On Exploiting Asymmetric Wireless Links via Oneway Estimation

1
On Exploiting Asymmetric Wireless Links via
One-way Estimation

• Lifeng Sang Anish Arora Hongwei Zhang

• Mobihoc 2007

2
Motivation

• Basic problem
• Reliability of data transfer, e.g. via
  convergecast routing, remains significantly below
  100 in many WSN scenarios
• Complicates assuring application quality
• Traditional solution bidirectional link model
• Estimates link quality via asynchronous beacons
• Avoids using asymmetric links

3
Our contribution

• Systematic experimental study of link asymmetry
  sync ACK
• Substantial age of links are asymmetric
• Significant age of links are unidirectional
• Reliability of sync ACK gt async ACK
• Propose one-way link estimation metric, ETF
• Experimental study of performance of convergecast
  routing that exploits asymmetric links
• ETF significantly improves performance over ETX
  on same convergecast protocol
• Holds with/without unidirectional link discovery

4
Outline

• Empirical study
• Link asymmetry
• Synchronous ACK
• Design of one-way metric ETF
• Link discovery
• Error control
• Experimental evaluation
• Same convergecast routing protocol, different
  metric
• Different density
• Different traffic trace
• Summary

5
Empirical study of link asymmetry

• Testbed
• 7 x 7 subgrid in Kansei, mimic the setup of a
  field sensor network
• XSM sensor motes
• 4MHz ATmega128L microcontroller, Chipcon CC1000
  radio, 128KB of flash, 4KB of RAM, running TinyOS
• Data Sets
• Round-robin broadcast experiments (about 2
  hours each), at different power level, collected
  at morning, afternoon, evening, and midnight

6
Empirical study of link asymmetry
symmetric
asymmetric
unidirectional

• Many links are asymmetric
• Traditional techniques tend to ignore asymmetric
  links
• Lower transmission power --gt more asymmetric links

• Symmetric links short asymmetric links long
• Exploiting asymmetric links can lead to more
  efficient routing

7
Outline

• Empirical study
• Link asymmetry
• Synchronous ACK
• Design of one-way metric ETF
• Link discovery
• Error control
• Experimental evaluation
• Same convergecast routing protocol, different
  metric
• Different density
• Different traffic trace
• Summary

8
Empirical study of synchronous ACK

• The link quality for synchronous ACK is better
  than that for beacons because
• Nodes in the neighborhood are backing off
• Size of sync ACK is usually much smaller joint
  reliability is higher
• Experiments on links with different quality
  (good, medium, bad) and different distance (long,
  short)
• Sync/Async ACK without intentional interference
• Sync/Async ACK with interference

9
Reliability of synchronous ACKs

• Significant improvement of using sync ACK over
  async messages, especially in the presence of
  interference
• Improvement occurs on both short and long links
• gt Norm of estimating link quality in both
  directions via async beacons underestimates the
  link reliability of asymmetric links

10
Outline

• Empirical study
• Link asymmetry
• Synchronous ACK
• Design of one-way metric ETF
• Link discovery
• Error control
• Experimental evaluation
• Same convergecast routing protocol, different
  metric
• Different density
• Different traffic trace
• Summary

11
What is ETF

• ETF expected of data transmissions required
  for a data packet over a forward link, without
  considering the delivery ratio of ACK packets

p(t) forward reliability, estimated in a window
of time
12
How to enhance ETF

• Link discovery
• How to detect asymmetric links that have good
  forward reliability but poor backward
  reliability?
• lt neighbor collaboration
• Error control
• How to deal with ack loss (especially for
  asymmetric links with low backward reliability)?
• lt dynamic retransmission threshold

13
Overhead and benefit

• Communication overhead
• Negligible discovery information is piggy-backed
  in beacons
• Relatively low discovery information is
  implemented as unicast control packets
• Benefit of using ETF
• Analytical comparison of ETF and ETX on expected
  number of transmissions

Cases where ETF is worse
Cases where ETF is better
14
Outline

• Empirical study
• Link asymmetry
• Synchronous ACK
• Design of one-way metric ETF
• Link discovery
• Error control
• Experimental evaluation
• Same convergecast routing protocol, different
  metric
• Different density
• Different traffic trace
• Summary

15
Experimental evaluation

• Compare under same convergecast routing protocol
• ETX metric ETX
• ETF metric ETF
• ETF-NU same as ETF, except that the explicit
  asymmetric-link discovery is disabled
• Performance metric
• End-to-End Reliability, Number of Transmissions
  Per Packet (TXPP)
• End-to-End Latency, Duplicates, Hop Count
• Traffic trace
• Lites an event traffic trace collected in a
  real-world intrusion detection sensor network
• Periodic traffic

16
Testbed density

• Power level 9 one can talk to almost everyone
  else
• Power level 1 one can only talk to a few nearby
• Power level 3 medium density, gives us a typical
  3-5 hops multihop network

17
Networks of different density
gt Significant improvement using ETF over ETX
18
Impact of link discovery

• ETF-NU achieves better performance than ETX
• Link discovery is effective
• ETF tends to choose links with higher reliability
  and stability, and they are relatively long

19
Impact of Maximum Transmission Count (C )
gt ETF outperforms ETX regardless of changing C
20
Evaluation on periodic traffic
gt Again, ETF significantly outperforms ETX
21
Summary

• Conclusion
• Systematically study on link asymmetry and sync
  ACK
• Propose one-way link metric ETF and its design
• Evaluate via testbed experiments ETF
  significantly improves the performance of
  functions such as convergecast routing in sensor
  networks
• Further questions
• Whether ETF is beneficial for routing problem in
  general?
• How about link asymmetry in different environment
  (mobile networks, other platforms, field sensor
  networks)?

22
Thank You!