Jiangchuan Liu with Xinyan Zhang, Bo Li, and T.S.P.Yum

1
CoolStreaming/DONet A Data-Driven Overlay
Network for Peer-to-Peer Live Media Streaming

• Jiangchuan Liu with Xinyan Zhang, Bo Li, and
  T.S.P.Yum
• Hong Kong University of Science and Technology,
  Clear Water Bay, Hong Kong
• Infocom 2005
• Lyu-Han Chen
• 
  2008/04/21

2
Some Facts

• DONet Data-driven Overlay Network
• CoolStreaming Cooperative Overlay Streaming
• Till March 2005
• Average online users 6,000
• Peak-time online users 14,000

3
Outline

• Motivation
• Background and related work
• Design of DONet/CoolStreaming
• Performance evaluation and implementation
• Conclusion

4
Motivation

• Enable large-scale live broadcasting in the
  Internet environment
• Capacity limitation
• Streaming 500Kbps, server outbound band 100Mbps
• Network heterogeneity
• No QoS guarantee

5
Client/Server Poor scalability
6
IP multicast Limited deployment
7
Collaborative Communications
8
Data-driven Overlay

• Data-centric design
• No prescribed roles like parent/child,
  upstreaming/downstreaming.
• The availability of data that guides the flow
  directions.

9
Data-driven Overlay (DONet)

• Target
• Live media broadcasting
• No IP multicast support
• Core operations
• Every node periodically exchanges data
  availability information with a set of partners
• Then retrieves unavailable data from one or more
  partners, or supplies available data to partners

10
Features of DONet

• Easy to implement
• no need to construct and maintain a complex
  global structure
• Efficient
• data forwarding is dynamically determined
  according to data availability, not restricted by
  specific directions
• Robust and resilient
• adaptive and quick switching among
  multi-suppliers

11
Outline

• Motivation
• Background and related work
• Design of DONet/CoolStreaming
• Implementation and empirical Study
• Conclusion

12
Related Solutions

• Content distribution networks
• Expensive
• Not quite scalable for a large number of
  audiences
• Self-organized overlay networks
• Application layer multicast
• Peer-to-peer communications

13
Application Layer Multicast

• Issue Node dynamics
• Structure maintenance
• Passive/proactive repairing algorithms
• Advanced coding

14
Tree-based protocols

• Centralized algorithm relies on a powerful root
  node.
• Experience frequent breaks in the highly dynamic
  Internet environment.

15
Gossip-based Dissemination

• Gossip
• Iteration
• Sends a new message to a random set of nodes
• Each node does similarly in the next round
• Pros Simple, robust
• Cons Redundancy, delay

16
Outline

• Motivation
• Background and related work
• Design of DONet/CoolStreaming
• Implementation and empirical Study
• Conclusion

17
Key Modules

• Membership manager
• mCache partial overlay view
• Update by gossip
• Partnership manager
• Random selection
• Partner refinement
• Transmission Scheduler

18
Node Join / Membership Management

• mCache (membership information)
• Each node periodically generates a membership
  message to announce it existence.
• lt seq_num, id, num_partner, time_to_live,
  (last_update_time) gt

partners
19
Buffer Map

• The availability of the segments of the buffer of
  a node can be represented by a Buffer Map.
• Sliding window 1, 1, 0, 0, 0, 1, 1, , 0
• Each node continuously exchange its BM with the
  partners.
• Schedules which segment is to be fetched from
  which partner accordingly.

20
Transmission Scheduling

• Problem From which partner to fetch which data
  segment ?
• Constraints
• Data availability
• Playback deadline
• Heterogeneous partner bandwidth

21
Scheduling algorithm

• Variation of Parallel machine scheduling
• NP-hard
• Heuristic
• Message exchanged
• Window-based buffer map (BM) Data availability
• Segment request (BM-like bit sequence) (piggyback
  by the data packets)
• Less suppliers first
• Multi-supplier Highest bandwidth within deadline
  first

22
Failure Recovery

• Graceful departure
• Node failure
• Choose better score partner
• More available segments
• Higher bandwidth

23
Analysis of Overlay Radius

• Reflect the end-to-end delivering latency.
• Modeled by a Breath-First Search tree.

24
Analysis of Overlay Radius
25
Analysis of Overlay Radius
26
Outline

• Motivation
• Background and related work
• Design of DONet/CoolStreaming
• Performance evaluation and implementation
• Conclusion

27
PlanetLab Experiments
28
Control overhead
29
Continuity index
30
Control overhead
31
Continuity index
32
Comparison with Tree-based Overlay
33
Continuity index
34
Online Statistics (Jun 21, 2004)
35
Outline

• Motivation
• Background and related work
• Design of DONet/CoolStreaming
• Implementation and empirical Study
• Conclusion

36
Conclusion