Reliability

1
Reliability

• Standard approaches
• Difference where to resend
• End to end (resend at edges)
• Hop by hop (resend at each hop)
• In sensor networks, not every bit has the same
  value
• Same goes for different nodes

2
Reliability (contd)

• Proposed approach Information-driven reliability
• When
• Is latency a concern?
• Is the information time-sensitive?
• What
• Is the information critical?
• Can the application operate fairly correctly
  without it?
• Where
• Which nodes along the path should retransmit?
• Who
• Clusterheads, aggregators and other important
  nodes in the path can request retransmission from
  neighbors
• Receiver-initiated reliability

3
First steps JR deployment

• Two major message types
• Control
• Requirement Low latency
• Data
• Requirement Energy efficiency

4
Control messages Reliability

• What we have
• RNC
• broadcast (or 1-hop multicast)
• Relatively low latency
• (at its current state) Energy inefficient
• Needs to be improved
• Less duplicates
• More energy efficient
• TinyRNC
• more?
• SMAC
• unicast
• Energy efficient
• More latency? (sleep cycles)

5
Data messages Reliability

• What we want
• Reliable low energy data
• Receiver driven
• Receiver(s) control sources
• How far along the tree do we go?
• All the way (end 2 end)?
• 1 hop?
• some hops until an aggregator is reached?
• Policy driven selective NACK

6
Different reliability policies

• Time/space width matrix
• Narrow time-narrow space
• Standard reliability (end to end)
• Narrow time-wide space
• Reverse anycast Any one of N spacially
  distributed nodes packets reliabily gets to
  destination
• Wide time-narrow space
• Statistical reliabilty some samples out of the
  data series (or some messages in general)
  reliably get to destination
• Wide time-wide space
• No reliability best effort

7
Applications and Reliability reqs

• Localization-tracking (Hanbio)
• TCP-like
• Mulihop
• 100 delivery (all data)
• Microclimate monitoring (JR)
• End-to-end through hop-by-hop (link-level
  retransmissions)
• 100 delivery
• Spatial aggregation (Simon)
• Link-level (single-hop extended to multihop)
• Estimation/error driven
• Combination of information importance and energy
  cost
• Not necessarily 100

8
Thoughts

• Application-specific reliability
• Application drives the reliability model
• Information importance only available through app
• Link quality determines energy cost
• Can be provided by a linkstats model
• Optimization problem (?)
• Given data importance cost, determine optimal
  retransmission request

9
Data importance

• If the spatial frequency of the physical
  phenomenon is known, things are relatively easy
• Dense deployment, redundant information
• What if it isnt?
• Entropy is one way to characterise importance
  (useful information carried per packet)
• Completely application dependent
• Any other way?
• What is known about the phenomenon? What needs to
  be known?
• Continuous refresh mode Data importance (and
  link quality) must be continuously computed for
  best accuracy

10
A slightly different approach

• Treat the retransmission selection problem as a
  data generation problem
• Deepaks suggestion
• Selects which nodes should generate data, out of
  those that are retransmission candidates
• For Deepaks application, can potentially use the
  same metric