Communication Paradigm for Sensor Networks

1
Communication Paradigm for Sensor Networks

• Sensor Networks
• Directed Diffusion
• SPIN

Ishan Banerjee ishan_at_cs.umd.edu
2
Sensor Networks
Conventional Networks

• Wired network
• Infinite power source
• Rapidly increasing bandwidth
• High performance workstations
• Attended nodes
• Low node to user ratio
• Manually configurable hosts
• Hosts are reparable and replaceable
• Complex global routing schemes
• Fixed, named nodes

3
Sensor Networks
Application of Sensor Networks

• Gathering accurate information in a distributed
  manner from
• Inaccessible geographic area
• Disaster area
• Industrial location
• Object tracking
• Traffic data
• Remote surveillance
• Global objective with local interaction

4
Sensor Networks
Current sensing methods
Architecture
Object
Signal analysis
Sensors

• Complex sensors far from object
• Sensors generate stream of data
• Sensors without computing power
• Signal processing to separate signal from noise
• Low signal to noise

5
Sensor Networks
Current sensing methods
Architecture
Object
Signal analysis
Sensors

• Sensors close to object
• Sensors generate stream of data
• Sensors without computing power
• Better signal to noise

6
Sensor Networks
Sensor Networks
Architecture
Object
Event analysis
Sensors Net

• Sensors net close to object
• Observation of each sensor is processed in-situ
• Sensors coordinate to make observation
• Tells host about result of observation

7
Sensor Networks
Sensor Networks

• Objectives
• Match-box sized devices
• In network processing
• Better Signal-tonoise ratio
• Extend life of devices
• Highly scalable
• Responsive to dynamic and hostile environment
• Implications
• Fixed wire-less network
• Low bandwidth. Avoid long distance
  communications
• No user attendance
• Deployed in large numbers
• Requires self configuration
• Device failure implies removal from network
• Requires simple energy efficient routing

8
Sensor Networks
Paradigm

• Data Centric
• Sensors net is queried for specific data
• Source of data is irrelevant
• No sensor-specific query
• Application Specific
• In-sensor processing to reduce data transmitted
• In-sensor caching
• Localized Algorithms
• Maintain minimum local connectivity save
  energy
• Achieve global objective through local
  coordination

9
Directed diffusion
Directed diffusion

• PULL model for obtaining information from a
  sensor-net

Object
Sensors

• Better than flooding, multicast
• Energy efficient
• Delay comparable to multicast
• Failure tolerant

10
Directed diffusion
Data naming

• Content based naming
• No globally unique ID for nodes (sensors)
• Name of sensors are irrelevant ephemeral nodes
• Task are named Attribute value pair
• Selecting naming scheme is important for the
  sensor net

Request
Interest ( Task ) Description Type temperature
increase Threshold 200 C Interval 100
ms Duration 10 hours Location -100, -100
100, 100
Reply
Node data Type temperature increase Intensity
5 C / sec Location 41, 73 Confidence
0.8 Time 101035
11
Directed diffusion
Interests

• Interest describes a task required to be done by
  the sensor-net
• Interest is injected at some point, sink
• Source is unknown at this point
• Interest diffuses through the network hop-by-hop
• Interest is broadcast by a node to its
  neighbours
• Loops are not checked for at this stage

12
a
Directed diffusion
c
b
Diffusion Gradient setup
13
Directed diffusion
Gradient setup
Object

• Interest diffuses through network
• Interest does not specify node information
  leads to scalability
• Caching is done to reduce traffic
• Specifies a data rate and a direction
• No global knowledge of the topology used
• Nodes aware only of neighbours
• Strictly local interaction
• Exhibits PULL paradigm

14
Directed diffusion
Data propagation
Source
Source

• In-situ processing is performed to identify
  event
• Data sent back is an event indication only low
  bandwidth
• Caching is used for loop detection

15
Directed diffusion
Reinforcement
Source
Source

• Sink may receive data from multiple sources
• Local rules are used to increase the data rate
  from a subset
• Done by sending renewed interest with higher
  rate
• Empirically determined path is reinforced
• Negative reinforcement used to close multiple
  paths

16
Directed diffusion
Performance

• DD
• Omniscient multicast
• Flooding

Key metric is dissipated energy per event received
Directed diffusion compared to flooding and
omniscient multicast Directed Diffusion A
scalable and Robust Communication Paradigm for
Sensor Networks http//lecs.cs.ucla.edu/estrin/pa
pers/diffusion.ps
17
Directed diffusion
Performance

• DD
• Omniscient multicast
• Flooding

Impact of node failure on directed
diffusion. Directed Diffusion A scalable and
Robust Communication Paradigm for Sensor
Networks http//lecs.cs.ucla.edu/estrin/papers/di
ffusion.ps
18
SPIN
Sensor Protocol for Information via Negotiation

• PUSH model for disseminating information to all
  nodes of a sensor-net

Detect
Object

• Broadcast of data
• Energy constrained network
• Limited computation capability
• Low bandwidth

19
SPIN
Broadcast characteristics
Detect
Object
Detect
Object
Implosion
Detect
Overlap
20
SPIN
SPIN Philosophy

• Application level framing
• Negotiation using meta-data
• Meta data describes actual data
• Used for negotiations
• Messages
• Advertise
• Request
• Data transfer
• Resource management
• Resource aware
• Protocols executed after considering energy

21
SPIN
SPIN-PP 3 way handshake
REQ

• Simple
• Adv, Req, Data
• Point-to-point

ADV
DATA
ADV

• Extended to energy aware variant
• May not participate in protocol if power too low

REQ
DATA
22
SPIN
SPIN-BC 3 way handshake
B
C
A
A and C suppress their REQ
ADV
B
C
A
B
REQ
C
A
DATA
23
SPIN
Performance
Data acquired by network over time
Corresponding Energy dissipated
24
Comments

• Demonstrate simple concepts in new domain
• Primary concern is energy usage
• Simulations only
• Assumed congestion free network