Watching Television Over an IP Network & TV-Watching Behavior Research presented by Weiping He.

Watching Television Over an IP Network & TV-Watching Behavior Research

presented by

Weiping He

Already Known:Television has been a dominant

and pervasive mass media since 1950’s

New media (# of channels, video signal)

IPTV

Still Unknown or Incompletely known:Ingrained TV viewing habits(Monitoring devices at individual

homes)Nielsen Media Research long-standing research effort

to estimate TV viewing behaviors through monitoring and surveys.

New Weapon to Explore the Unknown:IPTV:

Enable us to monitor user behavior andnetwork usage of an entire network;

More visibility on TV viewing activities; Large user base;

IPTV Service ArchitectureComponents: DSLAM, STB, home

gatewayIPTV channel switching logsRecord the ICMP messages

Timestamp in units of seconds IP address of the DSLAM IP address of the set-top-box (STB) IP address of the multiple group (channel) Multiple option of join or leave

The studies in this paperFirst in-depth analysis of IPTV workloads

based on network traces from one of the world’s largest IPTV systems.

250,000 households, over 6-month period

Characterize the properties of aggregate viewing sessions channel popularity dynamics geographical locality channel switching behaviors browsing pattern user arrival and departure pattern

Elements about the experiment:

Channel groups/genreFree, mixed, kids, docu, local, cine, sports, music, news, audio, rest.

Assumption on user modes Surfing Viewing Away

Note:

Different thresholds can be used according to particular

experiment environment and requirement.

Trace CollectionCollection of IPTV channel

switching logs from backbone provider.

Record the ICMP messages on the channel changes of 250,000 users.

Process the logs/data Pre-process the log by excluding non-video multicast

groups; Chronologically sort IGMP join messages; Analysis the data;

Perspectives for ObservationHigh-level viewing characteristicsChannel popularity and dynamicsGeographical locality

Factors that affect channel changesSwitching from one channel to anotherUser arrival and departure patterns

Section 4

Section 5

High-level viewing characteristics

Number of simultaneous online users

Session characteristicsAttention spanTime spent on each genre

return

Number of simultaneous online users

Friday and Saturday have the lowest evening peaks within the week On weekends:

# of viewers ramps up # of distinct viewers +5% total time spent on TV +30%

back

Session Characteristics

Average household per day: 2.54 hours and 6.3 distinct channels; Average length of each online session: 1.2 hours Median=8s Mean=14.8min

The frequency of a TV watching duration increases from 1-4 sec;The graph after 4-sec mark follows a “power-law-distribution”

back

Attention spanTwo steps for channel selections:a. browsing content to decide whether to continue or stop streaming

b. Switching through multiple channel for repeated browsing, until a desired channel is found

50th percentile values range from 6 to 11 seconds;90th and 95th percentile values range from 3 to 21 minutes.

back

Time spent on each genre

There might be some significant difference from those reported by sampling statistics for US population, in term of channel genre population back

Genre free mixed kids docu local cine

Viewing prob.Num. channels

38.6%6

21.5%19

12.5%7

6.6%12

4.9%17

3.9%6

Genre sports Music news audio rest total

Viewing prob.Num. channels

3.8%8

2.3%11

1.0%13

0.3%15

4.6%36

100%150

Table 2:Breakdown of popularity across genre (probability of a viewer watching each genre)

*Genre categorized “the rest” includes ppv, satellite, and promotional channels.

Channel popularity and dynamics (1)

The top 10% of channels account for nearly 80% of viewers;

(Pareto principle or 80-20 rule) This is consistent across different times of the day, regardless of the

changing of viewer base over the course of a day.o Calculate the effective number of viewers by the fraction of time a user

spent on each channel over a minute period. (Zipf-like distribution)

Channel popularity and dynamics (2)

The average viewer shares are similar to that shown in channel popularity; The graph shows significant fluctuation across the day. Dissimilarity coefficient ξ = 1 − ρ2, ξ greater than 0.1 is considered to have

substantial changes in ranks.

return

Geographical locality

Locality across regionsLocality across DSLAMs

return

Locality across regions

The most popular genres are similar across regions: free, mixed, and kids channels are consistently popular;

Users in some regions watch more local channels than those in other regions.

back

Locality across DSLAMs

backreturn

Factors that affect channel changes

Genre clearly affects the likelihood and frequency of channel changes; Potential factors: the time of day and program popularity;

return

Switching from one channel to another Linear vs nonlinear (EPG) Normalized average probability of channel changes between every

pair of channels; Examine the influence of channel change patterns on viewing.

Switching from one channel to another (cont.)

Several interesting channel switching habits in 1st case:1. Over 60% of channel changes are linear;

2. Certain genres show a distinctive pattern of non-linear channel changes within the genres, e.g., free, sports, and kids;

3. The pattern of linear channel changes continues through the less popular channels like music, satellite, and audio;

4. The remaining 18% of channel changes are non-linear across different genres.

Distinctive difference between the two cases: The consecutive viewing of the same channel in the 2nd case accounts

for 17% of all viewing instances; Non-linear viewing patterns in the 2nd case accounts for 67% of

viewing instances.

In summary:Viewers tend to continue watching the same channel even after switching for some time and with high probability.

return

User arrival and departure patterns

Arrival and departure ratesInter arrival and departure

times

return

Arrival and departure rates

The arrival and departure rates are similar on average. Several observations:

First, the arrival and departure rates vary over the day. Second, user departure patterns show consecutive spikes. Third, the user arrival is much less time-correlated than the departure.

back

Inter arrival and departure times Both median CDF of inter-arrival and inter-departure is 0.07 (the same rate); The arrival rate varies over time and the arrival process is not stationary

over the course of a day;

back return

Implications of findings:Existing and future IPTV systems;Design of the open Internet TV

distribution systems;Other emerging/potential

applications.

Any Questions?