Download Sensor Networks

Sensor Networks
Purnima Chandrasekaran
April 27, 2000
Why Sensor Nets?
• Civilian:
– Embed numerous distributed devices to monitor and interact with physical
world: in work-spaces, hospitals, homes, vehicles, etc.
– Inventory monitoring in factories.
– Map disaster stricken areas and coordinate recovery plans.
– Monitoring traffic conditions.
• Military:
– Stealthy monitoring of hostile / inhospitable terrain.
– Detect presence of biological agents in a battlefield.
– Perimeter surveillance
4/27/00
SensorNets
2
Overview
• Sensors :
– small wireless low-power unattended sensing devices with limited
communication & computation capabilities.
• SensorNets:
– Large scale dynamically changing robust sensor colonies.
– Coordinate amongst themselves to establish a communication network
and achieve a larger sensing task.
• Eg: Multiple sensors work together to find the speed, direction, size,
etc. of an approaching vehicle.
– As sensors are distributed, sensor network applications are designed
using distributed algorithms.
4/27/00
SensorNets
3
Constraints
• Energy
• Fault-tolerant
– Very limited (May not be
recharged :-)
– Avoid communication over long
distances.
• Self-configuring
• Dynamic Environments
– Sheer-number of sensors make
individual configuration
impossible.
– Sensors must reorganize when
additional sensors are added or
when old sensors fail.
4/27/00
– Sensors fail due to inhospitable
environments.
– SensorNets should be tolerant
of node failures.
– Dynamic operating conditions
– Dynamic resource availability
– Dynamic tasks
• Unattended and untethered
operation.
SensorNets
4
Comparison with Mobile Networks
SensorNets
Mobile Networks
 Nodes are unattended.
 Mobile nodes are mostly attended.
 Cheap sensors require
inexpensive power sources.
 Sensor life ends when power
runs out.
 Large numbers of sensors
coordinate to perform a task.
 Configuration of individual
sensors is infeasible.
 Designed for a specific
application.
 Communication is data-centric.
 Energy must be conserved, but nodes
can be recharged.
4/27/00
 Mobile nodes independently offer
service to the user.
 Mobile node can be configured by
the user.
 Designed to offer a wide-variety of
services.
 Communication is addressed to node.
SensorNets
5
Aspects of SensorNet Design
• Addressing
– Global Vs Local
• Naming and Binding
• Clustering
– To efficiently coordinate local interactions to achieve global
goals
• Information Dissemination
– Directed Diffusion
– Negotiation based
4/27/00
SensorNets
6
Address-Free Architecture
•
Applications do not address a particular sensor:
– What is the temperature at sensor #89768?
– Where are nodes whose temperature recently exceeded 30º?
•
•
•
•
Nodes and data are described by attributes rather than addresses.
Addressing, routing, and naming are removed from network layer to
application layer.
Addresses in Internet serve two purposes: Routing and Identity
In SensorNets
– nodes are not individually identified/addressed
– Message routing is Application specific
– Addresses do not serve the same purpose as traditional networks.
4/27/00
SensorNets
7
Address-Free Architecture (contd.)
•
Global addresses require long addresses to address individual nodes
– Amount of data communicated is small, but addressing overhead is high
– Power limitations impose local communication (but, sensing task is distributed)
– Locally unique identifiers require far fewer bits => Address Scalability
•
•
•
•
Locally unique identifiers - ephemeral, randomly selected, probabilistically
unique identifier
Conflicts are resolved by generating a new identifier
Detection of identifier collision is application-specific & is achieved by
detecting inconsistencies in data (eg. Missing sequence numbers)
Identifiers provide continuity among packets that make up a logical transaction
4/27/00
SensorNets
8
Clustering
• Clustering allows local coordination to achieve global goals.
– Scales well as the number of sensors increases
– Improved Robustness
– Efficient resource utilization
• Clustering builds hierarchical parent-child relationships
– Cluster-heads (parents) summarize the sensing information of children
– To conserve power, they may be the only ones performing some tasks
• Clustering Algorithm:
– Initially, all sensors are at level 0.
– Run a link level procedure to minimize transmission power, that still
provides full network connectivity.
– Each sensor follows the following algorithm:
4/27/00
SensorNets
9
Clustering Algorithm (contd.)
• Send an advertisement with a hop count ‘h’.
– Advertisements include hierarchical level, parent ID, remaining power
• Wait for time ‘t’ proportional to ‘h’
• Receive other sensor advertisements during ‘t’
• After ‘t’ , start a promotion timer ‘p’
– ‘p’ is inversely proportional to power left and advertisements received
• At the end of ‘p’, sensor promotes itself to level 1.
• Advertise itself as parent to level 0 sensors.
– includes a list of sensors it heard from; only they accept this as parent
– Symmetric connectivity between parent and child
• This process repeats at level 1, 2,... => setting up the hierarchy
4/27/00
SensorNets
10
Example: Triangulation
• Cluster heads independently run
the following algorithm:
– Each sensor determines its location
and the general direction of target.
– If all the neighboring cluster-heads
fall on the same side w.r.t. the target,
the cluster-head participates in
triangulation.
– A and B participate, while X
excludes itself.
4/27/00
SensorNets
11
Example 2: Adaptive Fidelity
• To get a better picture, turn on more sensors.
– Nodes adjust their coverage, sampling rate, communication
frequency based on neighbor density, power levels, and
reports from direct neighbors, etc.
4/27/00
SensorNets
12
Negotiation Based Protocols for Information
Dissemination
•
Sensor Protocols for Information via
Negotiation (SPIN)
– used to disseminate information in a
sensor network
•
•
•
•
•
Data is named & negotiated using
•
high-level descriptors - meta data
Data is also routed in application
controlled & application specific ways
based on data layout and state of
•
resources at sensors
Format of meta-data is application
specific
4/27/00
SensorNets
If x is the meta-data for data X, then
sizeof(x) < sizeof(X)
If data are distinguishable => metadata should also be distinguishable
Overcome implosion, overlap and
resource blindness deficiencies of
classic flooding
– By negotiation & resource-adaptation
Two classes of SPIN protocols– SPIN-PP for point-to-point networks
– SPIN-BC for broadcast networks
13
SPIN (contd.)
• Messages for communication in SPIN:
– ADV new data advertisement
– REQ request for data
– DATA data message
• ADV & REQ contain only meta-data
4/27/00
SensorNets
14
SPIN - PP
•
•
•
•
•
•
4/27/00
SensorNets
SPIN-PP is a simple protocol
Nodes know only about 1-hop
neighbors
Doesn’t involve much computation
to make decisions
Changes in network topology have
to travel only 1-hop
Can run in an unconfigured network
with a small startup to determine 1hop neighbors
SPIN-EC is SPIN-PP optimized for
energy conservation
15
SPIN - BC
•
•
•
•
•
•
4/27/00
Improvement over SPIN-PP as it uses
1-to-many communication
Effective resource conservation by
broadcast-message suppression
Nodes set a timer on receiving an ADV
and send a REQ only if no other REQ is
overheard for the same data.
SPIN-RL reliable version of SPIN-BC
Each node keeps track of every
advertisement & re-requests if data is
not received within some period of time
Re-requests a randomly picked
destination from the list of neighbors
that advertised the piece of data
SensorNets
16
Conclusions
• Sensor Networks is a very recent technology with lots of
research being carried out.
• Several approaches to solving the various issues have been
proposed and simulations done to study their effectiveness.
• Every approach has many advantages and disadvantages and
may/ many not be suitable for a particular implementation of
Sensor network.
4/27/00
SensorNets
17
References
•
•
•
•
Jeremy Elson, Deborah Estrin, “An Address-Free Architecture for Dynamic
Sensor Networks,” Under submission to SIGCOMM 2000.
Joanna Kulik, W Heinzelman, Hari Balakrishnan, “Negotiation-based Protocols
for Disseminating Information in Wireless Sensor Networks,” ACM
Mobicom’99.
Deborah Estrin, Ramesh Govindan, John Heidemann, Satish Kumar,“Next
Century Challenges: Scalable Coordination in Sensor Networks,” ACM
Mobicom’99.
William Adjie-Winoto, Elliot Schwatz, Hari Balakrishnan, “The design and
implementation of an intentional naming system,” ACM SOSP ‘99.
4/27/00
SensorNets
18