08032006 live p2p streaming talk itin

Live Video P2P Streaming: 

Live Video P2P Streaming Thomas SILVERSTON Olivier FOURMAUX Bénédicte Le GRAND

Presentation: 

Presentation Ph.D Student in Computer Sciences – Networks since Octobre 2005 University Paris 6 – LIP6/NPA Laboratory Lab Research Themes: Modeling Metrology Self-Organization Content Networks

Origin of Internet: 

Origin of Internet From universities interconnection network to world interconnection network Killer applications (email, web) Success of the Internet Growth of the Internet Modifications of its use

Current Internet: 

Current Internet Better equipment Faster CPU to get many processes More Bandwidth available to get bigger object More Resources available In the core of the Internet (routers, i.e. Internet Service Providers) At the edge of the Internet (clients)

Future of the Internet: 

Future of the Internet Interactive text and HTML is history Every Media converges on the Internet Voice, Audio, Video because of increasing Internet resources Because people want it

Background: 

Background We are focusing on live video streaming A bit different from: Video on Demand (not live) Interactive audio and/or video Telephony over IP, Videoconference

Outline: 

Outline I. Background II. Streaming and the Internet Architecture The Paradox of Video Streaming on the Internet III. Network Broadcasting IV. Motivations V. Works VI. Conclusion

Outline: 

Outline I. Background II. Streaming and the Internet Architecture The Paradox of Video Streaming on Internet III. Network Broadcasting IV. Motivations V. Works VI. Conclusion

Video Streaming : 

Video Streaming Reads and Decodes the video stream during reception Audio and Video Consumes a lot of Bandwidth, CPU Targets many receivers (TV or Radio) Internet resources have to be well managed Need powerful mechanisms

Traditional architecture of the Internet: 

Traditional architecture of the Internet Client/Server model Overloads Server CPU Consumes Server BW Poor performances !

Outline: 

Outline I. Background II. Streaming and the Internet Architecture The Paradox of Video Streaming on Internet III. Network Broadcasting IV. Motivations V. Current Works VI. Conclusion

Outline: 

Outline I. Background II. Streaming and the Internet Architecture The Paradox of Video Streaming on Internet III. Network Broadcasting IV. Motivations V. Current Works VI. Conclusion

Network Broadcasting: 

Network Broadcasting Video streaming applications target a lot of receivers (TV, Radio) Need Network Broadcasting mechanisms Duplicate all messages to receivers Manage the group membership Internet has no broadcasting mechanisms Need to outperforms Client/Server architecture

Network Broadcasting1st Proposal: inside the network: 

Network Broadcasting 1st Proposal: inside the network Routers duplicate all packets to receivers i.e. Multicast IP Pros Good utilization of network Cons Goes against driving principles of the Internet security problems etc. Suffers a lack of deployment

Network Broadcasting2nd Proposal: Content Distribution Network: 

Network Broadcasting 2nd Proposal: Content Distribution Network Content Servers are replicated strategically in the Internet All clients are redirected to closer Server Ex: Akamai

Network Broadcasting2nd Proposal: Content Distribution Network: 

Network Broadcasting 2nd Proposal: Content Distribution Network

Network Broadcasting2nd Proposal: Content Distribution Network: 

Network Broadcasting 2nd Proposal: Content Distribution Network Pros Server not overloaded scalable Cons Need commercial contract with intermediate (ISP) Very costly in infrastructure Infrastructure modifications complex to deploy

Network Broadcasting 3rd Proposal : Peer-to-Peer Network: 

Network Broadcasting 3rd Proposal : Peer-to-Peer Network Opposite to Client/Server model Client are both client and/or server Client get the video stream from other clients Duplication of video stream at client level

Network Broadcasting 3rd Proposal : Peer-to-Peer Network: 

Network Broadcasting 3rd Proposal : Peer-to-Peer Network

Network Broadcasting 3rd Proposal : Peer-to-Peer Network: 

Network Broadcasting 3rd Proposal : Peer-to-Peer Network Pros Easy to deploy No infrastructure modifications Software in end-system (client) Adding value to end-system Cons Suffers failures Opposite to dedicated entities (routers) Bad utilization of network Too many duplicated messages

Outline: 

Outline I. Background II. Streaming and the Internet Architecture The Paradox of Video Streaming on Internet III. Network Broadcasting IV. Motivations V. Current Works VI. Conclusion

Outline: 

Outline I. Background II. Streaming and the Internet Architecture The Paradox of Video Streaming on Internet III. Network Broadcasting IV. Motivations V. Current Works VI. Conclusion

Motivations: 

Motivations Multicast IP CDN P2P

Motivations: 

Motivations P2P Streaming Protocols have to be well designed Streaming uses many resources of the network Targets a lot of clients Internet has no broadcasting mechanisms P2P can implement these mechanisms But P2P has a bad utilization of network However, P2P is easily deployable

Main Goals: 

Main Goals Analyze P2P streaming protocols Increase performances Manage network utilization

Outline: 

Outline I. Background II. Streaming and the Internet Architecture The Paradox of Video Streaming on Internet III. Network Broadcasting IV. Motivations V. Works VI. Conclusion

Outline: 

Outline I. Background II. Streaming and the Internet Architecture The Paradox of Video Streaming on Internet III. Network Broadcasting IV. Motivations V. Works VI. Conclusion

Works: 

Works Designing a new Protocol: Real implementation Prototype implementation Mathematical model Simulations Simulations Efficient Simple

Works: 

Works Nowadays: 3 different approaches for Live P2P Streaming Source-driven Approach Receiver-driven Approach Data-driven Approach Try to compare them by simulation Implement a new discrete-event simulator Determine the better approach

Source-driven Approach: 

Source-driven Approach Server Clients are organized around the Source Message Client

Receiver-driven Approach: 

Receiver-driven Approach Server Clients organize other clients as resources Message Client

Data-Driven Approach: 

Data-Driven Approach Server Clients get the stream from other clients according to data availability Message Client

Simulator: 

Simulator A simple discrete-event simulator allows us: To obtain protocols comportment To compare protocols between them To evaluate quickly new protocols

Simulator: 

Simulator T=35ms SEND *next *ptrclient T=37ms HB *next *ptrclient T=38.4ms HB *next *ptrclient T=40.1ms HB *next *ptrclient T=39.2ms HB *next *ptrclient T=68ms SEND *next *ptrclient T=1037ms HB *next *ptrclient T=1038.4ms HB *next *ptrclient Id 0 Level 1 NA 208 Délai 1 state CNTD *next neighbor NULL Reading Clients Event list : simulation engine

Results: 

Results Average Time to First Packet Data-driven outperforms Source-driven

Results: 

Data Packet losses rates ac. To dynamicity number of leaving clients Results

Results: 

Data Packet Losses Rates ac. To dynamicity Leaving level of clients Results

Results: 

Results The simulator is a framework to Implement protocols Compare protocols performances Highlights differences between them We show: Data-driven outperforms source-driven approach

Outline: 

Outline I. Background II. Streaming and the Internet Architecture The Paradox of Video Streaming on Internet III. Network Broadcasting IV. Motivations V. Works VI. Conclusion

Outline: 

Outline I. Background II. Streaming and the Internet Architecture The Paradox of Video Streaming on Internet III. Network Broadcasting IV. Motivations V. Works VI. Conclusion

Existing Applications: 

Existing Applications http://www.pplive.com http://www.ppstream.com http://www.sopcast.com http://www.gridmedia.com => academic http://www.tvants.com

Future Works: 

Future Works Realistic input parameters to calibrate the simulator Simulations with realistic scenarios Scheduling and Caching in a peer Designing new protocols and approaches Beeing billionaire with live video p2p streaming: the new killer appz ;) i.e. Skype etc.

Slide43: 

Questions?

Slide44: 

Thank You ! ;)

References: 

References http://www.peercast.org CoolStreaming/DONet: A Data-driven Overlay Network for Peer-to-Peer Live Media Streaming, Xinyan Zhang, Jiangchuan Liu, Bo Li, and Tak-Shing Peter Yum, IEEE Infocom, April 2005 Source vs Data-driven Approach for Live P2P Streaming, Thomas Silverston and Olivier Fourmaux, IEEE ICN, April 2006 Streaming P2P: Impact of overlay managing protocols, master thesis, Septembre 2005 http://www.akamai.com Some Slides from http://aoeit.ie.cuhk.edu.hk/docs/Coolstreaming.ppt