Efficient Sub-stream Encoding and Transmission for P2P Video on Demand 1 Efficient Sub-Stream...

Efficient Sub-stream Encoding and Transmission for P2P Video on Demand

1

Efficient Sub-Stream Encoding and Efficient Sub-Stream Encoding and Transmission for P2P Video on Transmission for P2P Video on

DemandDemand

Zhengye LiuYanming Shen

Shivendra S. PanwarKeith W. Ross

Yao Wang

Polytechnic University, Brooklyn, NY, USA


2

OutlineOutline

• Motivation• P2P VoD system• Sub-stream encoding and transmission• Simulations• Conclusion


3

MotivationMotivation

• Video-on-demand services – Youtube, MSN video, google video,…– Content distribution networks (CDNs)

• P2P live streaming systems – PPLive, PPStream, UUSee, Coolstreaming,…

– Support thousands of users simultaneously


4

P2P VoD SystemP2P VoD System

• Multiple video architecture– Extension of CDNs: Peers act as video servers

– Contribute storage in addition to bandwidth

– Help each other with stored videos


5

Proposed System with Multiple Sub-Proposed System with Multiple Sub-Streams Streams

In this illustration, two simultaneous streaming sessions are requested from nodes 4 and 5. The system initially selects nodes 2 and 3 to serve node 4’s request, and selects nodes 4 and 1 to serve node 5’s request. After Node 2 goes down, the system finds node 6 as a replacement


6

Benefit of Using Layered Benefit of Using Layered Coding/MDCCoding/MDC

• Adaptive to the long-term bandwidth fluctuation due to peer churn– Uplink bandwidth fluctuation– Received video quality adapts to the available uplink bandwidth

• Robust to peer failure/disconnection– One supplier failure only affects one/several sub-stream(s)

– Video quality will not be impaired seriously


7

Multi-Stream Coding SchemesMulti-Stream Coding Schemes

(a) Layered coding

(b) MDC

(c) Ideal scheme

– Compare schemes (a), (b) – Design (c)


8

Push-Pull Delivery with Layered Push-Pull Delivery with Layered CodingCoding

• Store all layers of a video• Push-pull for layer delivery

• Storage consumed for one video: RM


9

The Ideal Scheme should consume The Ideal Scheme should consume minimum storageminimum storage

• Can we reduce the consumed storage?

• The minimum storage: R+R/2+,…,+R/M≈ R ln(M)


10

RS CodingRS Coding

RS (8,2) coding for Layer 2

Any two received chunks can recover the original two chunks

L22 R23 L21 L22

R23 L21 L22R25


11

RS Coding Instead of ReplicatingRS Coding Instead of Replicating

RS (8,k) coding, k=1,2,3,4


12

Redundancy-Free Transmission Based on Redundancy-Free Transmission Based on

Push-Pull ArchitecturePush-Pull Architecture

• A receiver schedules the chunks that should be delivered (Pull)

• A supplier pushes the chunks based on the schedule (Push)

R


13

Features of Proposed SchemeFeatures of Proposed Scheme

• Ideal Scheme– Equal importance (like MDC)– Redundancy free transmission (like Layered coding)

• Minimum storage consumed– R ln(M) vs. RM

• M=4, 2.08R vs. 4R, save about 50%• M=32, 3.47R vs. 32R, save about 89%

• Since a peer needs to store fewer substreams: save server bandwidth


SimulationsSimulations

• Setting– 3000 peers, different uplink bandwidth– 30 videos, different popularity (Zipf distribution with parameter 0.27)

• Simulate two video sequences– Foreman: low rate– Mobile: high rate

• Compare SLRS, Layered Coding, MD-FEC, RFMD• Performance metrics

– Discontinuity: number of undecodable GOPs/Total number of GOPs

– PSNR

14


Simulation ResultsSimulation Results

15

0 200 400 600 800 1000 1200 1400 1600 1800 20000

0.1

0.2

0.3

0.4

0.5

n

Dis

cont

inui

ty:

MD-FECLayered codingSLRSRFMD

0 200 400 600 800 1000 1200 1400 1600 1800 2000

16

18

20

22

24

26

28

30

32

34

36

38

n

Ave

rage

PSN

R(d

B)

MD-FECLayered codingSLRSRFMD

Discontinuity of different schemes vs. number of active users under the “Foreman” sequence

PSNR of different schemes vs. number of active users under the “Foreman” sequence


Simulation ResultsSimulation Results

16

0 200 400 600 800 1000 1200 1400 1600 1800 20000

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

n

Dis

cont

inui

ty:

MD-FECLayered codingSingle layer codingRFMD

0 200 400 600 800 1000 1200 1400 1600 1800 20000

5

10

15

20

25

30

n

Ave

rage

PS

NR

(dB

)

MD-FECLayered codingSingle layer codingRFMD

Discontinuity of different schemes vs. number of active users under the “Mobile” sequence

PSNR of different schemes vs. number of active users under the “Mobile” sequence


17

ConclusionConclusion

• Propose a redundancy-free transmission scheme based on a push-pull architecture

• RFMD– Advantages

• All substreams have equal importance: graceful quality degradation

• Only the source bits are transmitted: no transmission redundancy

• Coding instead replication: any combination of M or fewer substreams can be used in reconstructing video

– Additional cost:• Feedback from the client to each supplying peer: feasible in typical P2P VoD applications


18

Thanks!Thanks!

Efficient Sub-stream Encoding and Transmission for P2P Video on Demand 1 Efficient Sub-Stream...

Documents

Transcript of Efficient Sub-stream Encoding and Transmission for P2P Video on Demand 1 Efficient Sub-Stream...