Comparison of Raptor Codes with ARQ for Video Streaming on PeertoPeer Networks

1
Comparison of Raptor Codes with ARQ for Video
Streaming on Peer-to-Peer Networks

• Rohit Watve
• Deepan Gandhi

2
Outline

• Overview of Raptor codes
• Overview of SPPM
• Modeling for Raptor Code Scheme
• Modeling for ARQ Scheme
• Model evaluations
• Experimental results
• Conclusion

3
Overview of Raptor Codes

• A Type of Fountain Codes
• Suitable for Erasure Channels
• Rateless Code Variable Redundancy
• Systematic Coding is possible

4
Stanford P2P Multicast Protocol (SPPM)

• Multiple Tree-based protocol
• Losses can occur due to ungraceful disconnection
• Raptor Codes can be beneficial if simultaneous
  tree disconnections are limited

5
Model Parameters

• Homogenous Peers
• Balanced Trees

6
Modeling for Raptor Code Scheme

• A node is an interior node of only one tree
• N Total Number of nodes
• T Total Number of trees
• d depth of tree
• f fanout (number of children per peer)

7
Modeling for Raptor Code Scheme

• B Uplink Bandwidth
• r Data rate on each tree
• R Total Video Rate
• rt Maximum number of tree losses supported

8
Modeling for Raptor Code Scheme

• p Probability of node leaving ungracefully
• Pd Probability of getting disconnected from a
  tree
• Ploss Probability of losing a packet

9
Modeling for Raptor Code Scheme

• Worst Case Delay

10
Modeling for ARQ scheme

• High uplink bandwidth is required for raptor
  codes
• Same Bandwidth assumed here for fair comparison
• Losses due to congestion caused by retransmission
  are very small verified by model
• Model without considering congestion described
  here

11
Modeling for ARQ scheme

• Initially, let us consider only one tree
  disconnection
• probability that a node will be disconnected from
  a tree because of a node at the level m
• p(1-p)m-1
• Total number of other leaf nodes disconnected
  from this tree
• Total number of leaf nodes

12
Modeling for ARQ scheme

• Probability of loss for this particular case
• RetransNum the number of times retransmission
  requests are allowed

13
Modeling for ARQ scheme

• Probability that a packet will be lost
• Worst case delay additional waiting time

14
Model evaluation

• Uplink bandwidth 800kbps, Video rate 490kbps,
  Total nodes 10000, Number of trees varied from
  5 to 20, Node disconnection probability 0.05

15
Model evaluation More Analysis

• Uplink bandwidth 800kbps, Video rate 490kbps,
  Total nodes 10000, Number of trees varied from
  5 to 20, Node disconnection probability 0.05

16
Model evaluation More Analysis

• Uplink bandwidth 800kbps, Video rate 490kbps,
  Total nodes 10000, Number of trees varied from
  5 to 20, Node disconnection probability 0.05

17
Model evaluation More Analysis

• Uplink bandwidth 800kbps, Video rate 490kbps,
  Total nodes 10000, Number of trees varied from
  5 to 20, Node disconnection probability 0.05

18
Model evaluation

• Number of retransmissions 2 (other parameters
  same as before).

19
Model evaluation

• Uplink bandwidth 800kbps, Video rate 490kbps,
  Total nodes 10000, Number of trees varied from
  5 to 20, Node disconnection probability 0.1

20
Experimental results

• Uplink bandwidth 800kbps, Video rate 490kbps,
  Total nodes 80, Number of trees 8, Node
  disconnection probability 0.01

21
Experimental results

• Uplink bandwidth 800kbps, Video rate 490kbps,
  Total nodes 80, Number of trees 8, Node
  disconnection probability 0.1

22
Experimental results

• Uplink bandwidth 800kbps, Video rate 490kbps,
  Total nodes 80, Node disconnection probability
  0.01, delay 1000ms

23
Conclusion

• Loss probability can be decreased if more delay
  is acceptable
• More number of trees result into more delay but
  less loss probability
• Raptor codes more suitable for tight delay
  constraint and less disconnection probability
• Scheme with more than one retransmission is
  superior in terms of loss probability

24
Thank You