Network Impact of P2P-TV ZappingManxue Wang Olivier Fourmaux Yuko Nakamura Takumi Miyoshi

UPMC Sorbonne Universites Shibaura Institute of Technology

Introduction

I Strong increase of audio/video services in the Internet. wide distribution by expensive CDN infrastructure. local distribution by residential operators

I Alternative with P2P-TV applications. different from usual file sharing P2P : hard time constraints. P2P-TV overlay depend on content and user behavior

I IPTV user behaviors are zapping or steady [Cha & all, IMC’08]. most previous P2P-TV analysis works focus on steady state. our work focuses on transient state and zapping behavior

Experimentation

I 4 popular P2P-TV applications. SOPCast, PPStream, PPLive and UUSee

I 5 popular TV channels. CCTV1, CCTV2, CCTV4, CCTV10 and CCTV13

I 2 different locations: Paris and Tokyo (broadband access)

I 2 measurement periods (June and September 2011)

Measurement

I Black-box analysis. proprietary applications (internal mechanisms unknown). capture of all the traffic of one peer

I One channel preliminary results. active peers (other peers with whom the measured peer exchange). traffic split (upload/download, signaling/video...). traffic distribution (top ten peers)

I Five channel capture. peer and traffic distribution among channels

I Anonymised traces availables at. http://content.lip6.fr/traces/

Conclusion and further works

I Overload estimation in transient state: only preliminary results

I France and Japan results similar

I Further works. statistical results over the whole dataset. deeper application behavior analysis. combination with adverse network situations

Example with SOPCast, 1 channel (CCTV1)

I Measurement done on Tuesday, June 30, 2011 at 14h GMT

I Chronogram used with 1 channel:

channel 1

time (s)

viewing channel period (200s)

measurement period

0 30 230150

steady state (60s)(120s)

210

I Numerical results (simple signaling/video split based on packet size):

Trace Total data (KB) Duration (s) Download (KB) Upload (KB)Full 41894 238 18215 2269760s 8073 60 4042 3841

Trace Sig.Dl.(KB) Vid.Dl.(KB) % Sig.Up.(KB) Vid.Up.(KB) %Full 2165 16049 12 15216 7481 4960s 419 3623 10 2632 1209 46

Num

ber o

f pee

rs

Time (second)

peers_active_one_channel_sopcast_cctv1_14_25_200_1.0sagg

steady state: 60s

Fig. 1. Active peers associated with one SOPCast channel

0 25 50 75

100 125 150 175 200 225 250 275 300

0 30 60 90 120 150 180 210 240 270Amon

t of d

ata

dow

nloa

ded

(KBy

te)

Time (second)

data_active_down_one_channel_sopcast_cctv1_1_200

steady state: 60s

data downloadedtop 10

top 9top 8top 7top 6top 5top 4top 3top 2top 1

Fig. 2. Data downloaded with one SOPCast channel

Example with SOPCast, 5 channels, zapping time = 25s

I Measurement done on Tuesday, June 30, 2011 at 14h GMT

I Chronogram used with 5 channel:

54321

channel

time (s)

measurement period

0 30tztztztz viewing last channel period (200s)

steady state (60s)(120s)

I Peer distribution among channels:

Full CH 1 CH 2 CH 3 CH 4 CH 5601 74 48 50 50 228

0 2 4 6 8

10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40 42 44 46 48 50 52 54 56

0 30 60 90 120 150 180 210 240 270 300 330 360 390

Num

ber o

f pee

rs

Time (second)

peers_active_by_each_channel_five_channel_sopcast_1_25_25_25_25_200

steady state: 60s

tz = 25s tz = 25s tz = 25s tz = 25s number of peers downloaded channel 1: number of download peerschannel 2: number of download peerschannel 3: number of download peerschannel 4: number of download peerschannel 5: number of download peers

Fig. 3. Active peer per channel with SOPCast zapping among 5 channels

0 15 30 45 60 75 90

105 120 135 150 165 180 195 210 225 240 255 270 285 300 315 330 345 360 375 390

0 30 60 90 120 150 180 210 240 270 300 330 360 390

Amou

nt o

f dat

a do

wnl

oade

d (K

Byte

)

Time (second)

trafic_active_by_each_channel_five_channel_sopcast_1_25_25_25_25_200

tz = 25s full trace data downloaded channel 1: data downloadedchannel 2: data downloadedchannel 3: data downloadedchannel 4: data downloadedchannel 5: data downloaded

steady state: 60s

tz = 25s tz = 25s tz = 25s

Fig. 4. Traffic per channel with SOPCast zapping among 5 channels

olivier.fourmaux@upmc.fr miyoshi@shibaura-it.ac.jp