Network Coding A Brief Tutorial

1
Network CodingA Brief Tutorial

• Louai Al-Awami
• PhD Student
• Department of Electrical Computer Eng.
• Queen's University

2
Outline

• What is Network Coding (NC)
• Benefits of NC
• Design Issues with NC
• Types of Network Codes
• Applications of NC
• Conclusion

3
Network Coding (NC)

• Traditionally, coding is done on End-to-End
  bases.
• Intermediate nodes forward data blindly.
• Network coding removes this restriction by
  allowing intermediate nodes to manipulate data
  for the sake of better performance.

Network
4
Example 1
5
Example 2
6
Outline

• What is Network Coding (NC)
• Benefits of NC
• Design Issues with NC
• Types of Network Codes
• Applications of NC
• Conclusion

7
Benefits of NC

• Throughput
• Reliability
• Flexibility

8
Benefits of NC

• Throughput
• Increasing information content per transfer.
• Minimizing the number of transfers
• More bandwidth
• Less delay

9
Benefits of NC

• Flexibility
• Information is disseminated on all packets.
• results on all packets being equally important.
• Information can be retrieved using any set of
  packets no matter which.
• E.g. implicit ACK

10
Benefits of NC

• Reliability
• ARQ results in high delay
• FEC results in low rate
• Similar to information spreading/interleaving
• Duplicate information (overhearing) helps
  checking integrity of received data.

11
Outline

• What is Network Coding (NC)
• Benefits of NC
• Design Issues with NC
• Types of Network Codes
• Applications of NC
• Conclusion

12
Design Issues

• Encoding
• Let M1, M2, ....., Mn be messages to be sent
• Let g ltg1, g2, ....., gn gt be the encoding
  vector
• Then encoding can be done

13
Design Issues

• Decoding
• Can be viewed as solving a system of linear
  equations for the unknowns M1, M2, ....., Mn
• Done using decoding matrix
• Initially contains local data .
• Progressively, add received packets to the
  matrix.
• Perform Gaussian elimination.
• Whenever a row becomes of the form (ei,Mi), Mi is
  decoded.

14
Design Issues

• Decoding (Delay)
• We do not have always to wait for all packets
• Worst case will have average delay like ordinary
  encoding (wait for all packets).
• Do not forget the reduction in number of
  transfers --gt lower delay.

15
Design Issues

• Decoding (Matrix Size)
• Overcome using generations.
• Packets of the same generations are combined
  together.
• Larger matrix means better performance but higher
  memory requirements.
• It is a trade off between performance and cost.

16
Outline

• What is Network Coding (NC)
• Benefits of NC
• Design Issues with NC
• Types of Network Codes
• Applications of NC
• Conclusion

17
Types of Network Codes

• Deterministic Codes
• Random Codes

18
Types of Network Codes

• Deterministic Codes
• Encoding coefficients are know beforehand.
• No need to append decoding vector to packets.
• Decoding matrix size is fixed.
• Nodes can identify source of packets through
  decoding vector.

19
Types of Network Codes

• Random Codes
• Random coefficients are generated on-the-fly.
• Packets contains coefficient vector.
• Decoding matrix size can be dynamic.

20
Outline

• What is Network Coding (NC)
• Benefits of NC
• Design Issues with NC
• Types of Network Codes
• Applications of NC
• Conclusion

21
Applications of NC

• P2P Networks
• Wireless Networks
• Sensor Networks
• Network Monitoring
• Network Security

22
Applications of NC

• P2P Networks
• in P2P networks, server divides files into
  smaller blocks.
• Clients download files and distributes them among
  themselves.
• In Avalanche (developed by MS), server sends
  linear combinations of blocks.

23
Applications of NC

• P2P Networks
• Advantages over traditional P2P
• Reduced download time.
• Avalanche was shown to reduce download time by
  20-30
• Robustness against early server disappearance and
  high churn rates.
• Less suffering when cooperation-forcing mechanism
  is implemented.

24
Applications of NC

• Wireless Networks
• Bidirectional Traffic and Multiple Streams in
  Wireless Networks
• As we saw in the first example
• If multiple streams are coming to a router, it
  can mix traffic of multiple streams and send them
  at once.
• Bidirectional traffic can be very useful for
  telephony and video applications
• Implicit acknowledgments
• Experiment shows doubling throughput for IEEE
  802.11.

25
Applications of NC

• Sensor Networks
• Data Gathering
• The goal of some data gathering algorithms is to
  disseminate data over all nodes in order for the
  sink node to retrieve it from any node.
  (expensive memory requirements)
• With NC only 1 memory unit is needed per node,
  plus the sink node can retrieve it from any n
  nodes.
• Nodes broadcast their data and others multiplies
  it with their own data.

26
Applications of NC

• Network Monitoring
• Network monitoring keeps track of which nodes are
  operational/failed
• Existing solution use probes that send requests
  to nodes periodically (active monitoring)
• With NC monitoring nodes can analyze packets to
  know who is there (passive monitoring)

27
Applications of NC

• Network Security
• Eavesdropping
• Data is fully distributed, it is more difficult
  to grab.
• Modifying Data
• By replicating data, it is easy to check the
  integrity by using checking against multiple
  copies.

28
Conclusion

• What is Network Coding (NC)
• Benefits of NC
• Design Issues with NC
• Types of Network Codes
• Applications of NC
• Conclusion