Title: Directed Diffusion: A Scalable and Robust Communication Paradigm for Sensor Networks
1Directed DiffusionA Scalable and Robust
Communication Paradigm for Sensor Networks
2Motivation
- Properties of Sensor Networks
- Data centric
- No central authority
- Resource constrained
- Nodes are tied to physical locations
- Nodes may not know the topology
- Nodes are generally stationary
- How can we get data from the sensors?
3Directed Diffusion
- Data centric
- Individual nodes are unimportant
- Request driven
- Sinks place requests as interests
- Sources satisfying the interest can be found
- Intermediate nodes route data toward sinks
- Localized repair and reinforcement
- Multi-path delivery for multiple sources, sinks,
and queries
4Motivating Example
- Sensor nodes are monitoring animals
- Users are interested in receiving data for all
4-legged creatures seen in a rectangle - Users specify the data rate
5Interest and Event Naming
- Query/interest
- Typefour-legged animal
- Interval20ms (event data rate)
- Duration10 seconds (time to cache)
- Rect-100, 100, 200, 400
- Reply
- Typefour-legged animal
- Instance elephant
- Location 125, 220
- Intensity 0.6
- Confidence 0.85
- Timestamp 012040
- Attribute-Value pairs, no advanced naming scheme
6Directed Diffusion
- Sinks broadcast interest to neighbors
- Initially specify a low data rate just to find
sources for minimal energy consumptions - Interests are cached by neighbors
- Gradients are set up pointing back to where
interests came from - Once a source receives an interest, it routes
measurements along gradients
7Interest Propagation
- Flood interest
- Constrained or Directional flooding based on
location is possible - Directional propagation based on previously
cached data
Gradient
Source
Interest
Sink
8Data Propagation
- Multipath routing
- Consider each gradients link quality
Gradient
Source
Data
Sink
9Reinforcement
- Reinforce one of the neighbor after receiving
initial data. - Neighbor who consistently performs better than
others - Neighbor from whom most events received
Gradient
Source
Data
Reinforcement
Sink
10Negative Reinforcement
- Explicitly degrade the path by re-sending
interest with lower data rate. - Time out Without periodic reinforcement, a
gradient will be torn down
Gradient
Source
Data
Reinforcement
Sink
11Summary of the protocol
12Sampling forwarding
- Sensors match signature waveforms from codebook
against observations - Sensors match data against interest cache,
compute highest event rate request from all
gradients, and (re) sample events at this rate - Receiving node
- Find matching entry in interest cache
- If no match, silently drop
- Check and update data cache (loop prevention,
aggregation) - Resend message along all the active gradients,
adjusting the frequency if necessary
13Design Considerations
14Evaluation
- ns2 simulation
- Modified 802.11 MAC for energy use calculation
- Idle time 35mW
- Receive 395mw
- Transmit 660mw
- Baselines
- Flooding
- Omniscient multicast A source multicast its
event to all sources using the shortest path
multicast tree - Do not consider the tree construction cost
15- Simulate node failures
- No overload
- Random node placement
- 50 to 250 nodes (increment by 50)
- 50 nodes are deployed in 160m 160m
- Increase the sensor field size to keep the
density constant for a larger number of nodes - 40m radio range
16Metrics
- Average dissipated energy
- Ratio of total energy expended per node to number
of distinct events received at sink - Measures average work budget
- Average delay
- Average one-way latency between event
transmission and reception at sink - Measures temporal accuracy of location estimates
- Both measured as functions of network size
17Average Dissipated Energy
They claim diffusion can outperform omniscient
multicast due to in-network processing
suppression. For example, multiple sources can
detect a four-legged animal in one area.
0.018
0.016
Flooding
0.014
0.012
0.01
0.008
Omniscient Multicast
(Joules/Node/Received Event)
Average Dissipated Energy
0.006
Diffusion
0.004
0.002
0
0
50
100
150
200
250
300
Network Size
18Impact of In-network Processing
0.025
Diffusion Without Suppression
0.02
0.015
(Joules/Node/Received Event)
Average Dissipated Energy
0.01
Diffusion With Suppression
0.005
0
0
50
100
150
200
250
300
Network Size
19Impact of Negative Reinforcement
0.012
0.01
Diffusion Without Negative Reinforcement
0.008
Average Dissipated Energy
(Joules/Node/Received Event)
0.006
0.004
Diffusion With Negative Reinforcement
0.002
0
0
50
100
150
200
250
300
Network Size
Reducing high-rate paths in steady state is
critical
20Average Dissipated Energy (802.11 energy model)
0.14
Diffusion
0.12
Omniscient Multicast
Flooding
0.1
0.08
Average Dissipated Energy
(Joules/Node/Received Event)
0.06
0.04
0.02
0
0
50
100
150
200
250
300
Network Size
Standard 802.11 is dominated by idle energy
21Average energy and delay
- Average delay is misleading
- Directed Diffusion is better than Omniscient
Multicast? - Why dont they suppress messages in Omniscient
Multicast as done in Directed Diffusion? - Topology has little path diversity
22Failures
- Dynamic failures
- 10-20 failure at any time
- Each source sends different signals
- lt20 delay increase, fairly robust
- Energy efficiency improves
- Reinforcement maintains adequate number of high
quality paths - Shouldnt it be done in the first place?
23Analysis
- Energy gains are dependent on 802.11 energy
assumptions - Can the network always deliver at the interests
requested rate? - Can diffusion handle overloads?
- Does reinforcement actually work?
24Conclusions
- Data-centric communication between sources and
sinks - Aggregation and duplicate suppression
- More thorough performance evaluation is required
25Extensions
- One-phase pull
- Propagate interest
- A receiving node pick the link that delivered the
interest first - Assumes the link bidirectionality
26- Push diffusion
- Sink does not flood interest
- Source detecting events disseminate exploratory
data across the network - Sink having corresponding interest reinforces one
of the paths
27TEEN (Threshold-sensitive Energy Efficient sensor
Network protocol) IPDPS01
- Push-based data centric protocol
- Nodes immediately transmit a sensed value
exceeding the threshold to its cluster head that
forwards the data to the sink
28LEACH HICSS00
- Proposed for continuous data gathering protocol
- Divide the network into clusters
- Cluster head periodically collect
aggregate/compress the data in the cluster using
TDMA - Periodically rotate cluster heads for load
balancing
29 Discussions
- Criteria to evaluate data-centric routing
protocols? - Or, what do we need to try to optimize? Energy
consumption? Data timeliness? Resilience?
Confidence of event detection? Too many
objectives already? Can we pick just one or two?
30Questions?