Title: Directed Diffusion: A Scalable and Robust Communication Paradigm for Sensor Networks Charlmek Intanagonwiwat Ramesh Govindan Deborah Estrin
1Directed Diffusion A Scalable and Robust
Communication Paradigm for Sensor
NetworksCharlmek Intanagonwiwat Ramesh
Govindan Deborah Estrin
- Presentation By Hardik Shah
2.
Goal Embed computing (computing
device/sensors) close enough to the environment
to sense (detect) the phenomena, monitor and take
actions accordingly .
Key Issue Embedding the sensors in the
physical world requires network of such nodes
to co-ordinate to perform distributed sensing of
environmental phenomena.
3 Motivation
- Energy Efficiency
- Infeasible to transmit time-series data even
hop-by-hop. - Perform local computation and reduce data before
transmission. - Scalability
- Requires thousands of sensors to coordinate to
reach the decision. - Decisions should be done as much local as
possible. - Robustness
- Handle changing environment situations
4Architectural Requirements
- Application aware communication premetives
(expressed in terms of named data not in terms of
node who request data) - Achieve locality for decision making.
- (and reduce the communication)
- Application centric, data-driven networks.
- Achieve desired global behavior through localized
interactions, without global state.
5Directed Diffusion
- Data dissemination paradigm for distributed
network of sensors.
6Assumptions
- Sensor network's lower level communication is
topology independent.( not like IP networks mean
logical connectivity distinct from physical
geography). - Data aggregation is task dependant.( set of
tasks defined by application (or set of
applications) which defines interests for
network) - Naming scheme decides the expressiveness and
effectiveness of communication.
7Basic Directed Diffusion concepts
- Communication for named data not for those who
produces (its not our concern!) - Query generates (virtually from any node in the
network) interest (collection of attribute value
pair) - For specific data (which tries to map with
events supported by network ). - Interest diffused locally based on the naming
scheme (its most imp since communication done
for named data (hierarchical /flat)( mit ins
uses hierarchical approach).)
8- This sets the gradients (within network) to draw
events matching the interest. - Gradient represents both direction towards data
matching and status of demand with desired update
rate (active/inactive).
9Architectural elements
- Naming Scheme
- Interest propagation
- Data propagation
- Data caching and aggregation
- Reinforcement
10Naming
- Given Set of Tasks supported by sensor network
selecting a naming scheme is first step in
designing sensor networks. - Basically list of attribute value pairs.
- E.g. For tracking animal its attributes should
describe tasks like, type of animal, - geographic location to track, interval for
sending updates, duration for which it was
recorded (event occurrence time) -
11Interest propagation
- Flooding.
- Location aware routing (or geo casting).
- Directional propagation on previously cached
data. - In paper they have used flooding approach.
-
12Event
Source
interests
Sink
Have u seen any four leg animal???
QUERY DIFFUSED IN TO INTEREST WHICH IS LIST OF
ATTRIBUTE VALUE PAIRS
Interest Propagation (Flooding)
13YES I HAVE SEEN ONE.
INTIAL GRADIENTS SETUP(VALUEDIRECTION)
14Data Propagation
- Reinforcement to single path delivery.
- Multi path delivery with selective quality.
- Probabilistic forwarding with multi path
delivery. - For selecting neighbor who gave first or
either who has highest energy or lowest delay can
be chosen. (Its application dependant.)
15DATA DELIVERY THROUGH REINFORCED PATH
SINGLE PATH DELIVERY (CAN BE MULTIPATH ALSO)
16 Data caching and aggregation
- Robust data delivery in case of node failure.
- Validate with timestamps.
- May use hierarchical scheme with one or more
entry for distinct interest.
17IN CASE OF NODE FAILURE USE ALTERNATIVE PATHS
18Reinforcement
- When to reinforce ?(quality/delay matrices can be
chosen) - Whom to reinforce ?
- How many to reinforce?
- When to send negative reinforcement ?
19TinyOS Implementation
20(No Transcript)
21Summary of results
- Diffusion has achieved same delay of omniscient
multicast. - Application level data dissemination has
potential for energy saving. - This work did not develop the software
architecture necessary for realizing attributes
for in networking processing in an operational
system.
22 Comparison of Directed Diffusion to
flooding and omniscient multicast
23Work is influenced by
- Multicast routing join techniques for interest
propagation spt tree construction (or shared
tree) for deciding reinforcement policies. - Declarative routing is similar in approach except
no filters used. - Intentional naming system of mit has similar
concept for naming as directed diffusion(but
hierarchical not flat attribute value pair.) - In network processing for local repair is similar
to router assist for localized error recovery.
24 What it Proposes?
- A simple architecture that uses a topological-
independent naming for low-level communication
to achieve flexible, yet highly energy efficient
application designs.
25Discusses
- Design space of protocols underlying directed
diffusion. - (Where every sensor is task aware and possibly
knows where it is.)
26Evaluates
- Design questions concerning naming and
in-network processing encountered in deploying a
sensor network and presents experimental results.
27Issues of Concern
- Ad hoc, self organizing, adaptive systems with
predictable behavior - Collaborative processing, data fusion, multiple
sensory modalities - Data analysis/mining
28Issues yet to be resolved
- How to handle congested network?
- Semantics for gradients.
- (Variant of D.D. Is gradient directed
diffusion.) - Handling of more than one sources.
- Negative reinforcement increases delay and
contention (D.D. Uses mac layer unicast)
29Optimization
- Create processing points in the network.
- High level interests/queries for activity
triggers lower level local queries for particular
data modalities and signatures (e.g. acoustic and
vibration patterns that are mapped to the
activity of interest)As opposed to generating
detailed queries at sink points and relying on - opportunistic aggregation alone.
30Work In Progress
- Multi path reinforcing multiple upstream
neighbors for load balancing and robustness. - Disjoint paths selection.
- Opportunistic aggregation of source data
- Managing gradients/resources.
- Tiny diffusion for Motes.
- Diffusion under mobility objects, nodes
31Possible Areas of Future Work
- Adaptation to local node densities.
- How to map diffusions parameters to Diffusion
needs? - Diffusion to work on Asymmetric links.
- Intelligence in filters for decision making.
32Reference
- Design and implementation of INS.
- Location aware routing.
- Geocasting in mobile ad hoc networks
- Location based multicast algorithms
- Query localization techniques for on-demand
routing protocols in ad-hoc net. - Declarative routing.
33More Information
- SCADDS project
- http//www.isi.edu/scadds
- ns-2 network simulator
- http//www.isi.edu/nsnam/dist/ns-src-snapshot.tar.
gz - testbed and software
- http//www.isi.edu/scadds/testbeds.html