30
DASH in Twitch: Adaptive Bitrate Streaming in Live Game Streaming Platforms Karine Pires, Gwendal Simon Telecom Bretagne, France Nov 2014

DASH in Twitch: Adaptive Bitrate Streaming in Live Game Streaming Platforms

Embed Size (px)

Citation preview

DASH in Twitch:Adaptive BitrateStreaming in Live GameStreaming PlatformsKarine Pires, Gwendal SimonTelecom Bretagne, FranceNov 2014

Live Streaming : A New ContextCloud Delivery Process

2 / 19 K. Pires, G. Simon DASH in Twitch

Live Streaming : A New ContextCloud Delivery Process

more and morenon-professionalbroadcasters

2 / 19 K. Pires, G. Simon DASH in Twitch

Live Streaming : A New ContextCloud Delivery Process

CDNmore and moreheterogeneousend-viewers

2 / 19 K. Pires, G. Simon DASH in Twitch

Live Streaming : A New ContextCloud Delivery Process

CDN

Data-Center

2 / 19 K. Pires, G. Simon DASH in Twitch

Live Streaming : A New ContextCloud Delivery Process

CDN

Data-Center

Adaptive Streaming1 video stream = k representations

2 / 19 K. Pires, G. Simon DASH in Twitch

Live Streaming : A New ContextCloud Delivery Process

CDN

Data-Center

+55 millions unique visitors

4th in US internet traffic

2 / 19 K. Pires, G. Simon DASH in Twitch

Two options for the delivery process

Traditional Process

RawVideo

IngestServers

DeliveryServers

3 / 19 K. Pires, G. Simon DASH in Twitch

Two options for the delivery process

Traditional Process

RawVideo

IngestServers

DeliveryServers

Transcoding Process

RawVideo

IngestServers

TranscodeServers

DeliveryServers

3 / 19 K. Pires, G. Simon DASH in Twitch

Which live streams to transcode

TraditionalProcess↓ CPU

↓ QoE

↑ BW

TranscodingProcess↑ CPU

↑ QoE

↓ BW

Trade-off

Live stream ?

4 / 19 K. Pires, G. Simon DASH in Twitch

Our contributions

Adaptive Bit-Rate streaming in Twitch :Real traces from TwitchDecision of which channels should be delivered with ABRComparison of different decision strategies

5 / 19 K. Pires, G. Simon DASH in Twitch

A Tour on Twitch

6 / 19 K. Pires, G. Simon DASH in Twitch

What is a broadcaster

A broadcaster streaming two gaming sessions

online online

nb. of viewers

timet1 t ′

1 t2 t ′2

session 1 session 2

7 / 19 K. Pires, G. Simon DASH in Twitch

Data retrieval

For 3 months → from 6 Jan. to 6 Apr.

time

allchannelsof one

snapshot

9 :00 9 :05 9 :10 9 :15 9 :20every five minutes → one snapshot

Dataset available : http ://dash.ipv6.enstb.fr/dataset/twitch/8 / 19 K. Pires, G. Simon DASH in Twitch

How many online broadcasters

0 10 20 30 40 50 60 70 80 900

2K

4K

6K

8K

10K

Days

Nb.

ofon

line

chan

nels

min max

9 / 19 K. Pires, G. Simon DASH in Twitch

How much bandwidth is needed

0 10 20 30 40 50 60 70 80 900

1

2

Days

Ban

dwid

th(T

bps)

min max

10 / 19 K. Pires, G. Simon DASH in Twitch

Raw videos properties

240p 360p 480p 720p 1080p0

0.2

0.4

0.6

Representation

Sess

ions

ratio

11 / 19 K. Pires, G. Simon DASH in Twitch

Raw videos properties

240p 360p 480p 720p 1080p0

0.2

0.4

0.6

Representation

Sess

ions

ratio

0.1 1 2 100

0.2

0.4

0.6

0.8

1

Video bit-rate (Mbps)CD

Fof

the

sess

ions

480p

11 / 19 K. Pires, G. Simon DASH in Twitch

Raw videos properties

240p 360p 480p 720p 1080p0

0.2

0.4

0.6

Representation

Sess

ions

ratio

0.1 1 2 100

0.2

0.4

0.6

0.8

1

Video bit-rate (Mbps)CD

Fof

the

sess

ions

480p 720p 1080p

11 / 19 K. Pires, G. Simon DASH in Twitch

Selecting the ABRstreams

12 / 19 K. Pires, G. Simon DASH in Twitch

When to decide a stream should be transcoded

At-startupStrategy

transcoding decision

timet1 t ′

1

On-the-flyStrategy

timet1 t ′

1

transcoding decision

transcoding traditional

13 / 19 K. Pires, G. Simon DASH in Twitch

Strategies

None : no ABRAll : all streams are streamed with ABR

Top-50 (at-startup) : streams that were popularThreshold-1000 (on-the-fly) immediate popularity

14 / 19 K. Pires, G. Simon DASH in Twitch

Strategies

None : no ABRAll : all streams are streamed with ABRTop-50 (at-startup) : streams that were popular

Threshold-1000 (on-the-fly) immediate popularity

14 / 19 K. Pires, G. Simon DASH in Twitch

Strategies

None : no ABRAll : all streams are streamed with ABRTop-50 (at-startup) : streams that were popularThreshold-1000 (on-the-fly) immediate popularity

14 / 19 K. Pires, G. Simon DASH in Twitch

Evaluation settings

Sessions → Twitch datasetViewers → DASH datasetABR → proportional to raw video

15 / 19 K. Pires, G. Simon DASH in Twitch

Results : degraded viewers served

0 10 20

0

0.5

1

None

All

Days

Vie

wer

sra

tioTop Threshold

16 / 19 K. Pires, G. Simon DASH in Twitch

Results : costs evaluation

None All ThresholdTop0

50

100

150

-40%-39%

+5%

Strategies

Tota

lPric

e($

M)Delivery Transcode

17 / 19 K. Pires, G. Simon DASH in Twitch

Conclusion

18 / 19 K. Pires, G. Simon DASH in Twitch

Discussions

preliminary work=global picture→details neglectedmore formal and accurate formulationmore precise QoE estimationbetter transcoding needs modelmanagement of different hardware resources

more efficient strategies can be designedsophisticated strategiesstatistical approaches for popularity forecasting

19 / 19 K. Pires, G. Simon DASH in Twitch

Discussions

preliminary work=global picture→details neglectedmore formal and accurate formulationmore precise QoE estimationbetter transcoding needs modelmanagement of different hardware resources

more efficient strategies can be designedsophisticated strategiesstatistical approaches for popularity forecasting

19 / 19 K. Pires, G. Simon DASH in Twitch