PROTOCOL STACK

Task Management Plane

Mobility Management Plane
Plane
Power Management
Application Layer

Transport Layer
Network Layer

Link Layer
Physical Layer

1
 WHY CAN’T AD-HOC NETWORK
PROTOCOLS BE USED HERE?

 Number of sensor nodes can be several orders of magnitude higher

 Sensor nodes are densely deployed and are prone to failures

 The topology of a sensor network may change frequently due to node mobility and
node failure

 Sensor nodes are limited in power, computational capacities, and memory

 May not have global ID like IP address

 Need tight integration with sensing tasks

2
PART IV.
APPLICATION LAYER

3
 APPLICATON LAYER

 Query Processing
 Sensor Network Management

4
Query Processing

Query User
Dissemination Query

User
Storage

5
Query Processing

User
Query

Charts,
graphs, User
etc…
Storage

6
 QUERY PROCESSING
 User sends a query and gets response from valid sensor
nodes, e.g.,
WHICH AREA HAS HUMIDITY HGHER THAN 50?

Type= humidity
Timestamp=01/18/2007/16:35:28
Location=[60N,120W]
Humidity>50
7
 QUERY PROCESSING

Example: (Animal Tracking Query)

Type = four legged animal (detect animal location)

Interval = 30 s (send back events every 30 s)
Duration = 1h (.. for the next hour)
Rec = [-100,100,200,400] (from sensors within the rectangle)

8
.
 QUERY PROCESSING

Sensor detecting the animal generates the following data:

Type – four legged animal (type of animal seen)

Instance= elephant (instance of this type)
Location = (125,220) (node location)
Intensity = 0.6 (signal amplitude measure)
Confidence = 0.85 (confidence in the match)
Timestamp= 01:20:40 (event generation time)

9
 Sensor Query and Tasking Language (SQTL)
(C-C Shen, et.al., “Sensor Information Networking
Architecture and Applications”, IEEE Personal
Communications Magazine, pp. 52-59, August 2001.)

 SQTL is a procedural scripting language.

 It provides interfaces
to access sensor hardware:
- getTemperature, turnOn
for location awareness:
- isNeighbor, getPosition
and for communication:
- tell, execute.
10
 Sensor Query and Tasking Language (SQTL)

By using the upon command, a programmer can create

an “event handling block” for 3 types of events:

1. Events generated when a message is received by a

sensor node (RECEIVE),
2. Events triggered periodically (EVERY),
3. Events caused by the expiration of a timer (EXPIRE).

11
 Main Approaches in Query Processing

Push-based:
– Sensor initiated information delivery
Pull-based:
– Sink initiated query dissemination
Push-pull:
– Both sensors/sink actively involved

12
 Main Approaches in Query Processing

 PULL BASED – (INTEREST DISSEMINATION)

* Users send their interest to a sensor node,
a subset of the nodes or the entire network.
* This interest may be about a certain attribute
of the sensor field or a triggering event.

 PUSH BASED – (ADVERTISEMENT OF AVAILABLE DATA)

* Sensor nodes advertise the available data to
the users and the users query the data which
they are interested in.

13
 Query Classification

Continuous Queries:
Collect data commonly span some long period of time
Snapshot Queries:
Collect data about now or some other point in time
Historical Queries:
Collect summary data about past

14
 Query Types

 Data-Centric Query: (ATTRIBUTE BASED QUERY)

Search for “fire detection”, instead of

individual sensor node data
 Geographical Query: (LOCATION BASED QUERY)

Report values at {x,y,z}

15
Data-Centric Query: (ATTRIBUTE BASED QUERY)
“The locations of the nodes that sense
temperature higher than 70F”

 Geographical Query: (LOCATION BASED QUERY)

“Temperatures read by the nodes in region A”

16
 Data-Centric Query:
(ATTRIBUTE BASED QUERY)
Query:
71
Sensor nodes that read >70oF
temperature 68
75
67
Sink 66
71
71

68 71

69

17
 Geographical Query:
(LOCATION BASED QUERY)
Query: Region A
Region A nodes should send their 71
temperatures 75
68 67
Sink 66
71
71

Region C 68 71

69
Region B
Important for broadcasting,
multicasting, geocasting and anycasting
18
 Data Aggregation/Fusion

Impossible to continuously collect raw sensor data

– Limited memory and bandwidth
– Information overload
Individual sensor readings are of limited use:
– Interest in collective information rather than
individual sensor data
– Exploit in-network processing
Goal:
Save energy and increase network lifetime by
combining several sensor data
19
 Data Aggregation/Fusion

Data Aggregation:
Process of combining data or
information to estimate or predict events

Idea:
Take advantage of the routing hierarchy and
high network density
20
 DATA AGGREGATION/FUSION

Aggregation Query Processor

User
Query

User

Charts,
Storage graphs,
etc…

21
 Data Aggregation
Source 2
Source 1 Source 2

Source 1

Source 2

Aggregate
data before
Source 1 & 2
routing

Sink
22
 Data Aggregation/Fusion
The sink asks the sensor
71
nodes to report certain
conditions. Data coming 68 75
67
from multiple sensor 66
71
nodes are aggregated.
71
Sink
68 71

69

Query:
Sensor nodes that read >70oF temperature
23
 Data Aggregation Components
DATA STORAGE
Store sensor data in a memory efficient way, while
preserving the accuracy of the information.

AGGREGATION FUNCTIONS
Place aggregation points on the paths from sensors to sink.

AGGREGATION PATHS
Which are the optimal aggregation points? Which is the most suitable
path from source to sink to favor data aggregation?

24
 Data Storage Representations
 Data can be represented with different degree of accuracy.

Store individual data  histograms and lists

1 2 … k
Data samples 3 4 2 1 0 4 7 2

Store only mean and variance or other statistical representations

mx
Data samples
σx
25
 Aggregation Functions

Very simple functions

– Average, max, min, median
– Suppression of duplicates
More sophisticated functions
– Exploit spatial and temporal correlation
– Signal processing (convolution, filtering, etc.)

26
 Aggregation Paths

Sink
 Challenges:
– Find the optimal number of
aggregation points
– Selection of aggregation points
– Dynamic change of aggregation
points (energy efficiency)

Aggregation point
Sensor node
27
 Aggregation challenges

 Inherently unreliable environment, certain information

unavailable or expensive to obtain
– How many nodes are present?
– How many nodes are supposed to respond?
– What is the error distribution (in particular, what
about malicious nodes?)
– Trying to build an infrastructure to remove all
uncertainty from the application may not be feasible –
do we want to build distributed transactions?
28
 Optimal Data Aggregation

 Optimal Data Aggregation is NP-Hard

 Sub-optimal Algorithms:
– Opportunistic
 Just aggregate when possible
– Center at Nearest Source (CNS)
 The nearest source acts as the aggregation point
– Shortest Paths Tree (SPT)
 Sources send using shortest path if able to aggregate
– Greedy Incremental Tree (GIT)
 Recursively select the closest source to the tree
– Clustered Diffusion with Dynamic Data Aggregation
 Hybrid between diffusion and clustering with the ability to aggregate data
at the cluster heads
29
 Advantages of Aggregation

– In network processing reduces traffic load by

aggregating data en route
 Energy and memory efficiency

– Scalable to large numbers of both sinks and

sensors

30
 Disadvantages of Aggregation

– Requires careful design to tradeoff accuracy and

storage and message size

– Incur information loss, making robust estimation more

difficult:
 E.g. a single outlier reading can damage MAX/MIN aggregates

31
 Query Processing Challenges

High density: Rich and massive data

Correlation: Spatially distributed and correlated data
Uncertainty: Noise, erroneous data, outliers.
Semantics: Individual data  useful information
 Tools for Querying Sensor Networks

 TinyDB
– Information storage framework
– Tree-based data collection
 COUGAR
– Individual sensor data
– Distributed Gathering
– SQL-like
 TAG (Tiny Aggregation)
– Focus on Aggregation using SQL-like query language
– Integrated in TinyOS
 TinyDB
S. R. Madden, et. al., ``TinyDB: An Acquisitional Query Processing System for
Sensor Networks,’’ ACM Transactions on Database Systems, pp. 122–173, March 2005.

 A distributed query processor for networks of Mica motes

Goal: Eliminate the need to write C code for most
TinyOS users
Features
– Declarative queries
– Temporal + spatial operations
– Multihop routing
– In-network storage
 COUGAR-SENSOR DATABASE SYSTEM
S. Madden and J. Gehrke, ``Query Processing in Sensor Networks,’’
IEEE Pervasive Computing, vol. 3, No. 1., March 2004.

Performs in-network computation

Decreases energy consumption
Attempts to merge similar queries
Propagates results to sink
 COUGAR Architecture

Sink
Sensors send their
information towards
designated leaders
Partial aggregation is
performed at sensors
Collected data
aggregated optimally at
Leader the leaders
Sensor node
 COUGAR Components Leader Node
Sensor Node Towards the sink
Towards the leader
Select
Partial Aggregated
Results
Aggregation
Aggregate
Operator
Partially
aggregated
Received Collected results
Data (radio) Data (sensor) Received
Data (radio)

Data from neighbor node

Data from neighbor nodes

 COUGAR Leader Selection
Methods
– Fixed
– Randomly
Leader selection policy
– Dynamically maintained in case of failure
– Minimize communication distance
 Tiny Aggregation (TAG): Distributed Aggregation Scheme
S. Madden et al., “TAG: a Tiny Aggregation Service for
Ad-Hoc Sensor Networks” , in Proc. of OSDI, 2002

* A simple declarative interface for data collection

* Aggregation while distributively executing aggregation
queries
* It relies on the routing tree built by broadcast messages
sent by the sink.
* Based on the routing tree, parents wait for their children
to send data and discard irrelevant data combining relevant
readings into more compact records based on the
aggregation principles specified by the queries.
 TAG Approach
 Query Distribution: Aggregate queries are pushed down the
network to construct a spanning tree from the sink
– Root broadcasts the query, each node hearing the query
broadcasts.
– Each node selects a parent.
– The routing structure is a spanning tree rooted at the
query node.

 Data Collection: Aggregate values are routed up the tree.

– Internal node aggregates the partial data received from its
sub-tree.
 Sensor Network Management

Real world deployments need administration tool

Key issues in network management
– Data overload: Data increases dramatically with high data
rates and number of sensors
– Health monitoring: Environmental conditions and
constraints (i.e. power) make accurately monitoring motes
difficult
– Information visualization: Many challenges in presentation
of large amounts of data
41
 Sensor Network Management

System Administrators interact with Sensors using Network Management Tool.

TASKS:
 Turning sensors on and off
 Moving the sensor nodes
 Querying the sensor network configuration and the status
of nodes and re-configuring the sensor network
 Authentication, key distribution and security in data communication
 Time-synchronization of the sensor nodes
 Exchanging data related to the location finding algorithms
 Introducing the rules related to data aggregation,
attribute-based naming and clustering to the sensor nodes

42
 Network Management: Design Factors

Lightweight operation
– Minimum overhead and energy consumption
– Should not interfere with communication and sensing
operations
Robustness and fault tolerance
– Resilient to network failures (dropped packets, node
failures, link failures)
– Adaptive to network dynamics (topology changes,
node joins, node removals)
43
 Network Management: Design Factors

Minimal data storage

– Respect the memory constraints of sensor nodes

Scalability
– Operate efficiently in large scale networks

44
 Network Management Types

Passive
– The system collects information about network states.
– Post-processing of the collected data (offline analysis)
Reactive (Event-triggered)
– Network monitoring is initiated if an event of interest
has occurred
– Reconfigure the network according to an event
– Real-time processing of the collected data (online
analysis)
45
 Network Management Types

 Proactive
– Actively collects network state information
and analyses network performance
– Real-time processing of the collected data
(online analysis)

46
 Network Management Architectures:
Centralized

Sink
– Sink acts as the central manager
station
– Reduces the processing burden on
resource-constrained nodes
– High communication cost

Network Manager
47
 Network Management Architectures:
Distributed

Sink – Multiple manager stations

– Each manager controls a
cluster
– Managers communicate
directly with other managers
– Lower communication costs
than centralized management
– Complex and computationally
Network Manager expensive for sensor nodes
48
 Network Management Architectures

 Hierarchical
Sink L3 – Multiple manager stations
– Each manager reports to a
L2 higher-level manager
L2 L1
– No direct communication between
same-level managers
L1
L1 – Hybrid of centralized and
distributed architectures

Network manager

49
 Sensor Network Management Protocols

MANNA SNMS
BOSS

50
 Sensor Network Management Protocol:
BOSS (Bridge Of the SensorS)
H. Song et al., “UPnP- Universal Plug and Play- Based Sensor
Network Management Architecture,” in Proc. ICMU, 2005.

Sink node is used as BOSS:

– Sink node provides a bridge between WSN and a UPnP NW
– Centralized network management

51
 BOSS (Bridge Of the SensorS)

 Advantages:
– Different applications can be managed simultaneously
– Adaptive to topology changes
– Support proactive network management

 Disadvantages:
– Requires an end-user to observe network states and
take management actions accordingly (not automated)

52
 MANNA: A Management Architecture
L.B. Ruiz et al., “MANNA: A Management Architecture for Wireless
Sensor Networks,” IEEE Communications Magazine, pp. 116–125, 2003.

A policy-based management system:

– Collects dynamic management information using MNMP protocol
– Maps the collected information into WSN maps
– WSN maps maintain information about the overall state of the
network

 Examples of WSN maps:

– Topology map depicting node connectivity
– Residual energy map showing battery level of nodes in the network
– Sensing coverage area map describing the area covered by sensor
elements
– Security map describing the security status of the sensor network
53
 MANNA Network Management Protocol (MNMP)

 Three management components

– Sensor nodes
 Organized into clusters
 Send their states to the cluster-head

– Cluster-heads
 Responsible for executing local management functions
 Aggregate management data received from sensor nodes
 Forward management data directly to the base station

– Network manager (@ sink)

 Powerful management entity
 Responsible for complex management tasks requiring global knowledge of the
network

54
 Advantages and Disadvantages of MANNA

Advantages:
– Adaptive to network dynamics (distributed
management)
– Provide overall network state via WSN maps

Disadvantages:
– Require high processing and memory at sensor nodes

55
 Sensor Network Management System (SNMS)
G. Tolle and D. Culler, “Design of an Application-Cooperative Management System
for Wireless Sensor Networks,” in Proc. EWSN, Feb. 2005.

 An interactive system for monitoring the health of WSN

 Provide two main management functions
– Query-based network health data collection
 Sink initiates monitoring
– Event-driven data logging
 Sensors report their data if they meet the pre-specified event thresholds
 Support two traffic patterns
– Data Collection (sensors to sink)
– Query Dissemination (sink to sensors)
56
 Sensor Network Management System (SNMS)

Reply  Use a data gathering tree to collect

Sink network health information from the
Query
sensors

 Every sensor only maintains the single

best parent based on the strongest
received signal strength
– Memory efficient (not require
neighborhood table)

 Adaptive to changing network

conditions since nodes will select a
new parent if their existing parent
dies
57
 Advantages and Disadvantages of SNMS

Advantages:
– No neighborhood table required - Minimal
overhead to memory and network traffic

Disadvantages:
– Centralized approach requires continuous
polling of network (not energy efficient)

58
 Network Management System Comparisons

Network Energy Memory

Management Functionalities Robustness Scalability
Efficiency Efficiency
System
Network state retrieval,
localization,
BOSS Yes Yes Yes NO
synchronization, and
power management

Network state retrieval,

MANNA sampling frequency Yes Yes NO Yes
control, fault detection

Query-based network
SNMS health data collection Yes Yes Yes NO
and event logging

59