Resilient PeertoPeer Streaming Presented by Yun Teng Resilient
Resilient Peer-to-Peer Streaming Presented by: Yun Teng
Resilient Peer-to-Peer Streaming Authors l l l Venkata N. Padmanabhan Helen J. Wang Philip A. Chou From l Microsoft
Motivation Ø Distributing “live” streaming media content to a potentially large and highly dynamic population of hosts. l “Live” streaming refer to the simultaneous distribution of the same content to all clients
Challenge Ø Peer-to-peer content distribution is attractive because the bandwidth available to serve content scales with demand. Ø A key challenge: making content distribution robust to peer transience.
Approach Ø Introduce redundancy, both in network paths and in data Ø Multiple diverse distribution tree: provide redundancy in network paths Ø Multiple description coding (MDC): provide redundancy in data
Coop. Net Ø Makes selective use of P 2 P networking, placing minimal demands on the peers Ø Goal: help a server tide over crises such as flash crowds rather than replace the server with a pure P 2 P system
Assumption A node participates in and contributes bandwidth for content distribution only so long as the user is interested in the content. It stops forwarding traffic when the user tunes out Ø A node only contribute as much upstream bandwidth as it consumes in the downstream direction (applies to the total bandwidth in and out of a node aggregated over all trees Ø Nodes in Coop. Net are inherently unreliable Ø
Tree management
Goals Ø Short trees Ø Tree diversity Ø Efficiency Ø Quick join and leave Ø Scalability
Conflicts Ø Tree diversity versus efficiency Ø Quick join and leave versus scalability
Feasibility of the Centralized Protocol September 11 flash crowd at MSNBC Ø l l At peak, 18, 000 nodes, 1, 000 arrivals and departures per second On average, 10, 000 nodes, 180 arrivals and departures per second Resource requirement Ø l l l Memory: 10 MB Network bandwidth: 8 Mbps CPU: 40 ns memory cycle, allow 390 memory accesses per insertion
Centralized Tree Management Ø Randomized Tree Construction Ø Deterministic Tree Construction
Tree Efficiency / Topology Awareness Ø Need an efficient way to pick a proximate parent for a node without requiring extensive P 2 P network measurements Ø Each node maintains its “delay coordinates” of ping times to a small set of landmark hosts Ø Root pick the closest node for incoming node from a set of candidate parents
Tree Repair Ø Due to node leave Ø Two types l l Voluntary – Notify the root Failure – Detect failure
Multiple Description Coding (MDC)
Overview Ø Encoding an audio and/or video signal into M>1 separate streams, or descriptions, such that any subset of these descriptions can be received and decoded. Ø The distortion with respect to the original signal is commensurate with the number of descriptions received.
Overview (cont. ) MDC incurs a modest performance penalty relative to layered coding, which in turn incurs a slight performance penalty relative to single description coding. Ø The audio and/or video signal is partitioned into groups of frames (GOF), each group having duration of T (such as 1 second). Each GOF is independently encoded, error protected, and packetized into M packets. Ø
Coop. Net MDC System Architecture
Configuring MDC Ø GOF duration G = 1 second Ø M = 16 descriptions Ø Packet size P = 1250 bytes Ø T = 8 trees
Performance evaluation
Impact of Number of Distribution Trees
Effectiveness of MDC Ø Probability distribution of descriptions received vs. number of distribution trees Ø Root out-degree = 100 Ø Maximum client out-degree = 4
Impact of Repair Time
Related work
Related Work Ø Application-level Multicast Ø Source Coding and Path Diversity
References Ø V. N. Padmanabhan, H. J. Wang, and P. A. Chou. Resilient Peer-to-Peer Streaming. Technical Report MSR -TR-2003 -11, Microsoft Research, Redmond, WA, March 2003. Ø V. N. Padmanabhan, H. J. Wang, P. A. Chou, and K. Sripanidkulchai. Distributing Streaming Media Content Using Cooperative Networking. In Proc. NOSSDAV, May 2002.
Q&A
Thank you!
- Slides: 28