4. Motivation
Properties of Sensor Networks
Data centric approach: communication based named data,
not named nodes
No central authority
Resource constrained like limited power, computation
capacities and memory
Nodes are tied to physical locations
Nodes may not know the topology due to rapidly changes
of topology
Nodes are generally stationary
Q: How can we get data from the sensors?
5. Introduction(1/2)
A sensor network is composed of a large number of
sensor nodes, which are densely deployed either inside
the phenomenon or very close to it.
Random deployment
Cooperative capabilities
Sensor nodes scattered in a sensor field
Multi-hop communication is expected
Motivating factors for emergence
Applications
Advances in wireless technology
6. Introduction(2/2)
A region requires event-
monitoring
Deploy sensors forming a
distributed network
Wireless networking
Energy-limited nodes
On event, sensed and/or
processed information
delivered to the inquiring
destination
7. The Problem
Where should the data be
stored?
How should queries be
routed to the stored data?
How should queries for
sensor networks be
expressed?
Where and how should
aggregation be performed?
EventEvent
Sources
Sink Node
Directed
Diffusion
A sensor field
8. Directed Diffusion
Designed for robustness, scaling and energy efficiency
Data centric
Sinks place requests as interests for named data
Sources satisfying the interest can be found
Intermediate nodes can cache or transform data directly
toward sinks
Attribute-naming based
Data aggregation
Interest, data aggregation and data propagation are
determined by localized interactions.
9. Directed Diffusion
Four main features: Interests, Data, Gradients &
Reinforcement
Interest: a query or an interrogation which specifies what
a user wants.
Data: collected or processed information
Gradient: direction state created in each node that
receives interest.
Gradient direction is toward the neighboring node which the
interest is received
Events start flowing from originators of interests along
multiple gradient paths.
11. Directed Diffusion
Naming
Task descriptions are named by a list of attribute value pairs that
describe a task
eg:
type=wheeled vehicle // detect vehicle location
interval=20ms // send events every 20 ms
duration=10s // for the next 10s
rect=[-100,100,200,400]// from sensors within rectangle
Interests and Gradients
Interest is usually injected to the network from sink
For each active task, sink periodically broadcasts an interest
message to each of its neighbors
Initial interest contains the specified rect and duration attributes
but larger interval attribute
Interests tries to determine if there are any sensor nodes that
detect the wheeled vehicle(exploratory).
12. Interests
Interests: a query which specifies what a user wants by
naming the data.
Sink periodically broadcasts interest messages to each
neighbor.
Includes the rectangle and duration attributes from the
request.
Includes a larger interval attribute
All nodes maintain an interest cache
14. Sensor Node
Receives interest packet
Node is within the rectangle coordinates
Task the sensor system to generate samples at the
highest rate of all the gradients.
Data is sent using unicast
16. Exploratory versus Data
Exploratory use lower data rates
Once the sensor is able to report the data a reinforcement
path is created
Data gradients used to report high quality/high
bandwidth data.
17. Positive Reinforcement
Sink re-sends original interest message with smaller
interval
Neighbor nodes see the high bandwidth request and
reinforce at least one neighbor using its data cache
This process selects an empirically low-delay path.
18. Multiple Sources & Sinks
The current rules work for multiple sources and sinks
19. Negative Reinforcement
Repair can result in more than one path being reinforced
Time out gradients
Send negative reinforcement message
20. Repair
C detects degradation
Notices rate of data significantly lower
Gets data from another neighbor that it hasn’t seen
To avoid downstream nodes from repairing their paths C
must keep sending interpolated location estimates.
22. Simulation Environment
NS2, 50 nodes in 160x160 sqm., range 40m
Random 5 sources in 70x70, random 5 sinks
Average node density constant in all simulations
Comparison against flooding and omniscient multicast
1.6Mbps 802.11 MAC
Not realistic (reliable transmission, RTS/CTS, high power, idle
power ~ receive power)
Set idle power to 10% of receive power, 5% of transmit
power
23. Metrics
Average dissipated energy
per node energy dissipation / # events seen by sinks
Average packet delay
latency of event transmission to reception at sink
Distinct event delivery
# of distinct events received / # of events originally sent
Both measured as functions of network size
24. Average Dissipated Energy
In-network aggregation reduces DD redundancy
Flooding poor because of multiple paths from source to sink
flooding
DiffusionMulticast
Flooding
DiffusionMulticast
25. Delay
DD finds least delay paths, as OM – encouraging
Flooding incurs latency due to high MAC contention,
collision
flooding
Diffusion
Multicast
26. Average energy and delay
Average delay is misleading
Directed Diffusion is better than Omniscient Multicast!?
Omniscient multicast sends duplicate messages over the
same paths
Topology has little path diversity
Why not suppress messages with Omniscient Multicast
just as in Directed Diffusion?
27. Event Delivery Ratio under node failures
Delivery ratio degrades with higher % node failures
Graceful degradation indicates efficient negative
reinforcement
0 %
10%
20%
28. Analysis
Energy gains are dependent on 802.11 energy assumptions
Directed Diffusion has lowest average dissipated energy
Data aggregation and negative reinforcement enhance
performance considerably
Differences in power consumption disappear if idle–
time power consumption is high
Can the network always deliver at the interest’s requested
rate?
Can diffusion handle overloads?
Does reinforcement actually work?
29. Continued….
Pros
Energy – Much less traffic than flooding.
Latency – Transmits data along the best path
Scalability – Local interactions only
Robust – Retransmissions of interests
Cons
The set up phase of the gradients is expensive
Need of and adequate MAC layer to support an efficient
implementation. The simulation analysis uses a modified
802.11 MAC protocol
Design doesn’t deal with congestion or loss
Periodic broadcasts of interest reduces network lifetime
Nodes within range of human operator may die quickly
30. Conclusions
Directed diffusion, a paradigm proposed for event
monitoring sensor networks
Energy efficiency achievable
Diffusion mechanism resilient to fault tolerance
Conservative negative reinforcements proves useful
More thorough performance evaluation is required
MAC for sensor networks needs to be designed
carefully for further performance gains