1.The Impact Of Data Aggregation in Wireless Sensor Networks. 2.The ACQUIRE Mechanism for Efficient Querying In Sensor Networks.

1
1.The Impact Of Data Aggregation in Wireless
Sensor Networks.2.The ACQUIRE Mechanism for
Efficient Querying In Sensor Networks.

• By
• Kinnary Jangla
• Rishi Kant Sharda

2
Date 04-19-06
The Impact of Data Aggregation in Wireless Sensor
Networks

• Paper By
• - Bhaskar Krishnamachari
• - Deborah Estrin
• - Stephen Wicker
• Presented By
• - Kinnary Jangla
• - Rishi Kant Sharda
  

3
Basic Idea..

• To exploit the data redundancy
• Packets from different nodes, are combined in
  network.
• Implementation
• Who carries the data with redundancy
• Data-centric routing
• Differences
• Data-centric routing
• Based on contents of the packets.
• Address-centric routing
• Routing based on an end-to-end manner.

4
The Impact Of Data Aggregation On Wireless Sensor
Networks
Overview

• Sensor Network Models
• Event-Radius Model
• Random Source Models
• Impact of
• Source-Destination Placements
• Communication Network Density
• On
• - Energy Costs
• - Delay

5
(Cont..)
The
Impact Of Data Aggregation On Wireless Sensor
Networks

• Data Centric routing - Significant
  Performance Gain
• Complexity of Data Aggregation
• NP-Hard Problem.

6
Sub - Titles
The Impact Of Data Aggregation On
Wireless Sensor Networks

• Introduction.
• Routing Models.
• AC
• DC
• Data-Aggregation
• Optimal Suboptimal Aggregation
• Sensor Network Models
• Energy Savings
• Theoretical Results
• Simulation Results
• Delay

7
Introduction.
?
??
?
?

• Concepts.
• Sensor Network ?
• Sensor Node ?
• Unattended Operation ?
• Data Aggregation ?
• Data Redundancy !
• Wireless Sensor Network.
• Applications.
• Network Topology of a Sensor Network.

8
Introduction cont..

• Network Topology of a Wireless Sensor Network.

9
(cont..)
The Impact Of Data Aggregation On Wireless
Sensor Networks

• Data Aggregation in WSN ?
• - Address-centric approach
• - Data-centric approach

10
Routing Models
The Impact Of Data Aggregation On Wireless
Sensor Networks

• Address Centric Approach

11
The Impact Of Data Aggregation On Wireless
Sensor Networks

• Data Centric Approach

12
Data Aggregation
The Impact Of Data Aggregation On
Wireless Sensor Networks

• Result 1
• - The optimum number of transmissions required
  per datum for the DC protocol is equal to the
  number of edges in the minimum steiner tree in
  the network which contains the node set (s1, . ,
  Sk, D).
• - Hence, assuming an arbitrary placement of
  sources and a general network graph G, the task
  of doing DC routing with optimal data aggregation
  is NP-Hard.
• - Steiner Tree?
• - NP-Hard Problem?

13
Optimal Data Aggregation

• The optimal data aggregation problem is NP-Hard.
• An optimal multicast problem
• A well-known problem
• A minimum Steiner tree problem NPC
• SoNO optimal Solution
• Thus, sub-optimal solutions.

14
Data Aggregation
The Impact Of Data Aggregation On Wireless
Sensor Networks

• Section 1
• 3 suboptimal Schemes
• Center at Nearest Source
• Aggregation center nearest node to the sink.
• Shortest Paths Tree
• Shortest path routing with data aggregation in
  the overlap nodes.
• Greedy Incremental Tree
• Node closest to the tree connects to the path and
  forms a new tree until all the source nodes are
  vertices.

15
(cont..)
The Impact Of Data Aggregation On
Wireless Sensor Networks

• Section 2
• Sensor Network Models- for source placement.
• Factors affecting the performance gains of
  sensor network..
• Position of the sources
• communication network topology.
• Event Radius Model.
• Random Sources Model.

16
The Impact Of Data Aggregation On
Wireless Sensor Networks

• Event Radius Model.
• Location of an event.
• Sensing Range, S.
• (Pi)S2n average number of sources.

17
The Impact Of Data Aggregation On Wireless
Sensor Networks

• Random Sources Model.
• Sources not clustered.
• K random nodes, that are not sinks,are chosen to
  be sources

18
Energy Savings due to data aggregation

• Notations
• di the distance of the shortest path from
  source i to
• the sink
• NA the total number of transmissions required
  for the optimal
• address-centric protocol
• ND the total number of transmissions required
  for the optimal
• data-centric protocol
• X the diameter of the graph formed by a set of
  connected nodes
• K the number of the sources in the RS model
• R communication range
• S sensing range in the ER model

19
Energy Savings Due to Data Aggregation
The Impact Of Data Aggregation On
Wireless Sensor Networks

• Main performance gain ? When sources are far away
  from the sink.
• NA d1 d2 . Dk sum (di)
• Diameter X max of pairwise shortest paths.
• Theoretical Results
• Result 2
• If the source nodes S1, S2, , Sk have a
  diameter X gt 1. The total number of
  transmissions (Nd) required for the optimal DC
  protocol satisfies the following bounds
• ND lt (k-1)X min(di) X gt 1
• ND gt min(di) (k-1) X 1
• Corollary If diameter X lt min(di), then ND lt NA.

20

• Proof
• data aggregation tree consists of
• (k - 1) sources sending their packets to the
  remaining
• source which is nearest to the sink.
• This tree has no more than (k-1)X min(di)
  edges,
• Next result is obtained by considering the
  smallest
• possible Steiner tree which would happen if the
• diameter were 1.
• The shortest path from the source node at
  min(di) must be part of
• the minimum Steiner tree, and there is exactly
• one edge from each of the other source nodes to
• this node.
• Conclusion The optimum data-centric protocol
  will perform strictly
• better than the Address-centric
  protocol.

21
Cont

• Result 3

ND/NA 1/k
- DC Protocol gives k-fold savings.
22
Cont
The Impact Of Data Aggregation
On Wireless Sensor Networks

• Result 4
• If the subgraph G of the communication graph G
  induced by the set of source nodes (S1Sk) is
  connected, the optimal data aggregation tree can
  be formed in polynomial time.
• Corollary
• In the ER model, when R gt 2S, the optimal data
  aggregation tree can be formed in polynomial
  time.

23

• Proof
• The tree is initialized with the path from the
  sink to the nearest source.
• At each additional step of the GIT, the next
  source to be connected
• to the tree is always exactly one step away
  (such a source is guaranteed to exist since G is
  connected).
• At the end of the construction, the number of
  edges in the tree is therefore
• dmin (k - 1).
• Therefore, the GIT construction runs in
  polynomial time w.r.t. the number of
• nodes .

24
Summary

• Result 1
• The number of transmissions for the DC protocol
  number of edges in the minimum Steiner tree.
• Result 2
• Nd lt (k-1)X min(di)
• Nd gt (k-1) min(di)
• Result 3
• Result 4
• The optimal data aggregation tree can be formed
  in polynomial time.

ND/NA 1/k
25
Simulation Results
The Impact Of Data Aggregation On Wireless
Sensor Networks

• Figure 1
• - Comparison of Energy costs versus R in the
  ER model.
• Figure 2
• - Comparison of energy costs versus R in the RS
  model

26
The Impact Of Data Aggregation On Wireless
Sensor Networks

• Figure 3
• Comparison of energy costs versus S in the ER
  model
• Figure 4
• - Comparison of energy costs versus k in the
  RS model.

Sensing Range
27
Energy Savings.
The Impact Of Data Aggregation On Wireless
Sensor Networks

• Summary of experiments
• Energy Savings due to data aggregation can be
  quite significant, particularly when there are a
  lot of sources (large S or large k) that are
  many hops from the sink - (small R).

28
Delay due to Data Aggregation
The Impact Of Data Aggregation On Wireless
Sensor Networks

• Tradeoff
• Greater Delay !!
• Data from sources have to be held back at an
  intermediate node in order to be aggregated.
• Worst Case- Latency due to aggregation will be
  proportional to the number of hops between sink
  and the farthest source.

29
The Impact Of Data Aggregation On Wireless
Sensor Networks

• Figure 5
• Max(di) and Min(di) versus R in the ER Model
• Figure 6
• Max(di) and Min(di) versus S in the ER Model.

30
Conclusions
The Impact Of Data Aggregation On Wireless
Sensor Networks

• The formation of an optimal data aggregation tree
  is NP Hard.
• Energy Gains possible with data aggregation.
• Large when
• - number of sources large
• - Sources located close to each. Other
  and far from sink
• Aggregation Latency (Delay) non-negligible

31
Thank You.
32
The ACQUIRE Mechanism for Efficient Querying in
Sensor Networks

• Written By
• Narayanan Sadagopan
• Bhaskar Krishnamachari
• Ahmed Helmy
• Presented By
• Rishi Kant Sharda
• Kinnary Jangla

33
The Basics

• A sensor network is a computer network of many,
  spatially distributed devices using sensors to
  monitor conditions at different locations, such
  as temperature, sound, vibration, pressure,
  motion or pollutants.
• Each device is equipped with a radio transceiver,
  a small microcontroller, and an energy source,
  usually a battery. The devices use each other to
  transport data to a monitoring computer.
• Usually these devices are small and inexpensive,
  so that they can be produced and deployed in
  large numbers, and so their resources in terms of
  energy, memory, computational speed and bandwidth
  are severely constrained.
• Therefore not feasible to collect all
  measurements from each device for centralized
  processing.

34
Introduction

• Best to view them as distributed databases.
• Central querier/data sink issues queries.
• Due to energy constraints it is desirable for
  much of the data processing to be done
  in-network.
• This leads to the concept of data centric
  information routing i.e. queries and responses
  are for named data.

35
Categories of Queries

• Continuous Queries
• e.g Report the measured temperature for the next
  7 days with a frequency of 1 measurement per
  hour.
• One-Shot Queries
• e.g Is the current temperature higher than 70?
• Aggregate Queries
• e.g Report the calculated average temperature of
  all nodes in region X.
• Non-Aggregate Queries
• e.g What is the temperature measured by node x?
• Complex Queries
• e.g What are the values of the following
  variables X, Y , Z?
• Simple Queries
• e.g What is the value of the variable X?
• Queries for Replicated data
• e.g Has a target been observed anywhere in the
  area?
• Queries for Unique data

36
Flooding-based query mechanisms (Directed
Diffusion data-centric routing scheme)
37
Expanding Ring Search
38
Why ACQUIRE?

• Earlier Flooding-based query methods such as
  Directed Diffusion data-centric routing scheme
  are well suited only for continuous-aggregate
  queries.
• One-size-fits-all approach unlikely to provide
  efficient solutions for other types.
• If it is not continuous then flooding can
  dominate the costs associated with querying.
• Similarly in data aggregation duplicate responses
  can lead to suboptimal data collection in terms
  of energy costs.

39
Example Bird Habitat Monitoring
40
Example Continued

• Task Obtain sample calls for the following
  birds in the reserve Blue jay, Nightingale,
  Cardinal, Warbler
• Complex
• One-shot
• For replicated data

41
ACQUIRE
LEGEND Active Query Complete Response Update
Messages Sensor
42
Analysis of ACQUIRE

• Basic Model and Notation
• Local update
• Forward
• Steps to Query Completion
• Local Update Cost
• Total Energy Cost
• Optimal Look Ahead

43
Basic Model and Notation

• X number of sensors.
• V V1,V2,VN are the N variables tracked.
• Q Q1,Q2,QM consisting of M sub-queries, 1 lt
  M N and for all i i lt M, Qi ? V.
• Let SM be the average number of steps taken to
  resolve a query consisting of M sub-queries.
• d Look ahead parameter
• Size of a sensors neighborhood f(d)
• Assumed that all queries Q are resolvable by this
  network.
• x be the querier which issues the query Q.

44
ACQUIRE Process

• Local Update
• If current information not up-to-date, x sends
  request to all sensors d hops away.
• Request forwarded hop-by-hop.
• Sensors who get the request then forward their
  information to x.
• Let the energy consumed in this phase be Eupdate
• Forward
• After answering the query based on information
  received.
• x forwards the remaining query to a randomly
  chosen node d hops away.

45
ACQUIRE Process 2

• Since updates are triggered only when the
  information is not fresh, it makes sense to try
  and quantify how often such updates will be
  triggered.
• We model this as amortization factor c.
• An update is likely to occur at any given node
  only once every c queries.
• c such that 0 lt c 1. e.g if on average an
  update has to be done once every 100 queries, c
  0.01.
• a denotes the expected number of hops from the
  node where the query is completely resolved to x

46
ACQUIRE Process 3

• The average energy consumed to answer the query
  of size M with look-ahead d can be expressed as
• Case dD , where D is the diameter of the
  network.
• Case d too small.
• SM ? when d ?
• Eupdate ? when d ?

47
Steps to Query Completion

• If there are M queries to be resolved the
  probability of success in each trial is p M/N
  and failure is p (N-M)/N.
• Expected number of trials till 1st success
  1/pN/M.
• The whole experiment can be repeated with one
  less query and time to answer another query is
  N/(M-1) and so on.
• Let sM be the number of trials till M successes
  i.e complete resolution. Then

48
Steps to Query Completion 2

• H(M) is the sum of the first M terms of the
  harmonic series.
• H(M) ln(M) ?, where ? 0.57721 Eulers
  constant, thus
• and

49
Local Update Cost

• Eupdate Energy spent in updating the
  information at each active node.
• The number of transmissions needed to forward
  this request is the no. of nodes within d-1 hops,
  f(d-1).
• N(i) Number of nodes at hop i.

50
Total Energy Cost

• If the response is returned along the reverse
  path i.e a lt dSM
• Special case d 0 Random Walk.
• E(sM) steps to resolve and return the query.

51
Optimal Look-ahead

• Ignoring boundary effects, it can be shown that
  N(i) 4i and
• f(d) (2d(d1))1 for a grid of sensors, each
  node having 4 immediate neighbors.
• Combining expression for SM, Eupdate, Eavg , N(i)
  and f(d) we get

52
Optimal Look-ahead 2

• We determine the value of the look-ahead
  parameter which minimizes this energy cost by
  taking the derivative with respect to d and set
  it equal to 0, we get d by
• In general the lower c is, higher will be the
  look ahead parameter d

53
Optimal Look-ahead 4
54
Optimal Look-ahead 5
55
COMPARISON
56
Conclusions

• Proposed ACQUIRE as a scalable protocol for
  complex, one-shot queries for replicated data in
  sensor networks.
• Developed an analytical comparison of ACQUIRE,
  FBQ and ERS.
• With optimal parameter settings ACQUIRE
  outperforms all other schemes for complex,
  one-shot queries.
• Optimal ACQUIRE performs many orders of magnitude
  better than flooding-based schemes.
• Can reduce energy consumption by more than 60.

57
Future Work

• The efficiency of ACQUIRE can also be improved if
  the neighborhoods of the successive active nodes
  in the query trajectory have minimal overlap.
• Guided trajectories may also be helpful in
  dealing with non-uniform data distributions
• Taking into account that receptions can also
  influence energy consumption. This is the case
  especially for broadcast messages.

58
THANK YOU