A Local Data Abstraction and Communication Paradigm for ... - MPC

5 downloads 8441 Views 3MB Size Report
Mar 21, 2007 - Applying sensor networks to pervasive computing. ○. Unpredictability of ... on network path (to support different cost functions for the same metric). ○ .... Complete network performance evaluation that measures query ...
A Local Data Abstraction and Communication Paradigm for Pervasive Computing

Sanem Kabadayı and Christine Julien [email protected] March 21, 2007

Overview     

Motivation and Challenges Our Solution: Scenes Scene Model Evaluation Conclusion

2

Introduction 





Motivation: Potential to enhance human experience and simplify life  Aware homes, intelligent construction sites, firstresponder deployments, battlefield scenarios Challenge: Highly dynamic operating contexts and vast amounts of raw information, need to limit the scope of interactions Solution: Declarative specification of the region of the network with which to interact

3

Intelligent Construction Site

4

Challenges

remote distributed sensing

 

pervasive computing

Sensor networks have almost exclusively been used in the former manner Applying sensor networks to pervasive computing  

Locality of interactions Mobility-induced dynamics



Unpredictability of coordination 5

Our Solution: Scenes 

Pervasive computing environments present highly dynamic operating contexts and vast amounts of raw information 



Applications must be able to limit the scope of interactions

Scenes 



Novel communication abstraction that allows application to limit scope of interactions to include only data that matches current needs Results not specific to sensor networks

6

Scenes: Abstractions of Local Data 

Programmer specifies three parameters for a scene: 





Metric: property of network/environment defining connection cost Cost function: function that operates on network path (to support different cost functions for the same metric) Threshold: value path cost must satisfy for sensor to be member (to avoid flooding the entire network) For all j < k, cj THRESHOLD k

i ci = COST_FUNCTION(METRIC(pi, i)) 0

7

Persistent (Continuous) and One-Time Queries 

One-time queries 





Return a single result from each scene member One location reading for each crane in the scene on a construction site

Persistent queries 



Return periodic results from scene members Crane location readings every 30 seconds

8

Scene Construction

For all j < k, cj THRESHOLD k

i 0

ci = COST_FUNCTION(METRIC(pi, i))

9

An Example Scene 

Construct a scene: 



From sensors within 5m that can deliver data to host in less than 15 ms No messages are sent yet

Metric

Cost function

Metric value

Threshold

Distance component

SCENE_DISTANCE

SCENE_DFORMULA

Location of source

5m

Latency component

SCENE_LATENCY

SCENE_SUM

Latency on path so far

15ms

10

An Example Scene 

Volatile organic compound 





VOC cloud

Organic chemical compounds that can vaporize under normal conditions May have adverse health effects

Query for gas cloud concentrations Metric

Cost function

Metric value

Threshold

Distance component

SCENE_DISTANCE

SCENE_DFORMULA

Location of source

5m

Latency component

SCENE_LATENCY

SCENE_SUM

Latency on path so far

15ms

11

An Example Scene 

Query for gas cloud concentrations > 85 ppm (parts per million) every second 



VOC cloud

All the sensors in the scene receive the message Only scene members who can satisfy the query respond

Metric

Cost function

Metric value

Threshold

Distance component

SCENE_DISTANCE

SCENE_DFORMULA

Location of source

5m

Latency component

SCENE_LATENCY

SCENE_SUM

Latency on path so far

15ms

12

An Example Scene 

VOC cloud

Gas cloud moves, changing responses

Metric

Cost function

Metric value

Threshold

Distance component

SCENE_DISTANCE

SCENE_DFORMULA

Location of source

5m

Latency component

SCENE_LATENCY

SCENE_SUM

Latency on path so far

15ms

13

An Example Scene 

Client moves, changing scene membership 



VOC cloud

Maintained by beacons

Sensors could move as well

Metric

Cost function

Metric value

Threshold

Distance component

SCENE_DISTANCE

SCENE_DFORMULA

Location of source

5m

Latency component

SCENE_LATENCY

SCENE_SUM

Latency on path so far

15ms

14

An Example Scene 

Metric value changes (e.g., latency increases) changing scene membership 

VOC cloud

Maintained by beacons Latency increases on this link, disqualifying these nodes from the scene

Metric

Cost function

Metric value

Threshold

Distance component

SCENE_DISTANCE

SCENE_DFORMULA

Location of source

5m

Latency component

SCENE_LATENCY

SCENE_SUM

Latency on path so far

15ms

15

Scene Feasibility Study  

TOSSIM (Simulator for TinyOS) implementation of scene protocol Parameters   

100 sensor nodes in a 200ft X 200ft square Single client device moves among sensors Topologies  



Regular grid with 20ft internode spacing Uniform random placement

TOSSIM empirical radio model 

Based on real Mica measurements in outdoor environment

16

Scene Feasibility Study (cont.) 

Variables 



Client device speed: 0 mph, 2 mph, 4 mph (brisk walk), or 8 mph (slow moving vehicle) Scenes based on hop count metric (1, 2, 3 hops) 



Implementation easily supports other types of scenes

Beacon interval set inversely proportional to speed 



Empirically shown to provide stable scenes for varying mobility Requires use of global knowledge and is not how beacon intervals should be assigned

17

Example Scene 

 



Fixed radius (10 ft) radio model Scene threshold 3 hops Red LEDs turn on for scene members Blue circles represent nodes currently sending beacons

18

Performance Metrics for Scenes 

Average number of scene members  



Number of messages sent per scene member 





Measures performance of beacon assignment heuristic Number of scene members found should stay same as speed increases Measures sensor node's cost of participation in scene Correlated with battery dissipation for participating sensors

Number of messages sent per unit time  

Measure scene scalability (i.e., for scenes of increasing sizes and increasing client mobility)

Measure of average activity in network Activity takes place only within scene 19

Numbers of Scene Members  

Number of scene members independent of client speed Setting beacon interval directly proportional to client speed accurately keeps track of moving client

20

Messages Sent per Scene Member 

Beacon frequency set directly proportional to client speed 



For 4mph, beacon every 0.5s; for 8mph, beacon every 0.25s

Linearity due mainly to increased beacons and nothing else

21

Messages per Second of Scene Registration  

Beacons sent more frequently as speed increases; more messages packed into a time interval As size increases, more nodes become members and beacon

22

Conclusion 



Communication protocol that enables opportunistic communication with the local immersive sensor environment for pervasive computing applications Future work  Complete network performance evaluation that measures query response times and direct energy usage  Real-world deployment on mixture of embedded and client devices  Optimal beacon assignment

23

Questions?

Thank you! Sanem Kabadayı [email protected] http://mpc.ece.utexas.edu

24

Suggest Documents