do people browse through channels?  Do...
Channel change probability <ul><li>Probability of joining channel  y  after joining channel  x </li></ul>60% linear
Channel viewing probability <ul><li>Probability of viewing channel  y  after viewing channel  x </li></ul>67% non-linear 6...
User arrival and departure rates <ul><li>Batch-like  arrivals and departures </li></ul><ul><li>Inheritance  (continued vie...
Implications of channel change patterns <ul><li>Disparity  in how we change and view channels </li></ul><ul><li>=> Design ...
Summary <ul><li>The first work  to analyze television viewing patterns from complete raw data of IPTV users </li></ul><ul>...
<ul><li>When static 2-hour threshold used for demarcating active and inactive sessions </li></ul>Backup: inferring user mo...
Backup: IPTV hot issues <ul><li>How is IPTV different from traditional TV?  Why telcos deploy IPTV? </li></ul><ul><li>Mode...
Upcoming SlideShare
Loading in …5
×

Slide 1 - Advanced Networking Lab

494 views

Published on

  • Be the first to comment

  • Be the first to like this

Slide 1 - Advanced Networking Lab

  1. 1. Watching Television Over an IP Network Meeyoung Cha MPI-SWS Pablo Rodriguez Telefonica Research Jon Crowcroft U. of Cambridge Sue Moon KAIST Xavier Amatriain Telefonica Research ACM IMC 2008
  2. 2. Internet TV (IPTV) <ul><li>Delivering television channels over an IP network </li></ul><ul><li>20 M subscribers worldwide in 2008 </li></ul><ul><li>Popular types </li></ul><ul><ul><li>1. Telco’s nation-wide provisioned service </li></ul></ul><ul><ul><ul><li>By AT&T, France Telecom, Korea Telecom, Telefonica </li></ul></ul></ul><ul><ul><li>2. Web TV </li></ul></ul><ul><ul><ul><li>Joost, Zatoo, VeohTV, Babelgum, BBC’s iPlayer </li></ul></ul></ul><ul><ul><li>3. Box-based video-on-demand </li></ul></ul><ul><ul><ul><li>Apple TV, Vudu box, Sony’s Internet video link </li></ul></ul></ul>
  3. 3. Why study TV viewing patterns? <ul><li>Understanding of human viewing behaviors </li></ul><ul><ul><li>Identify social and demographic aspects, user profiling </li></ul></ul><ul><li>Cost-efficient design of distribution architectures </li></ul><ul><ul><li>Evaluate existing designs and explore new ones </li></ul></ul><ul><li>Design better channel guides and advertisements </li></ul><ul><ul><li>Help people find interesting programs more quickly </li></ul></ul>
  4. 4. Challenges in traditional TV research <ul><li>Nielsen TV rating </li></ul><ul><ul><li>Select representative samples </li></ul></ul><ul><ul><li>Install metering devices at sampled homes </li></ul></ul><ul><ul><li>Extrapolate statistics across a nation </li></ul></ul><ul><li>< Drawbacks > </li></ul><ul><ul><li>Potential bias in sampling </li></ul></ul><ul><ul><li>Awareness to metering may alter user behaviors </li></ul></ul><ul><li>Gathering data from a large number of samples challenging </li></ul><ul><ul><li>IPTV allows for continuous and detailed TV analysis! </li></ul></ul>
  5. 5. A first study on Telco’s IPTV workloads <ul><li>Collected raw data of everybody watching TV </li></ul><ul><ul><li>A quarter million users from a large IPTV system </li></ul></ul><ul><ul><li>(entire subscribers within a nation) </li></ul></ul><ul><ul><li>150 channels including various genres </li></ul></ul><ul><ul><li>(free-to-air, children, sports, movies, music, etc) </li></ul></ul><ul><ul><li>Collected traces for 6 months </li></ul></ul><ul><li>Largest scale study on TV viewing patterns </li></ul><ul><ul><li>User base 10 times larger than the Nielsen’s </li></ul></ul>
  6. 6. Telco’s IPTV service architecture Set-top box TV DSLAM customer premise IP backbone TV head end All 150 channels 1-2 channels
  7. 7. <ul><li>User’s channel change input </li></ul><ul><ul><li>IGMP messages collected across all 700 DSLAMs </li></ul></ul><ul><li>Trace example </li></ul><ul><ul><li>Timestamp </li></ul></ul><ul><ul><li>DSLAM IP </li></ul></ul><ul><ul><li>Set-top box IP </li></ul></ul><ul><ul><li>Multicast channel IP </li></ul></ul><ul><ul><li>Action (join or leave) </li></ul></ul>Data collection Collected here set-top-box DSLAM
  8. 8. Part1. IPTV overview and dataset Part2. Analysis of viewing patterns Part3. Channel change probability
  9. 9. <ul><li>60% channel changes happen within 10 seconds </li></ul><ul><li>Infrastructure must support fast channel changes </li></ul>Channel holding times
  10. 10. Assumptions about user modes <ul><li>Difficulty in inferring user away mode </li></ul><ul><ul><li>TV is OFF; or left ON without any viewer </li></ul></ul><ul><li>Determined active users as those who change channels within a one hour threshold period </li></ul><ul><ul><li>Tested with longer thresholds </li></ul></ul><ul><li>Demarcate viewing from surfing by the minute </li></ul><ul><ul><li>Nielsen also uses 1 minute threshold </li></ul></ul>
  11. 11. <ul><li>Each user in one of the three states at any given time </li></ul><ul><li>Active session: consecutive time spent on surfing or viewing </li></ul>Three user modes
  12. 12. <ul><li>Durations </li></ul><ul><ul><li>An average household watched 2.54 hours of TV and 6.3 channels (distinct) a day </li></ul></ul><ul><ul><li>Each active session lasted 1.2 hours </li></ul></ul><ul><ul><li>Each viewing event lasted 14.8 minutes </li></ul></ul><ul><li>Per content genre </li></ul><ul><ul><li>Average surfing time longer for documentaries and movies (9-11 sec) than news, music, and sports (6-7 sec) </li></ul></ul>Session characteristics
  13. 13. <ul><li>Viewing hours across users highly correlated </li></ul><ul><li>Two peaks at lunch (3PM) and dinner (10PM) times </li></ul>Diurnal pattern
  14. 14. <ul><li>Applied 2-hour thresholds for certain genres (movies, documentaries, sports, etc) </li></ul>Diurnal pattern with longer away threshold
  15. 15. <ul><li>90% of concurrent viewers watch 20% of channels </li></ul><ul><li>Follow the Pareto principal </li></ul>Channel popularity
  16. 16. <ul><li>Viewer share of top channels higher at peak times </li></ul><ul><li>Popularity of top channels reinforced at peak times </li></ul>Time evolution of channel popularity
  17. 17. Implications of viewing patterns <ul><li>60% of channel changes within 10 seconds ( surfing ) </li></ul><ul><li>=> Challenges for P2P-based IPTV systems </li></ul><ul><li>User focus followed the Pareto principal </li></ul><ul><li>=> IP multicast not efficient for unpopular channels </li></ul>
  18. 18. Part1. IPTV overview and dataset Part2. Analysis of viewing patterns Part3. Channel change probability
  19. 19. Channel change patterns <ul><li>Our goal is to understand </li></ul><ul><ul><li>How do people browse through channels? Do they use electronic program guide? </li></ul></ul><ul><ul><li>Do channel changes result in viewing? </li></ul></ul><ul><ul><li>How do users join and leave a particular channel? </li></ul></ul>
  20. 20. Channel change probability <ul><li>Probability of joining channel y after joining channel x </li></ul>60% linear
  21. 21. Channel viewing probability <ul><li>Probability of viewing channel y after viewing channel x </li></ul>67% non-linear 60% within genre 17% to the same channel
  22. 22. User arrival and departure rates <ul><li>Batch-like arrivals and departures </li></ul><ul><li>Inheritance (continued viewing even after channel changes) </li></ul>arrival departure
  23. 23. Implications of channel change patterns <ul><li>Disparity in how we change and view channels </li></ul><ul><li>=> Design of efficient program guide </li></ul><ul><li>High churn, especially during commercial breaks </li></ul><ul><li>=> Challenging for P2P-based IPTV systems </li></ul>
  24. 24. Summary <ul><li>The first work to analyze television viewing patterns from complete raw data of IPTV users </li></ul><ul><li>Implications on the architecture </li></ul><ul><ul><li>Support fast channel changes </li></ul></ul><ul><ul><li>Handle high churn during commercials </li></ul></ul><ul><ul><li>Reflect Pareto channel popularity </li></ul></ul><ul><li>Implications on the viewing guide </li></ul><ul><ul><li>Devise a better way to browse channels </li></ul></ul><ul><ul><li>Personalize suggestions for users </li></ul></ul>
  25. 25. <ul><li>When static 2-hour threshold used for demarcating active and inactive sessions </li></ul>Backup: inferring user modes
  26. 26. Backup: IPTV hot issues <ul><li>How is IPTV different from traditional TV? Why telcos deploy IPTV? </li></ul><ul><li>Modeling TV viewing habits </li></ul><ul><li>Implications on P2P </li></ul>

×