Spatial Views: Space-aware Programming for Networks of Embedded Systems

Spatial Views: Space-aware Programming for Networks of Embedded

Systems

Yang Ni, Ulrich Kremer, and Liviu Iftode

Rutgers University

This work has been supported by NSF ITR Award ANI-0121416.

LCPC 2003 EEL Lab, Rutgers University 2

NES: A New Computing Platform


NES: A New Computing Platform


NES: A New-Generation Computing Platform

(contd.) Many potential applications for NES

What’s the traffic 10 miles ahead? Is there a man in red near Terminal A? Find a cab.

How to program them?Will existing models work?


What’s Special About NES?Location-sensitivity.

A node is interesting because of its location in addition to its functionality.Context-awareness often translates into location-awareness.

Volatility.Nodes join and leave at any time.

Resource-limitation.Execution time and battery life are limited.


How to program NES?Existing programming models do not sufficiently address all those issues specific to NES.We need a high-level programming model with:

Location-awarenessLocation specification at the programming levelLocation-based operation at the programming level

Dynamic integration of servicesDiscover and use services dynamically in an ad-hoc network

Adaptation to resource-limitationsQoR (Quality of Results) and resource tradeoff


MotivationSpatial Views Programming ModelSpatial Views Application Example

ExperimentsRelated WorkConclusions


Basic Programming Abstraction in Spatial Views

Nodes are abstracted as (service, location)

A physical node providing multiple services represent multiple virtual nodesA moving physical node represent multiple virtual nodesTime is an implicit dimension, the value of which is always the time when a node is visited and used


Spatial Views

A dynamic collection of virtual nodes that provide an interesting service and are located in an interesting space.

SV(service,space) {virtual node x | x.ofClass isA service and x.location space}.SV(service) = SV(service, anywhere).

Service: named with interfaces (abstract classes) with agreed-on semantics


Iterators and SelectorsIterators and Selectors instantiate Spatial Views.

Iterators enumerate all possible nodes in a view within a time constraint.Selectors stop after the first successful discovery, or timeout.

Time Constraint is necessary because Spatial Views can be infinite.


Application Example

Find a person in red on the third floor of a building

STARTSTOP


SpatialView cameraView = new SpatialView("Camera", BuschCampus.CoRE.Floor3);

foreach camera in cameraView do within 30000 {

Picture pic = camera.getPicture();

Rectangle redRegion = pic.findRegionInColor(Color.Red);

if (redRegion != null) {

SpatialView detectorView = new SpatialView("FaceDetector");

Rectangle face;

forany detector in detectorView do within 10000

face = detector.detectFaceInPicture(pic);

Location loc = camera.getLocation();

if (face !=null && face.isCloseTo(redRegion))

System.out.println("A person in red is found at " + loc);

else

System.out.println("Something red is found at " + loc);

}

}


MotivationSpatial Views Programming ModelSpatial Views Application Example

ExperimentsRelated WorkConclusions


Experiments: Implementation

Spatial Views Program

Spatial Views Compiler

Java Bytecode

Spatial Views Virtual MachineSpatial Views Library

Modified Sun javac 1.3.1Supporting space definition and Spatial View iteration.

Modified Sun KVM/CLDC 1.3Supporting migration and discovery


Experiment 1: Person Search Application

Execution TimeWith face detection: 23.1 secondsWithout face detection: 10.0 seconds

Routing: gossiping with backtracking


Experiment 2:Quality of Result v.s. Time/Resource Usage

QoR QoR

Resource Resource


Experiment 2:Effects of Time Constraints

Perfect answer: 8 nodes

Perfect Answer: 5 nodes


Execution time v.s. Network Link Failure Rate

(1)

0

2000

4000

6000

8000

10000

12000

60 70 80 90 100

Link Failure Rate (% )

Execu

tion T

ime (

ms)

w/ o constr.w/ constr.timeout


Quality of Result (QoR) v.s.

Network Link Failure Rate

0

1

2

3

4

5

6

7

8

Number of Nodes Visited

60 70 80 82 85 87 90 92 95 97 98


line


Execution time v.s. Network Link Failure Rate

(2)

0500

100015002000250030003500400045005000

60 70 80 90 100


Execu

tion T

ime (

ms)

w/ o constr.w/ constr.timeout


Quality of Result (QoR) v.s.

Network Link Failure Rate

00.5

11.5

22.5

33.5

44.5

5

Number of Nodes Visited

60 70 80 82 85 87 90 92 95 97 98


net


Related WorkLocation technology: GPS, APS(Rutgers), Cricket (MIT), RADAR (Microsoft Research), Active Badges (AT&T), Bats (AT&T).Service discovery: Jini (SUN), INS (MIT), SDS(Berkeley), SLP, Salutation, Bluetooth SDP.Migratory execution: Smart Messages (Rutgers), Active Messages (Berkeley) and Mobile Agents including D’Agent (Dartmouth), Java Aglet (IBM).Sensor Networks: nesC(Berkeley)


ConclusionsSpatial Views is a programming model for dynamic and resource-limited networks of embedded systemsThe first space-aware programming model with best-effort semanticsSimple and expressive, prototype and applications implemented


Thank you!


Experimental Setup

10 iPAQ PDA’sH3800 & H3600 Mercury camera sleeves

802.11b wireless networkFamiliar Linux, Kernel 2.4.18Customized KVM/CLDC 1.0.3Customized Javac 1.3.1


Experiment 2:One-hop Migration Time

050

100150200250300350400450

6928

4

7761

2

8580

8

9400

0

1021

92

1103

84

1185

76

Live data size

Mig

rati

on t

ime

iPAQ-wirelessPC-wired


Syntax<SpatialView> <identifier><Iterator> foreach <var> in <SpatialView> do <TimeConstraint> [constraints] <statement><Selector> forany <var> in <SpatialView> do <TimeConstraint> [constraints] <statement><TimeConstraint> within <NumberOfMilliSeconds>

Spatial Views: Space-aware Programming for Networks of Embedded Systems

Documents