1. CS 414 - Spring 2012 CS 414 – Multimedia Systems Design Lecture 38 – P2P Streaming (Part 2) Klara Nahrstedt

2. Administrative MP3 deadline Saturday April 28, 5pm  Demonstrations of MP3, April 30, 5-7pm Groups should sign up as follows:  5-6pm – 3 rd floor 3401 SC for groups that work with laptops and need wireless connectivity for MP3  6-7pm - 216 SC basement for groups who work with PCs and need wired connectivity for MP3 Top four groups will be decided Monday, April 30 at 7pm (via email, also posted on the newsgroup/classwebsite) - these groups will compete in front of the judges on Tuesday, May 1 CS 414 - Spring 2012

3. Administrative Competition of final four groups on Tuesday 5-7pm in 3401 SC/ 216 SC  ByteMobile Inc. company – judging competition (and TA/Instructor)  The top four groups should prepare 3-4 power-point slides to present Intro Slide – name of your system and your names (1 slide) Surveillance System Design – overall architecture (1 slide) Features of Your System - interface (1 slide) Features of Your System – other features (1 slide)

4. Administrative Homework 2 is posted  Deadline May 2, Wednesday midnight 11:59pm Peer Evaluations – due Friday, May 4, midnight  Peer Evaluation Form and Explanation - available on the class website  Submit your Peer Evaluation to klara@illinois.edu  Note: if you do not submit your peer evaluations, you get 0 for self-evaluation and 100% for your group mates. ¼ Unit projects – due Friday, May 4 midnight (if you need more time, arrange deadline with instructor) CS 414 - Spring 2012

5. Outline Summary of P2P File Sharing P2P Streaming CS 414 - Spring 2012

6. Napster Created in 1999 to share music CS 414 - Spring 2012 Napster’s servers Store file directory (Filename/location) Clients/peers Store files Centralized Peer Management: Each Peer registers with the Napster server and Napster server(s) keep Peer List Centralized File Management: Searching for other peer file is central, i.e., Peers go to Napster servers which keep file list of files and which peer keeps what files

7. Gnutella – Example of Unstructured P2P CS 414 - Spring 2012 Servents (peers) Peer pointer Peers store: Their files Peer pointers (peer management) Distributed Peer Management (Using Ping/Pong) Distributed File Management (Maintain my own files and search of others via flooding)

8. DHTs (Distributed Hash Tables) – Example of Structured P2P Hash table allows these operations on object identified by key:  Insert  Lookup  Delete Distributed Hash Table – same but in a distributed setting (object could be files) CS 414 - Spring 2012

9. DHT performance comparison MemoryLookup latencylookup overhead NapsterO(1) at client; O(N) at server O(1) GnutellaO(N) Chord (DHT)O(log(N)) CS 414 - Spring 2012

10. Streaming from servers Problem: Bandwidth at video service and number of servers have to grow with demand  Flash crowds have to be taken into account CS 414 - Spring 2012 … … clients Video service servers

11. P2P Streaming P2P Streaming is a response to elevate the demand on bandwidth in video servers  Issue: In-band and out-band bandwidth of peers (in/out bandwidth) P2P Streaming can distribute the bandwidth demand across peers  Issue: find peer that has enough out-band BW P2P Streaming will require management  Peer management  Chunk management P2P Streaming will require streaming distribution protocols

12. Peer Management for P2P Streaming One could use  Centralized peer management Live source keeps peer list, i.e., each peer registers with live source Separate session server keeps peer list, i.e., each peer registers with session server  Distributed peer management Peers advertise among each other and create peer list of its neighbors (Gnuttela-like) CS 414 - Spring 2012

13. Chunk Management for P2P Streaming Video is divided into chunks in P2P streaming  Chunk size can be size of GOP (group of pictures)  Chunk size can be agnostic to video semantics (e.g., 4K, or 8K or 32K chunk sizes). Peers hold chunks (not files) in P2P streaming Need chunk management  Centralized chunk management Server (live source or session manager) keeps information which peer has what chunks  Distributed chunk management Peers keep their own chunk table and other peers send queries to neighbors for requested chunks

14. P2P Streaming Use the participating node’s bandwidth  More nodes watching stream = more shared bandwidth App-level multicast  How? CS 414 - Spring 2012 Live stream source (could be a member of the p2p network) P2P network Peers watching stream

15. P2P Streaming Common arrangements to multicast the stream  Single Tree  Multiple Tree  Mesh-based  All nodes are usually interested in the stream They all have to deal with node dynamism (join/leave/fail/capacity-changes) CS 414 - Spring 2012

16. Streaming in a single tree CS 414 - Spring 2012 Source Frames codercoder 321 packets 3 2 1 3 2 1 (using RTP/UDP)

17. Single Tree Peers interested in the stream organize themselves into a tree CS 414 - Spring 2012 … …… Source Nodes send as many copies of a data packet as they have children

18. Joining the tree Find a node with spare capacity, then make it parent If contacted node lacks capacity, pick child according policy  Random child or  Round robin or  Child closest in physical network to joining node CS 414 - Spring 2012 “Parent?” “Try one of my children”

19. Leaving the tree or failing Orphan nodes need a new parent Policies for new parent  Children pick source  Subtree nodes pick source  Children pick grandfather  Subtree nodes pick grandfather  …then repeat join procedure CS 414 - Spring 2012 Orphan children after parent leaves Ex-parent grandfather

20. Single tree issues Leaves do not use their outgoing bandwidth Packets are lost while recovering after a parent leaves/fails Finding unsaturated peer could take a while Tree connections could be rearranged for better transfer CS 414 - Spring 2012

21. Multiple Trees Approach: a peer must be internal node in only one tree, leaf in the rest CS 414 - Spring 2012 1 2 2313 3 12 Source Are nodes 1, 2, 3 receiving the same data multiple times? -No, we stripe chunks across multiple nodes -Example: node 1 receives chunk 1, node 2 receives chunk 2, node 3 receives chunk 3 from source and then they distribute to other nodes in the subtree their chunks 1 3 2 2 31 1 3 2

22. Multiple Trees – Other Approach: Multiple Description Coding (MDC) Each description can be independently decoded (only one needed to reproduce audio/video)  More descriptions received result in higher quality CS 414 - Spring 2012 codercoder Frames 3030 2020 1010 Packets for description 0 Packets for description n … 3n3n 2n2n 1n1n

23. Streaming in multiple-trees using MDC Assume odd-bit/even-bit encoding -- description 0 derived from frame’s odd- bits, description 1 derived from frame’s even-bits CS 414 - Spring 2012 4 1 23 3 1 2 3030 2020 1010 3131 2121 1 (using RTP/UDP)

24. Multiple-Tree Issues Complex procedure to locate a potential- parent peer with spare out-degree  Degraded quality until a parent found in every tree Static mapping in trees, instead of choosing parents based on their (and my) bandwidth  An internal node can be a bottleneck CS 414 - Spring 2012

25. Mesh-based streaming Basic idea  Report to peers the packets that you have  Ask peers for the packets that you are missing  Adjust connections depending on in/out bandwidth CS 414 - Spring 2012 Description 0/1/2 Description 1 Description 0 Description 1,2 Description 0,1,2 (Nodes are randomly connected to their peers, instead of statically) Description 0/1/2 (mesh uses MDC)

26. Content delivery CS 414 - Spring 2012 1 4 101112 5 2 6 131415 7 3 8 1617 9 Description 0Description 2 Description 1 (1) Diffusion Phase ( ) (2) Swarming Phase ( ) (Levels determined by hops to source)

27. Diffusion Phase As a new segment (set of packets) of length L becomes available at source every L seconds  Level 1 nodes pull data units from source, then level 2 pulls from level 1, etc.  Recall that reporting and pulling are performed periodically CS 414 - Spring 2012 3030 2020 1010 … 32322 1212 4040 4242 Segment 0 Segment 1 … … Have segment 0 3 Send me Segment 0 2 1212 (during period 0) 3 Have segment 0 9 2 1212 (during period 1) Send me Segment 0 (drawings follow previous example)

28. Swarming Phase At the end of the diffusion all nodes have at least one data unit of the segment Pull missing data units from (swarm-parent) peers located at same or lower level Can node 9 pull new data units from node 16?  Node 9 cannot pull data in a single swarm interval CS 414 - Spring 2012 (drawings follow previous example) 1011 12 2 131415 7 1617 9 2020 1010 2121 1 21211 1 2121 1010 2020

29. Conclusion P2P streaming – strong alternative to CDN (Content Distribution Networks) networks Examples of P2P streaming technology  PPLive  Skype CS 414 - Spring 2012

30. Many more details in references and source code M. Castro, P. Druschel, A-M. Kermarrec, A. Nandi, A. Rowstron and A. Singh, "SplitStream: High-bandwidth multicast in a cooperative environment," SOSP 2003. H. Deshpande, M. Bawa, H. Garcia-Molina. "Streaming Live Media over Peers," Technical Report, Stanford InfoLab, 2002. N. Magharei, R. Rejaie. "PRIME: Peer-to-Peer Receiver drIven MEsh-Based Streaming," INFOCOM 2007. N. Magharei, R. Rejaie, Y. Guo. "Mesh or Multiple-Tree: A Comparative Study of Live P2P Streaming Approaches," INFOCOM 2007. http://freepastry.org CS 414 - Spring 2012