Transcript
Page 1: Towards Peer-Assisted Dynamic Adaptive Streaming over HTTP

Towards Peer-Assisted Dynamic Adaptive Streaming over HTTP

Stefan Lederer, Christopher Müller and Christian Timmerer

19

th I

nte

rnati

on

al

Pac

ket

Vid

eo

Wo

rksh

op

| P

V 2

01

2

M a y 1 0 - 1 12 0 1 2

M u n i c hG e r m a n y

Page 2: Towards Peer-Assisted Dynamic Adaptive Streaming over HTTP

PV 2012 | Peer-Assisted DASH Slide 2

Introduction

• Video streaming needs huge bandwidth ressources

• Can other peers be used to reduce the server load and save costs?

• Can this be integrated into DASH?

Page 3: Towards Peer-Assisted Dynamic Adaptive Streaming over HTTP

PV 2012 | Peer-Assisted DASH Slide 3

Towards Peer-Assisted DASH• Reduction of server load

• Clients offer their downloaded segments– Segment requests are monitored by server– Integration in DASH MPD for future clients

• Intelligent Scheduling Algorithms necessary– When to load from peer, when from server? – Error Handling, etc.

Page 4: Towards Peer-Assisted Dynamic Adaptive Streaming over HTTP

PV 2012 | Peer-Assisted DASH Slide 4

Peer Assisted Streaming

• Peer Traffic– Non-symmetric Internet Connection– Bottleneck: Low Upload Resources

• Split segments to smaller sub-chunks • Restrict number of accepted connection at client• Limit connection duration to prevent stalls

• Target: – Reduce server bandwidth by 10 - 20 %

Page 5: Towards Peer-Assisted Dynamic Adaptive Streaming over HTTP

PV 2012 | Peer-Assisted DASH Slide 5

Peer Assisted Streaming

Page 6: Towards Peer-Assisted Dynamic Adaptive Streaming over HTTP

PV 2012 | Peer-Assisted DASH Slide 6

Implementation

• Proof of Concept using DASH VLC Plugin + PHP for server-side implementation– Provides basic proof of concept and shows

weaknesses to improve– Problem: Amount of clients too low

• Detailed Evaluation in Omnet++– Larger scale simulation with 40 clients– Deterministic client behaviour scenarios

Page 7: Towards Peer-Assisted Dynamic Adaptive Streaming over HTTP

PV 2012 | Peer-Assisted DASH Slide 7

MPD Tracking Server (1)

• MPD Import– Imports original MPD to database

• MPD Generator– Generates MPD based on database– Including other peers using <BaseURL>

– Client has the possibility to update its MPD– Use @MediaRange to signal segment size – Use ZIP compression to reduce MPD size

Page 8: Towards Peer-Assisted Dynamic Adaptive Streaming over HTTP

PV 2012 | Peer-Assisted DASH Slide 8

MPD Tracking Server (2) andClient Modifications• File Tracker

– Gateway for segment access– Segment + Representation as parameter– Stores client IP + timestamp for upcoming MPD

requests– Response: the requested segment

• Local HTTP Server + Buffer at Client– Store segments to disc and provide them on

demand via a local HTTP server

Page 9: Towards Peer-Assisted Dynamic Adaptive Streaming over HTTP

PV 2012 | Peer-Assisted DASH Slide 9

<MPD> <BaseURL> http://www.cdn.com/tracker.php?file= </BaseURL> <Period> <AdaptationSet bitstreamSwitching="true"> <Representation bandwidth="2000000"....>

<BaseURL>http://client1-IP/example</BaseURL>

<BaseURL>http://client2-IP/example</BaseURL> <SegmentList duration="4"> <SegmentURL

media=“rep2MBit_segment1.mp4"> </SegmentList> </Representation> <Representation bandwidth="4000000"....

<BaseURL>http://client1-IP/example</BaseURL> <!-- further base urls and Segments --> </Representation> <!-- further representations --> </AdaptationSet> </Period>...

MPD Example

Server URL withFile Tracker

Peers offering the segment

Page 10: Towards Peer-Assisted Dynamic Adaptive Streaming over HTTP

PV 2012 | Peer-Assisted DASH Slide 10

Peer Assisted DASHExample

Page 11: Towards Peer-Assisted Dynamic Adaptive Streaming over HTTP

PV 2012 | Peer-Assisted DASH Slide 11

Evaluation

• OMNet++– Simulation framework– INET framework for protocol stack– HTTP Client/Server implementation– DASH Client based on DASH VLC Plugin / libDASH– MPD Generator + Segment Tracker using external

MySQL database

Page 12: Towards Peer-Assisted Dynamic Adaptive Streaming over HTTP

PV 2012 | Peer-Assisted DASH Slide 12

Evaluation Settings

Bitrate Resolution

101 kbit/s. 320x240

201 kbit/s. 480x360

395 kbit/s. 480x360

800 kbit/s. 854x480

1372 kbit/s. 853x480

1992 kbit/s. 1280x720

2995 kbit/s. 1920x1080

3992 kbit/s. 1920x1080

4979 kbit/s. 1920x1080

5936 kbit/s. 1920x1080

Page 13: Towards Peer-Assisted Dynamic Adaptive Streaming over HTTP

PV 2012 | Peer-Assisted DASH Slide 13

Simulation 1:Results – Server

- 15 %

• 6 Mbps maximumRepresentation limit

• Clients select different representations according to their downlink speed- Number of clients offering

one specific segment is low

Page 14: Towards Peer-Assisted Dynamic Adaptive Streaming over HTTP

PV 2012 | Peer-Assisted DASH Slide 14

Simulation 2:Results - Server

- 25 %

• 1,4 Mbps maximumRepresentation limit

• Clients select the same maximum representation- Downlink speed of all

clients is sufficient - Lower upload time for

segments

Page 15: Towards Peer-Assisted Dynamic Adaptive Streaming over HTTP

PV 2012 | Peer-Assisted DASH Slide 15

Simulation ResultsExample Client

• Client:– 8 Mbps Downlink– Activation at

second 214 of the simulation

• Simulation 2:– 1,4 Mbps max.

Representation limit

Page 16: Towards Peer-Assisted Dynamic Adaptive Streaming over HTTP

PV 2012 | Peer-Assisted DASH Slide 16

Cost Saving Possibilities

• Simulation 1: 15 % traffic cost reduction– Total costs: US$ 4.14 per hour – Savings: US$ 0.62 per hour

• Additionally: Reduced reserved bandwidth capacity

Page 17: Towards Peer-Assisted Dynamic Adaptive Streaming over HTTP

PV 2012 | Peer-Assisted DASH Slide 17

Conclusions

• Torwards Peer-Assisted DASH– Peer-assited streaming using standard-compliant

DASH MPDs– Maintainance of DASH advantages – Relative simple system design and implementation

work

• Evaluation simulation– Up to 25 % bandwidth savings– Directly convertable to CDN cost reductions

Page 18: Towards Peer-Assisted Dynamic Adaptive Streaming over HTTP

PV 2012 | Peer-Assisted DASH Slide 18

Conclusions & Further Work

• Much more possibilities– Intelligent client clustering in larger scale

environments– Peer management & download algorithm

improvements– MPD update improvements– Detailed CDN cost analysis – Evaluation of some Set-Top box scenarios– Integration to Content Centric Networking (CCN)

Page 19: Towards Peer-Assisted Dynamic Adaptive Streaming over HTTP

The END

http://dash.itec.aau.at

19

th I

nte

rnati

on

al

Pac

ket

Vid

eo

Wo

rksh

op

| P

V 2

01

2

M a y 1 0 - 1 12 0 1 2

M u n i c hG e r m a n y


Recommended