Upload
clifford-anthony
View
228
Download
0
Tags:
Embed Size (px)
Citation preview
1
A Context Discovery Middleware for Context-Aware Applications with
Heterogeneous Sensors
Yu-Min Tseng
2
Sensor & Sensor Network
Deployed freely in the environmentContext information may be sensed & extracted by various sensors - Temperature, pressure, location, audio, image…Heterogeneous (Different sensing & computing capabilities)
Sensor mote are named via high-level descriptionsSelf-organization, power efficiency, large scale, reliabilityNeed ad hoc interaction for energy conservation & operation efficiency
3
The fundamental issue - How to deploy a sensor network that provide ad hoc comm. between sensors - An atomic ad hoc talk is end-to-end communicationBriefly - A robust & efficient ad hoc data-centric routing infrastructure
Problem Statement
4
Assumption
Each sensor is capable ofWireless comm.
•Only can talk to its geographical neighborComputationBuffering
Each sensor does not need GPS, MAC, or IP address supportTransmit requests & responses of sensory data only between those interested partiesOther nodes shouldn’t be involved, except those nodes that must relay messages
5
Support data-centric routing, request can be directed easily & efficiently. This reduces the overhead of address mapping & directory lookup.Self-calibration, self-management, self-healing.
Assumption (cont’d)
6
Data-centric Routing
Concept
Q: Prof. King’s cellular phone?
•A friend is contacted. Who ever taught by Prof. King
•The friend may ask his graduate classmate who joined the lab hosted by Prof. King
•The graduate student may ask the lab assistant
7
Data-centric Routing (cont’d)
Numerical perspectives
Translate each query to a number
Search a given number by numerically approaching
Example
•Q: Search a given number N1=Prof. King’s cellular phone
•N2=the friend
•N3=the graduate student
•N4=the assistant
•N2 > N3 > N4 > … Nm = N1
8
Data-centric naming
A sensor is characterized by its meta descriptor about its sensory data
The meta descriptor is attribute-value pairs
Example
Data-centric Routing (cont’d)
9
Data-centric Routing (cont’d)
Convert each sensor’s meta descriptor to a value by adopting a hash function (ex. SHA-1)
Requirement for the hashing
Generate a unique hashing value
•For naming various sensor
Uniform distribution
•Prevent overloading some particular sensors
10
v0 v1v2v3
Reference
R/2
R/4
R/8
u1u2u3 v4
R/4
R/8
Leader
Data-centric Routing (cont’d)
Example:
Vo receive a request for V4
Vo V1
V1 V4
11
Operations - Route
12
Operations - Join
13
Operations - Update
Cost(x, y) denotes the total energy consumed by routing arequest torwards y from xThe energy of signal consumed between 2 sensor is 1/d4, where d is the distance between the antennas
14
Operations - Construct
15
When to trigger a mote to update its discovery & neighbor tables?
1. A mote detects that leader stop forwarding requests
• Failure of link or motes
2. A node cannot communicate with its neighbor nodes
3. Periodical update
Maintenance
16
Evaluation
1 ~ 5000 sensors
Uniformly & randomly deploy over 210x210 meters square
Each sensor is randomly named
17
Result
18
Result (cont’d)
19
Result (cont’d)
20
Optimization
The logical network may not follow the actual network topologyExample
A given named value u not appear in mote vIf the cost of path from v to u is relatively economic than the one from v to a leader representing mote v’s i-th discovery scopeMote u replace the leader
21
Optimization (cont’d)
Each mote v locally eliminates loops
Ex: a b c a d c e d f a d c e d f a d f
22
It is possible that multiple leaders & neighbors appeared in he discovery & neighbor tables in a mote are the same
The path to the same leaders & the neighbors may not have same route
Replace those with the one have the minimally routing cost
Optimization (cont’d)
23
Summary
Heterogeneity Sensor motes are named via high-level descriptions
Data-centric
Discovery is based on named dataNamed data resolution and routing are integrated
Robustness
Each mote (a)periodically refreshes its discovery and/or neighbor tablesMultiple routes are constructed by having multiple leaders for a given particular discovery scopeMultiple neighbors in a mote are also maintained
Large-scale
Each mote only maintain O(log n) leadersThe discovery overhead is O(log n) in terms of messages and energy consumptionVicinity motes help the discovery
Maintenance-free
See the operation algorithm (Join, Construct, Update)Each mote refresh its discovery and neighbor tables
Energy-constrained
Via the vicinity discovery