managing quality of context in pervasive computing

34
Intelligent Intelligent S S pace pace 國國國國國國國國國國國國國 國國國國國國國國 國國國國國國國國國國國國國 國國國國國國國國 Managing Quality of Managing Quality of Context in Pervasive Context in Pervasive Computing Computing Authors Authors Y.Bu, T.Gu, X.Tao, J.Li, S.Chen, and J.Lu Y.Bu, T.Gu, X.Tao, J.Li, S.Chen, and J.Lu Proceedings of 6th IEEE International Conference on Quality Software (QSIC’06) Reporter Reporter C.F.Liao ( C.F.Liao ( 廖廖廖 廖廖廖 ) Apr 27,2007 Apr 27,2007

Upload: abdul-cline

Post on 02-Jan-2016

33 views

Category:

Documents


4 download

DESCRIPTION

Managing Quality of Context in Pervasive Computing. Authors Y.Bu, T.Gu, X.Tao, J.Li, S.Chen, and J.Lu Proceedings of 6th IEEE International Conference on Quality Software (QSIC’06) Reporter C.F.Liao ( 廖峻鋒 ) Apr 27,2007. HCI Journal. IEEE Transactions on Software Engineering. - PowerPoint PPT Presentation

TRANSCRIPT

Page 1: Managing Quality of Context in Pervasive Computing

IntelligentIntelligent SSpacepace國立台灣大學資訊工程研究所 智慧型空間實驗室國立台灣大學資訊工程研究所 智慧型空間實驗室

Managing Quality of Context Managing Quality of Context in Pervasive Computingin Pervasive Computing

AuthorsAuthors

Y.Bu, T.Gu, X.Tao, J.Li, S.Chen, and J.LuY.Bu, T.Gu, X.Tao, J.Li, S.Chen, and J.Lu

Proceedings of 6th IEEE International Conference on Quality Software (QSIC’06)

ReporterReporter

C.F.Liao (C.F.Liao ( 廖峻鋒廖峻鋒 ))

Apr 27,2007Apr 27,2007

Page 2: Managing Quality of Context in Pervasive Computing

IntelligentIntelligent SSpacepace國立台灣大學資訊工程研究所 智慧型空間實驗室國立台灣大學資訊工程研究所 智慧型空間實驗室

2/30

Context-Aware Middleware for the Smart Context-Aware Middleware for the Smart EnvironmentsEnvironments

OSGi

Middleware forSmart Home

Middleware forSmart Environments

Univ. of Florida (USA)(OSCAR)(OSCAR)

Semantic Web Ontology

新加坡大學新加坡大學(SOCAM)(SOCAM)

Agent Oriented

Georgia TechGeorgia Tech(Context-Toolkit)(Context-Toolkit)

Maryland Univ.Maryland Univ.(CoBra, SOUPA)(CoBra, SOUPA)

HK PolytechnicHK Polytechnic(MobiPADS)(MobiPADS)

Washington UniversityWashington University (LIME)(LIME)

Jini

BerkleyBerkley(Context-Fabric)(Context-Fabric)

Univ. College LondonUniv. College London(CRISMA)(CRISMA)

HCI JournalIEEE Transactions on Software Engineering

ACM Transactions on Software Engineering and Methodology

Page 3: Managing Quality of Context in Pervasive Computing

IntelligentIntelligent SSpacepace國立台灣大學資訊工程研究所 智慧型空間實驗室國立台灣大學資訊工程研究所 智慧型空間實驗室

3/30

OutlineOutline

IntroductionIntroduction Quality-based Context ManagementQuality-based Context Management

Context Quality Measurements ER-Ontology Context Model Quality-based Context Processing Context Pooling

ExperimentsExperiments ConclusionConclusion

(RLR and the Case Study sections are skipped in this presentation)

Page 4: Managing Quality of Context in Pervasive Computing

IntelligentIntelligent SSpacepace國立台灣大學資訊工程研究所 智慧型空間實驗室國立台灣大學資訊工程研究所 智慧型空間實驗室

4/30

Using RDF as a Common Context Using RDF as a Common Context Representation FormatRepresentation Format

Sensor

ContextProvider

ContextProvider

Id=John,activity=lie down,place= bed

Activity Recognition

Module

Activity Recognition

Module

raw data Context(John,has posture,lie-down)

(John,location,bed)RDF

SOCAMSOCAM

RDF = Resource Description FrameworkRDF = Resource Description Framework

Page 5: Managing Quality of Context in Pervasive Computing

IntelligentIntelligent SSpacepace國立台灣大學資訊工程研究所 智慧型空間實驗室國立台灣大學資訊工程研究所 智慧型空間實驗室

5/30

Describing Data with RDFDescribing Data with RDF

RDF is a W3C standard, which has the following RDF is a W3C standard, which has the following capabilitiescapabilities Able to describe most kinds of data. Able to describe the structural design of data sets. Able to describe relationships between data.

Format: Format: Example:Example:

(bedroom, contains, light1) (light1, state, “on”)

(subject, predicate, object)

Actually, all resources are represented by URI, for example:http://www.foo.bar/myhome/mybedroom#light1

Page 6: Managing Quality of Context in Pervasive Computing

IntelligentIntelligent SSpacepace國立台灣大學資訊工程研究所 智慧型空間實驗室國立台灣大學資訊工程研究所 智慧型空間實驗室

6/30

Representing Context with RDF NetworkRepresenting Context with RDF Network

LightSwitch1 state

on Literal

Resource

Bedroom

locatedIn

size

9

contains

TV1

Page 7: Managing Quality of Context in Pervasive Computing

IntelligentIntelligent SSpacepace國立台灣大學資訊工程研究所 智慧型空間實驗室國立台灣大學資訊工程研究所 智慧型空間實驗室

7/30

The Structure of this PaperThe Structure of this Paper

Current Context Applications can not work well in real world

Low Context Quality!

What do we mean by low “Context Quality”?Context Quality Model

A Context Management mechanism to Improve Context Quality.

startTimeedgeecurrentTim

ttledgefrequencyedge

rfedgei

ii

i .

..

.

Page 8: Managing Quality of Context in Pervasive Computing

IntelligentIntelligent SSpacepace國立台灣大學資訊工程研究所 智慧型空間實驗室國立台灣大學資訊工程研究所 智慧型空間實驗室

8/30

MotivationMotivation

Context-awarenessContext-awareness plays a key role in a plays a key role in a paradigm shift from traditional desktop paradigm shift from traditional desktop computing to pervasive computing.computing to pervasive computing.

Most context-aware applications are unlikely to Most context-aware applications are unlikely to work well in the real world.work well in the real world.

Two major factors:Two major factors: Inconsistent contexts The limited data gathering frequency

Page 9: Managing Quality of Context in Pervasive Computing

IntelligentIntelligent SSpacepace國立台灣大學資訊工程研究所 智慧型空間實驗室國立台灣大學資訊工程研究所 智慧型空間實驗室

9/30

Context Repository

Context Inconsistency Context Inconsistency

Room 311 Aisle3

(Mary,walkIn,Room311)

(Mary,walkIn,Room311)

(Mary,walkIn,Aisle3)

(Room311,disjointWith,Aisle3)

(Mary,walkIn,Aisle3)

t t+2t+1

Conflict!

It seems that we either have to check context repository constantly or some conflict-resolving techniques have to be developed.

Page 10: Managing Quality of Context in Pervasive Computing

IntelligentIntelligent SSpacepace國立台灣大學資訊工程研究所 智慧型空間實驗室國立台灣大學資訊工程研究所 智慧型空間實驗室

10/30

Data Gathering FrequencyData Gathering Frequency

t t+5 t+10

10 10 10 11 12 14 10 12 10 10 12

12

Real World

System

The temperature data gathering period is 2 seconds.

10 1010 10 12 12 10 10 10 10

Page 11: Managing Quality of Context in Pervasive Computing

IntelligentIntelligent SSpacepace國立台灣大學資訊工程研究所 智慧型空間實驗室國立台灣大學資訊工程研究所 智慧型空間實驗室

11/30

OutlineOutline

IntroductionIntroduction Quality-based Context ManagementQuality-based Context Management

Context Quality Measurements ER-Ontology Context Model Quality-based Context Processing Context Pooling

ExperimentsExperiments ConclusionConclusion

(RLR and the Case Study sections are skipped in this presentation)

Page 12: Managing Quality of Context in Pervasive Computing

IntelligentIntelligent SSpacepace國立台灣大學資訊工程研究所 智慧型空間實驗室國立台灣大學資訊工程研究所 智慧型空間實驗室

12/30

Evaluating Context QualityEvaluating Context Quality

Context Quality MeasurementsContext Quality Measurements Delay Time Context Correctness Probability Context Consistency Probability

A well-designed context-aware system should A well-designed context-aware system should have:have: Low Delay Time High Context Correctness Probability High Context Consistency Probability

Context Pooling

RCIR / RLR

Page 13: Managing Quality of Context in Pervasive Computing

IntelligentIntelligent SSpacepace國立台灣大學資訊工程研究所 智慧型空間實驗室國立台灣大學資訊工程研究所 智慧型空間實驗室

13/30

Delay TimeDelay Time

t t+k

An event happens

System know what happens in the real world

Sensor Data Gathering

ContextProcessing

Service Provision

Delay Time

The time interval between an event happens in real world and when it is recognized by the system.

Page 14: Managing Quality of Context in Pervasive Computing

IntelligentIntelligent SSpacepace國立台灣大學資訊工程研究所 智慧型空間實驗室國立台灣大學資訊工程研究所 智慧型空間實驗室

14/30

Context Correctness ProbabilityContext Correctness Probability

t t+5 t+10

TemperatureContext

10

Context Correctness Probability = 7/ 11 = 0.64Context Correctness Probability = 7/ 11 = 0.64

10 10 11 12 14 10 12 10 10 12

10 10 11 10 10 12

Real World

System

10 10 11 10 10

The raw context gathering period is 2 seconds

Error due to context conflict resolution

Page 15: Managing Quality of Context in Pervasive Computing

IntelligentIntelligent SSpacepace國立台灣大學資訊工程研究所 智慧型空間實驗室國立台灣大學資訊工程研究所 智慧型空間實驗室

15/30

OutlineOutline

IntroductionIntroduction Quality-based Context ManagementQuality-based Context Management

Context Quality Measurements ER-Ontology Context Model Quality-based Context Processing Context Pooling

ExperimentsExperiments ConclusionConclusion

(RLR and the Case Study sections are skipped in this presentation)

Page 16: Managing Quality of Context in Pervasive Computing

IntelligentIntelligent SSpacepace國立台灣大學資訊工程研究所 智慧型空間實驗室國立台灣大學資訊工程研究所 智慧型空間實驗室

16/30

Context and Context RepositoryContext and Context Repository

Room342CSIE

BuildinglocatedIn

BABA nodeedgenode ,,

Context Graph (Extended RDF Network)Context Graph (Extended RDF Network)

Context RepositoryContext Repository

ContextContext

Page 17: Managing Quality of Context in Pervasive Computing

IntelligentIntelligent SSpacepace國立台灣大學資訊工程研究所 智慧型空間實驗室國立台灣大學資訊工程研究所 智慧型空間實驗室

17/30

Context Graph is essentially an extended RDF Context Graph is essentially an extended RDF Network. Network.

Context GraphContext Graph

Mary

Room311

CSIE Building

locatedIn

locatedIn

locatedIn

Node

Implicit Edge Meta Edge

Raw Edge

What are the benefits of this extension…?

Page 18: Managing Quality of Context in Pervasive Computing

IntelligentIntelligent SSpacepace國立台灣大學資訊工程研究所 智慧型空間實驗室國立台灣大學資訊工程研究所 智慧型空間實驗室

18/30

Persistent and Dynamic EdgesPersistent and Dynamic Edges

Room342CSIE

BuildinglocatedIn

TomCSIE

BuildinglocatedIn

Persistent Edge.The relationship that is unlikely to change.

Dynamic Edge.The relationship that is changing with time.

Page 19: Managing Quality of Context in Pervasive Computing

IntelligentIntelligent SSpacepace國立台灣大學資訊工程研究所 智慧型空間實驗室國立台灣大學資訊工程研究所 智慧型空間實驗室

19/30

OutlineOutline

IntroductionIntroduction Quality-based Context ManagementQuality-based Context Management

Context Quality Measurements ER-Ontology Context Model Quality-based Context Processing Context Pooling

ExperimentsExperiments ConclusionConclusion

(RLR and the Case Study sections are skipped in this presentation)

Page 20: Managing Quality of Context in Pervasive Computing

IntelligentIntelligent SSpacepace國立台灣大學資訊工程研究所 智慧型空間實驗室國立台灣大學資訊工程研究所 智慧型空間實驗室

20/30

Context Processing ProcedureContext Processing Procedure

Raw Context Gathering

InconsistencyResolution

Row LevelRefactoring

ContextRepository

Rule-basedReasoning

Rules

TriggeringApplications

Updating ContextRepository

Ontology-basedReasoning

OntologyContextRepository

RCIR RLR

JENA

Not-addressed in this paperJENA is a Semantic Web Framework for Java, Welcome to the lecture on 5/17 at R310

Page 21: Managing Quality of Context in Pervasive Computing

IntelligentIntelligent SSpacepace國立台灣大學資訊工程研究所 智慧型空間實驗室國立台灣大學資訊工程研究所 智慧型空間實驗室

21/30

Inconsistency Resolution (Definitions)Inconsistency Resolution (Definitions)

Conflict PairConflict Pair

Conflict Set Conflict Set ba edgeedge ,

Mary Room311locatedIn

Mary Room311locatedIn

Conflict

ba edgeedge ,

dc edgeedge ,

fe edgeedge ,

hg edgeedge ,We use an edge to represent a context instance here.

Page 22: Managing Quality of Context in Pervasive Computing

IntelligentIntelligent SSpacepace國立台灣大學資訊工程研究所 智慧型空間實驗室國立台灣大學資訊工程研究所 智慧型空間實驗室

22/30

Inconsistency Resolution by Inconsistency Resolution by RFRF

Core ideaCore idea When resolving conflicts, more frequent contexts

have more priority than infrequent ones. RF (Relative Frequency): Using TTL (Time to live) to

transform static frequency to dynamic frequency.

Term definitionsTerm definitions Edge TTL

The time period in which a context is valid.

Edge Frequency Edge Start Time

Page 23: Managing Quality of Context in Pervasive Computing

IntelligentIntelligent SSpacepace國立台灣大學資訊工程研究所 智慧型空間實驗室國立台灣大學資訊工程研究所 智慧型空間實驗室

23/30

Relative Frequency ( Relative Frequency ( rf ))

ExampleExample TTL = 2s Frequency = 1/6 (次 /s)

t t+6t+2 t+12t+8

startTimeedgeecurrentTim

ttledgefrequencyedge

rfedgei

ii

i .

..

.(for persistent edges)

(for dynamic edges)

Page 24: Managing Quality of Context in Pervasive Computing

IntelligentIntelligent SSpacepace國立台灣大學資訊工程研究所 智慧型空間實驗室國立台灣大學資訊工程研究所 智慧型空間實驗室

24/30

Raw Context Inconsistency Resolution Raw Context Inconsistency Resolution (RCIR)(RCIR)

Raw Context Sets(Mary,walkIn,Room311)

(John,walkIn, A)

(Mary,walkIn, Aisle3)

(Tom,walkIn, A)

Jena’s Conflict Detection Mechanism

(edge,edge)

Conflict Sets (edge,edge)(edge,edge)

(edge,edge)

(edge,edge)

Next edge type

Consistent Sets(edge,edge)(edge,edge)

No more edges

(walkIn,walkIn),rf=0.9

(walkIn,walkIn),rf=0.8

(walkIn,walkIn),rf=0.6

(walkIn,walkIn),rf=0.4

Preserve a pair that have highest rf value.

Page 25: Managing Quality of Context in Pervasive Computing

IntelligentIntelligent SSpacepace國立台灣大學資訊工程研究所 智慧型空間實驗室國立台灣大學資訊工程研究所 智慧型空間實驗室

25/30

Context RefactoringContext Refactoring

If a raw edge is changed, its related implicit edgIf a raw edge is changed, its related implicit edges should also be changed.es should also be changed.

The RLR (Raw Level Refactoring)algorithm aims The RLR (Raw Level Refactoring)algorithm aims to remove edges that are dependent to in-existinto remove edges that are dependent to in-existing raw edges.g raw edges.

Page 26: Managing Quality of Context in Pervasive Computing

IntelligentIntelligent SSpacepace國立台灣大學資訊工程研究所 智慧型空間實驗室國立台灣大學資訊工程研究所 智慧型空間實驗室

26/30

Conflict!

Context Refactoring: An ExampleContext Refactoring: An Example

LightSwitch state

on

Toilet 1

locatedIn

Aisle 3

contains

Tom

Bedroom contains

contains

contains

(Toilet, contains,”Tom”) (Aisle 3, contains, “Tom”)

(Bedroom, contains, “Tom”)

Page 27: Managing Quality of Context in Pervasive Computing

IntelligentIntelligent SSpacepace國立台灣大學資訊工程研究所 智慧型空間實驗室國立台灣大學資訊工程研究所 智慧型空間實驗室

27/30

Context PoolingContext Pooling

Context Repository

Application A RDQL

Context Pool

Context Change

InvalidateContext Manager

Pooling the unchanged context Pooling the unchanged context nodes in local cache to reduce nodes in local cache to reduce network traffic overhead.network traffic overhead.

Page 28: Managing Quality of Context in Pervasive Computing

IntelligentIntelligent SSpacepace國立台灣大學資訊工程研究所 智慧型空間實驗室國立台灣大學資訊工程研究所 智慧型空間實驗室

28/30

OutlineOutline

IntroductionIntroduction Quality-based Context ManagementQuality-based Context Management

Context Quality Measurements ER-Ontology Context Model Quality-based Context Processing Context Pooling

ExperimentsExperiments ConclusionConclusion

(RLR and the Case Study sections are skipped in this presentation)

Page 29: Managing Quality of Context in Pervasive Computing

IntelligentIntelligent SSpacepace國立台灣大學資訊工程研究所 智慧型空間實驗室國立台灣大學資訊工程研究所 智慧型空間實驗室

29/30

Performance EvaluationPerformance Evaluation

2 Intel Xeon CPUs, 4G RAM, Linux OS2 Intel Xeon CPUs, 4G RAM, Linux OS SensorSensor

Mica / Cricket (MIT)

PlatformPlatform OSGi Platform 1257 RDF triples

Page 30: Managing Quality of Context in Pervasive Computing

IntelligentIntelligent SSpacepace國立台灣大學資訊工程研究所 智慧型空間實驗室國立台灣大學資訊工程研究所 智慧型空間實驗室

30/30

ConclusionsConclusions

The authors proposed a Context Quality The authors proposed a Context Quality Measurements Model based on their Measurements Model based on their experiences of designing context-aware experiences of designing context-aware applications.applications.

Several mechanisms are proposed to Several mechanisms are proposed to increase the context quality:increase the context quality: ER-Ontology Context Model RCIR / RLR Context Pooling

Page 31: Managing Quality of Context in Pervasive Computing

IntelligentIntelligent SSpacepace國立台灣大學資訊工程研究所 智慧型空間實驗室國立台灣大學資訊工程研究所 智慧型空間實驗室

31/30

DiscussionsDiscussions

The limitation of context resolution mechanism.The limitation of context resolution mechanism. Raw context gathering period.Raw context gathering period.

Page 32: Managing Quality of Context in Pervasive Computing

IntelligentIntelligent SSpacepace國立台灣大學資訊工程研究所 智慧型空間實驗室國立台灣大學資訊工程研究所 智慧型空間實驗室

32/30

Limitations of Context ResolutionLimitations of Context Resolution

Sensor

Activity Recognition Agent

OSGiPlatformApplications

Bio-informationAgent

??

Raw Data

Bill is walking

Bill is sleeping

Actually, I’m sleep walking

Page 33: Managing Quality of Context in Pervasive Computing

IntelligentIntelligent SSpacepace國立台灣大學資訊工程研究所 智慧型空間實驗室國立台灣大學資訊工程研究所 智慧型空間實驗室

33/30

Raw Context Gathering PeriodRaw Context Gathering Period

The gathering period is important to both The gathering period is important to both performance and effectiveness.performance and effectiveness. To short – the processing mechanism will degrade to

piece by piece processing. To long – to much inconsistency, the RCIR algorithm

will have low performance.

Page 34: Managing Quality of Context in Pervasive Computing

IntelligentIntelligent SSpacepace國立台灣大學資訊工程研究所 智慧型空間實驗室國立台灣大學資訊工程研究所 智慧型空間實驗室

34/30

OutlineOutline

IntroductionIntroduction Quality-based Context ManagementQuality-based Context Management

Context Quality Measurements ER-Ontology Context Model Quality-based Context Processing Context Pooling

ExperimentsExperiments ConclusionConclusion