Thesis Plan

1
Thesis Plan

• Lewis Girod
• 17 Sept 2003

2
Thesis in a Nutshell

• Develop and study mechanisms to support
  in-network coordination
• Mechanisms centered around use of clusters and
  identification of stable topology
• (Generation of stable topology work by Alberto
  Cerpa)
• Models of uncertainty as basis for control loop
• Sample application used to demonstrate utility
• Ranging application on possibility
• Other applications might be more useful?

3
Examples of Coordination Tools

• Codebook compression
• Multihop state sync
• Multihop resource discovery

4
Codebook compression

• Used to reduce data transmission
• Shared codebook with broadcast domain enables
  repeated data to be replaced by shorter codes
• Reduces the cost of soft-state refresh
• All the benefits of soft-state simplicity at
  lower cost
• How it works
• Assumes clustered topology
• Maintain shared state across wireless link to
  cluster head

5
Multihop state synchronization

• Used by node to publish slowly varying state info
• API allows nodes to publish key-value pairs
• KVPs visible to receiving nodes, notification on
  update
• Reliable convergence to steady state
• Does not guarantee all state transitions seen
• Cheaper, faster than soft-state equivalent
• No need to refresh old data sequence number
  refreshed
• Faster recovery time from errors explicit retx

6
Multihop resource discovery

• Annotated neighbor discovery
• Group membership, e.g. imagers, NIMS nodes
• Current capabilities, power levels, etc.
• Can work across multiple hops
• Multicast groups?

7
Importance of Clusters

• Many of these concepts assume stable, clustered
  topology
• Alberto Cerpa working on clustering and topology
  control in parallel

8
Uncertainty and Control

• Elements of distributed systems maintain local
  estimates of attributes of the world,
• e.g. link quality to neighbors
• Local estimates
• Often require discrete measurements
• Measurement entails cost
• e.g. pinging a neighbor
• Model uncertainty of estimate to decide when to
  pay cost of experiment

9
Uncertainty Models

• Uncertainty changes as a function of
• Time
• Time awake as well as time cycled off..
• Power
• Remaining power, usage rate, estimated system
  lifetime
• Rate of environmental change
• Including node mobility
• Serendipitous acquisition of app-specific
  evidence
• e.g. reception of packet from neighbor, or
  reception of reference to neighbor from third
  party

10
Uncertainty based control loop

• Explicitly represent uncertainty of estimate
• e.g. neighbor probability rather than binary (or
  binary uncertainty estimate)
• Evidence from other sources filters into
  uncertainty
• e.g. monitor packet stream
• Actuate measurement with a probability derived
  from the uncertainty
• Similar to ideas in robotics, but applied to
  coordination protocols
• Control loop itself modified by interest from
  apps
• e.g. different apps might have different
  operating points in terms of certainty of info

11
Sample application

• Want to start pulling together sample app early
  in process it motivates the work
• Originally planned to do acoustic ranging
• Perhaps other apps are needed more urgently?
• Triggering imager needs resource discovery and
  possibly group communication to imagers?
• Calibration data fits sync model well
• Diffusion interest reduction fits codebook?
• Perhaps other coordination primitives needed?

12
Related work

• Satish Kumar work in clustering
• Use of uncertainty in robotics
• Bayesian networks
• MHR sensor uncertainty models
• Various sensor net reliable transport protocols
• Google search yields RNC, RMST, ESRT, PSFQ
• Differences
• Assumption of clustered topology
• State sync semantics