Gradient Setup Techniques

1

Gradient Setup Techniques Professor - Dr
Ajay Gupta Presented By -vivek Kinra CS691
Spring 2003 Source -http//www.cs.ucla.edu/classe
s/fall01/cs218/l1/project/student/ http//lecs.cs
.ucla.edu/estrin/talks
2
Recap

• Characteristics of a Sensor Network
• Data centric sensors are addressed by interests
  expressed by user queries
• Scalable no sensor has complete knowledge of the
  whole system
• Energy-sensitive sensors have limited battery
  power
• Data dissemination is application-specific data
  paths are determined by the needs of the
  application using the data
• Adaptive path selection may change as the energy
  level of certain paths decrease with use

3
Recap contd

• Sink-gtGenerate Interest-gtPropagate Interest
  throughout N/W.
• Sensor with that info (SOURCE) -gtrespond
  -gtcollecting required info-gtforwarding it back
  towards sink.
• User addresses entire N/W with there quires
  instead of global addressing scheme

4
contd

• Naming
• Gradient Setup
• Reinforcement (When, Whom, How Many, Negative
  reinforcement)

5
Illustrating Directed Diffusion
Setting up gradients
Source
Sink
6
Interest Propagation
Setting up gradients
Source
Sink
7
contd

• I-P also becomes G establishment
• Each node become sink for propagating interest to
  its neighbor
• Receiving path becomes sending path
• Upon receiving interest from a neighbor, an
  intermediate node will create a gradient for the
  link in reverse direction back to its neighbor

8
contd

• Interest Message may specify the loc of target
  query (x10, y4)
• Non spatial interest are more vague and may
  require more flooding
• Single Sink Case
• User identification,
• initiation time,
• max hop count, localized identification on the
  one previous hop away

9
Interest Propagation Steps

• 1st time -node will record interest and
  calculate gradient for the reverse path of the
  link to its neighbor (previous hop)
• Based on gradient just established, the sensor
  can determine if the reverse path likely to carry
  data back to the sink
• Hop count gt 0 and intermediate sensor decrease
  hop count and forwards interest
• Sensor doesnt forward interest twice

10
contd

• Intermediate sensors must forward the data along
  the link with the greatest gradient and decide
  whether to forward it to other links that may
  have a comparable tendency to reach the sink
• The source tags each piece of data with a
  timestamp so that intermediate nodes will drop
  duplicate data and prevent a data-forwarding loop
• Finally if the node can obtain the data in query,
  it becomes the source node and sent back the
  collected data.
• Multiple Sink Case

11
Gradient

• G in N/W shape the data routing paths to that
  network
• Also G establishment is determined by application
  req, diff paths may be established for same data
• Data aggregation-gtdata propagation more
  efficient. Multiple (sink or source)
• Energy awareness

12
Schemes

• Direction-based Scheme (COS)
• Data is directed towards the sink in a straight
  line with as little deviation as possible
• Each sensor knows its relative position
• The interest message includes the sinks position
  and the position of the previous hop neighbor

13
contd

• The gradient can be given by the function cos?
• The gradient 0 when ? 90 (when s?nb is
  perpendicular to s?sink)
• The gradient gt 0 when ? lt 90
• The gradient lt 0 when ? gt 90 (a negative
  gradient discourages data forwarding along that
  link)

cos
?
s
nb
?
neighbor
source
Sink
14
Distance-based Scheme (ttl)

• Data is directed towards the sink along the
  shortest path
• The interest message contains a maximum hop count
  that each intermediate sensor decreases as it
  forwards
• Based on this hop count, a node can determine its
  neighbors distance to the sink

15
contd

• Distance-based gradient is determined by
  1/(ttl1)
• (ttl1) is used because ttl 0 when nb is
  exactly the sink
• Distance-based gradient is the inverse of hop
  count

1/(ttl1)
s
nb
ttl hops
source
sink
16
Gravity based
Treat data forwarding like gravitational
attraction where the sink exerts a gravity that
is inversely proportional to the hop
count Gradients are calculated by
cos?/(ttl1)2 The gravity-based gradient is a
combination of the distance and direction
gradients
cos? /(ttl1)2
s
nb
?
source
sink
17
(2, 6)
(2, 7)
(2, 5)
(0, 4)
(2, 4)
(2, 3)
(0, 2)
(2, 2)
A sensor node
(2, 1)
(2, 0)
A sample topology
18
0
0.707
1
1
1
1
Data flow
A link
1
0.894
A source node
1
A sink node
1
A sensor node
1
Data flow with direction-based gradient
19
1/2
1/3
1/6
1
1/5
1/2
1/4
Data flow
A link
1/3
A source node
1/2
A sink node
1
A sensor node
1
Data flow with distance-based gradient
20
0
.079
.028
1
0.04
0.224
0.063
Data flow
A link
0.111
A source node
0.25
A sink node
1
A sensor node
1
Data flow with gravity-based gradient
21
Energy Awareness

• Consider Energy consumption level w/ gradient
• If same gradient but different energy, choose
  higher energy
• 2 Goals
• Each sensor minimize energy consumption
• Even out energy dissipation in entire network
• Energy Watermark determines when to consider
  minimizing energy

22
contd

• If sensor overused and has lower energy compared
  to other neighbors then link should be assigned a
  lower gradient to decrease the chance it will
  used again
• Sensor maintains up-to-date energy info of all
  its neighbors.
• Inefficient

23
Data Aggregation

• Duplicate interests to node from same link
• Save energy by eliminating duplicate packets
• Goal Send data on link which best satisfies all
  interests

24
Experiments

• 3 schemes ( COS, TTL, Gravity)
• Total Energy Consumption
• Success Ratio (delivery of packets)
• Data Aggregation
• Average energy consumption at each node

25
Experimental Setup Energy Consumption

• 10 random topologies (64 x 64)
• 7 sensor density in the network
• 3 schemes with/without Energy Awareness
• Guarantee 100 success ratio
• 99.9 Confidence Intervals
• Long Hop (7-10 hops)
• Short Hop (3-4 hops)

26
contd

• Each sensor has transmitting range of 8 units in
  the area
• Sensor uses -
• Size of message
• Energy units to transmit an interest
• Size of message energy units for data propagation

27
Ave Number of HopsLong Hop
28
Experimental Setup Aggregation

• 1 random topology (64x64)
• 6.9 sensor density in the network
• 3 different positions
• 3 schemes with/without Energy Awareness

U2
U2
S
S
S
U2
U1
U1
U1
29
Aggregation Energy Consumption
30
Conclusion

• Gravity Scheme uses less Energy, but unreliable
  when multiple users in different directions
• Energy Awareness ?doesnt improve the evenness
  of consumption per node

31
Naming schemes

• Internet where IP address provides low level
  names for routing
• Web and the search engines provide document and
  object naming scheme
• We need naming scheme that doesnt relay on
  network topologies
• Rather we need low level communication based on
  names that are external to N/W topologies

32
contd

• We need application relevant or it can be based
  on sensor types or geographic locations
• Attribute based naming system

33
Tiny Diffusion

• Implementation of Diffusion on resource
  constrained USB motes
• 8 bit CPU, 8k program memory, 512 bytes data
  memory
• Subsets of full system
• Retains only gradients and condenses attributes
  to a single tag
• Entire system runs for less than 5.5 KB memory

34
contd

• Tiny OS adds 3.5 KB and 144 bytes of data
  (inclusive support for radio and photo sensor
• Diffusion adds 2k code and 110 bytes of data to
  tiny OS

35
Tiny Diffusion Functionality

• Resource Constraint
• Limited Cache size-currently 10 entries of 2
  bytes each
• Limited ability to support multiple traffic
  stream. currently support 5 concurrently active
  gradients

36
TinyDiffusion Architecture
37
Description

• Fig shows interconnects b/n Filters,
  TinyDiffusion, AM, Timers and the photo
  components
• Filters connected -gt diff ports of the Timers
  which export alarms for periodic events
• SRC is connected to both 0 and 1 port of Timers

38
conts

• Filters connect at different ports to the
  TinyDiffusion to send and receive packets.
• They specify which attribute types they are
  interested in AND when the attribute type in
  packed matches the specified type, the packed is
  sent to the corresponding filter.
• Active Messaging layer

39
(No Transcript)
40
Gateway Architecture
MOTE ATMEL 8586 4MHz MCU 8K program memory 512
Bytes Data Memory RFM Radio 900 MHz
PC104 AMD ElanSC400 66MHz CPU 16MB RAM Form
Factor 3.6"  x  3.8"  x  0.6"
41
Tiered Testbed

• PC-104(linux) with MoteNIC
• Tags, Sensor Card
• UCB Motes w/TinyOS
• Yet to come SmartDust (highly specialized nodes)

PC104
TAG
USB Mote