Problem setting: time-shifted streaming in P2P LIVE STREAMING SYSTEM. We want to preserver the scalability of the P2P system. How? Make use of peer’s local storage in addition to their uplink bandwidths!
System overview: SPPM for LS, and its extension Explain “video availability Intro to Fast prefetching as parent selection algorithm Video availability Server load Video quality
SPPM originally supports live streaming to a population of peers
Define LS peers
Peer database at the server Peer identifier, join time, initial video time-stamp Latest video time-stamp is estimated (and periodically corrected) Peers cache received video packets Protocol can handle video time-stamp
Explain time-video plot. But, no needs for explaining “video trajectory”! Needs for defense on the two assumptions?
150 peers. 90% of them are active -> 135. Maximum # of LS reach 60.
Explain how to interpret the video availability plot. Each point. Higher, better. Latest part is covered by LS peers. In early stage, all portion of video is well covered. Averaged 100 simulations. Intro to NS2 simulator (packet-level)
Say: 2R is sufficient to support 100 LS peers. For 100 TS peers, more than 20R bandwidth required.
Explain time-video plot again. Explain “video trajectory”! A peer can receive video faster than playout speed Why fast prefetching? Data are available. BW is available. (good use of peer resources) Benefit: Good for children ( relaxed playout deadline) Larger video coverage -> may reduce server load
How to maximize fast prefetching? LS peer parent selection: number of logical hops to server -> it’s not optimal here. Learn more about parent candidates and choose the best for maximum expected throughput
EXPLAIN COLORS EXPLAIN SIMULATION SETTINGS
Time 0 ~ 900s : server loads are almost the same for different schemes. But, there is a difference in video quality!
At the last step, the child download rate >= r’
Peers discard cached packets only when they leave the system Uniform access: worst case video random access
Weight the region with lifetime probability
Time-Shifted Streaming in a P2P Video Multicast System Jeong h un Noh , Aditya Mavlankar, Pierpaolo Baccichet 1 , and Bernd Girod Information Systems Laboratory Stanford University 1 Now with OnLive Inc.