160 likes | 308 Views
http://napa-wine.eu. Delay-Aware Push/Pull Protocols for Live Video Streaming in P2P Systems. Alessandro Russo, Renato Lo Cigno DISI – University of Trento, Italy locigno @ disi.unitn.it http://disi.unitn.it/locigno. P2P Multimedia Streaming.
E N D
http://napa-wine.eu Delay-Aware Push/Pull Protocols for Live Video Streaming in P2P Systems Alessandro Russo, Renato Lo Cigno DISI – University of Trento, Italy locigno @ disi.unitn.it http://disi.unitn.it/locigno
P2P Multimedia Streaming • P2P is cool, but why streaming? And why live, real-time streaming • Think of out-of-country TV broadcasting • easier to get Internet connection than a satellite dish • Think of the cost of starting a new TV channel • traditional TV broadcasting vs. client-server vs. P2P • P2P-TV could become one of the dominant multimedia applications on the Internet • Some systems already deployed: PPLive, TVAnts, CoolStreaming, … with hundreds of channels already available www.disi.unitn.it/locigno ICC 2010 - NGS, Cape Town, June 26 2010
P2P Multimedia Streaming contd. • P2P-TV is resource-hungry • previously unseen traffic volumes to/from the users • 1+ mbit/s sustained download • Even higher upload (if available) • P2P-TV is challenging to design • large peer count with heterogeneous networking resources • This is not VoD, potentially millions of users watching the same live channel • tight delay constraints • This is not file sharing, delay is the design objective www.disi.unitn.it/locigno ICC 2010 - NGS, Cape Town, June 26 2010
Outline of Talk • P2P streaming systems, definitions • Protocols for Chunk Trading • Push & Pull, why both? • Delay Aware Peer Selection • Wrap-Up and Future Work www.disi.unitn.it/locigno ICC 2010 - NGS, Cape Town, June 26 2010
P2P Streaming Systems • As in the previous talk ... yes, we do talk to each other before presentations • 1 source generates media chunks at Bs Mbit/s • Peers receive and transmit chunks • The system is unstructured and chunks swarms through the overlay topology • No fixed distribution tree • Each peer is connected to a subset of the other peers • Neighborhoods are stable (in this study) • Peers are autonomous and not synchronized: they evolve solely based on the protocol, no other coordination supposed www.disi.unitn.it/locigno ICC 2010 - NGS, Cape Town, June 26 2010
Network Model • We consider an n-regular topology with symmetric connectivity • Good approx. of a random topology • Easy to construct and maintain • Access is the bottleneck and ADSL-like • Both upload and download bandwidth follow a simplified reservation sharing mechanism (on UDP) • Congestion is avoided setting a maximum to parallel transmissions 3-regular topology www.disi.unitn.it/locigno ICC 2010 - NGS, Cape Town, June 26 2010
Chunk Trading • Each peer • Receives chunks from the other peers • Redistributes chunks to neighbour peers • Two main drivers of the Chunk Trading Logic: • The protocol • The scheduling (local choices of Peers and Chunks) • We focus here on the protocol • Scheduling is “plain” (or trivial if you prefer) • Major “fights” discuss benefits of Push or Pull-based protocols (these latter called also data-driven ... with no reason ) • Indeed there is no reason to use one OR the other, they can be mixed in the same application www.disi.unitn.it/locigno ICC 2010 - NGS, Cape Town, June 26 2010
Traded Push & Pull Peers to trade with are chosen at random, or may follow some “logic”: distance, av. bandwidth, delay, ... Push can also be “blind” (many studies assume so), but duplicated chunks waste bandwidth www.disi.unitn.it/locigno ICC 2010 - NGS, Cape Town, June 26 2010
Node Active and Passive Behavior Peers transmit and receive chuncks Peers transmit offers/requests: they are active protocol entities (or clients in Internet terminology) the node starts requests/offers Peers satisfy offers/requests: they are passive protocol entities (or servers in Internet terminology) the node receives requests/offers www.disi.unitn.it/locigno ICC 2010 - NGS, Cape Town, June 26 2010
Selecting Peers and Chunks • Peers are selected in the neighborhood • At random – R • Following a distribution weighted by 1/RTT: Delay Aware – D • Selected peers are “poisoned” to avoid deterministic patterns and starvations (e.g. one peer very close and the others far away) • In Push: chunks are offered selecting the w most recent available • In Pull: chunks are requested selecting the w most needed, i.e., those closer to the playout time (or oldest) • Push and Pull phases are asynchronous and compete for the bandwidth resources • Offers are put forward and requests satisfied only if there are available resources on the local link • Model suitable for applications that enforce some sort of shaping • Signaling and data transmission are sequential www.disi.unitn.it/locigno ICC 2010 - NGS, Cape Town, June 26 2010
The role of Push and Pull: Diffusion delay distribution • RTT = [10,250] ms • = 1,1 Bp = 1.9 ... 3.4 Bs Peer Choice = R www.disi.unitn.it/locigno ICC 2010 - NGS, Cape Town, June 26 2010
D R The impact of R/D, parallel tx. a, and window w RTT = [10,250] ms www.disi.unitn.it/locigno ICC 2010 - NGS, Cape Town, June 26 2010
Delay Aware bandwidth reduction Tail Behavior: 95-th percentile, different RTTs www.disi.unitn.it/locigno ICC 2010 - NGS, Cape Town, June 26 2010
Summary and Future Work – Summary • Analysis and insight in a flexible protocol (Push/Pull) for P2P streaming • Assessment of the impact (very positive) of selecting peers in your neighborhood based on their RTT, one of the few easy-to-measure network characteristics • Study of some tuning parameters of the basic protocol www.disi.unitn.it/locigno ICC 2010 - NGS, Cape Town, June 26 2010
Summary and Future Work – Work ... • Implementation of Push/Pull in GRAPES libraries: done! http://napa-wine.eu • Implementation of P2PTV streamers in NAPA-WINE peers based on Push/Pull protocols, to compare with other offer/select protocols: under wayhttp://napa-wine.eu • Exploration of tradeoffs between building delay-aware topologies with random peer choice, vs. random topologies vs delay-aware peers selection: to be done • Integration of delay-aware techniques with other network-aware strategies: in discussion & first tests • Improvements and open issues for parallel signaling and chunk transfer: under way (both simulation & implementation) www.disi.unitn.it/locigno ICC 2010 - NGS, Cape Town, June 26 2010
THE END Thank you! Questions? Comments? www.disi.unitn.it/locigno ICC 2010 - NGS, Cape Town, June 26 2010