Automated Generation of Context-Aware Tests

Automated Generation of Context-Aware Tests

by

Zhimin Wang (University of Nebraska–Lincoln)

Sebastian Elbaum (University of Nebraska–Lincoln)

David S. Rosenblum (University College London)

Research funded in part by the EPSRC and the Royal Society

BackgroundUbiquitous Computing Systems

• Context-Aware– Execution driven by changes

in execution environment• Sensed through middleware

invocation of context handlersContext is an additional input space that

must be explored adequately during testing

• Adaptive– Execution must adapt to context changesChanges to multiple context variables

may occur simultaneously

An important emerging class of software systems

Nearby device IDs

LocationRadio signal

strength

Available memory

Battery level

Problem StatementTesting Ubiquitous Systems

Discovering concurrencyfaults in context-awareubiquitous systems

• Failures occur frequentlyduring attempts to handlemultiple context changes

• Existing testing techniques havelimited effectiveness in discoveringthe underlying faults

New SMS Found Wi-Fi

Additional Challenges during Testing

• Hard to control when and how to input contexts– Middleware can introduce noise– Interference can occur between context handlers

• Hard to define a precise oracle– Execution differs under various vectors of context inputs

• Real environment is not available– Too many sensors are required

• Sensed contexts can be inconsistent– Example: The border between rooms at an exhibition

Contributions

1. CAPPs: Context-Aware Program Points• A model of how context affects program execution

2. CAPPs-Based Test Adequacy Criteria• Criteria for evaluating test suite effectiveness• Defined in terms of sets of test drivers

• Sequence of CAPPs to cover

3. CAPPs-Driven Test Suite Enhancement• Automated exploration of variant interleavings of

invocations of context handlers• Schedules interleavings via special instrumentation

Application for Case StudyTourApp Released with the Context Toolkit

PDA

CommunicationMiddleware

DemoWidget 1

RegistrationWidget

Sensor Sensor

RegistrationBooth

Room 1

InterpreterWidget

TourApp

Remote dataconnection

Service …

Visitor

Application

EndWidget

Sensor

Exit Room

DemoWidget 2

Sensor

Room 2

Service

Tag

Application for Case StudyTourApp Released with the Context Toolkit

PDA

CommunicationMiddleware

DemoWidget 1

RegistrationWidget

Sensor Sensor

RegistrationBooth

Room 1

InterpreterWidget

TourApp

Remote dataconnection

Service …

Visitor

Application

EndWidget

Sensor

Exit Room

DemoWidget 2

Sensor

Room 2

Service

Tag• Location:

Registration Room

• Application Response:Pop-Up Registration Form

Application for Case StudyTourApp Released with the Context Toolkit

PDA

CommunicationMiddleware

DemoWidget 1

RegistrationWidget

Sensor Sensor

RegistrationBooth

Room 1

InterpreterWidget

TourApp

Remote dataconnection

Service …

Visitor

Application

EndWidget

Sensor

Exit Room

DemoWidget 2

Sensor

Room 2

Service

Tag• Location:

DemoRoom 1

• Application Response:Display Lecture Information

Application for Case StudyTourApp Released with the Context Toolkit

PDA

CommunicationMiddleware

DemoWidget 1

RegistrationWidget

Sensor Sensor

RegistrationBooth

Room 1

InterpreterWidget

TourApp

Remote dataconnection

Service …

Visitor

Application

EndWidget

Sensor

Exit Room

DemoWidget 2

Sensor

Room 2

Service

Tag• Power:

LowBattery!

• Application Response:Confine Display Updates

Overview of Testing Infrastructure

CAPPs Identifier

Program Instrumentor

Context Manipulator

Context Driver Generator

Context-Aware Program (P)

AnnotatedFlow Graph

Test Suite (T)

Selected ContextAdequacy Criteria

P

Achieved CoverageAnd

Test Case ExtensionFeedback on Coverage

Test Drivers (D)

Test Adequacy CriteriaContext Adequacy (CA)

• Test driver covering at least one CAPP in each type of context handler

• Examples– { capp1, capp2 }

– or { capp3, capp2 }

– or { capp2, capp1 }

11

12

16

13 19

20

21

22

23

Exit

Exit

11

14

16

15 26

27

32

Exit

Exit

28

31

29

30

Handler CFGtype="power"

Handler CFGtype="demo"

capp1

capp2

capp3

capp4

capp5

capp6

Test Adequacy CriteriaSwitch-to-Context Adequacy (StoC-k)

• Set of test drivers covering all possible combinations of k switches between context handlers

• StoC-1 Example:– { capp1, capp2 },

{ capp5, capp3 },

{ capp3, capp3 },

{ capp5, capp5 }

11

12

16

13 19

20

21

22

23

Exit

Exit

11

14

16

15 26

27

32

Exit

Exit

28

31

29

30

Handler CFGtype="power"

Handler CFGtype="demo"

capp1

capp2

capp3

capp4

capp5

capp6

Test Adequacy CriteriaSwitch-with-Preempted-Capp Adequacy(StoC-k-FromCapp)

11

12

16

13 19

20

21

22

23

Exit

Exit

11

14

16

15 26

27

32

Exit

Exit

28

31

29

30

Handler CFGtype="power"

Handler CFGtype="demo"

capp1

capp2

capp3

capp4

capp5

capp6

• Set of test drivers covering all possible combinationsof k switches between context handlers, with each switch exercised at every CAPP

• StoC-1-FromCapp Example:– { capp1, capp2 },

{ capp3, capp2 },

{ capp4, capp5 },

{ capp2, capp3 },

{ capp5, capp1 },

{ capp6, capp3 },

{ capp3, capp3 },

{ capp5, capp5 }

Case Study Design and SettingsTourApp

• 11 KLOC of Java, 4 seeded faults• Test suite of 36 end-to-end test cases• Executing test suite takes 10 minutes• Studied 4 versions:

– originalTourApp: unmodified originalmanipulatedTourApp: instrumented with calls to our

scheduler methods– delayShortTourApp: instrumented with 1–3 seconds

random delays (sleep())– delayLongTourApp: instrumented with 1–10 seconds

random delay (sleep())

Results: CostTimings with manipulatedTourApp

• Summary of study– Execution time increases with more demanding context

coverage criteria as more context scenarios are required

Results: FeasibilityDrivers in manipulatedTourApp

• Summary of study– Some test drivers were not realised in the application

within the set timeouts– Improvements in generation of D may be needed via

better flow-sensitive analysis

Results: EffectivenessPercentage of Contextual Coverageand Fault Detection

• Summary of study– Coverage decreases with more powerful criteria

• But only slightly for manipulatedTourApp

– manipulatedTourApp performs the best, especially with more powerful criteria

Related Work

• Deterministic testing of concurrent programs(Carver & Tai, Taylor et al.)– Concurrency intrinsic to program, not execution environment

• Metamorphic testing for context-aware applications(Tse et al.)– Oracles embodying metamorphic properties must be defined by

tester

• Data flow coverage criteria for context-aware applications(Lu et al.)– Does not support notion of CAPPs or manipulation of test

executions

• Random sleeps for test perturbation (Edelstein et al.)– Inferior to controlled scheduling of context switches

Conclusion

• Defined the CAPPs model of how context changes affect context-aware applications

• Defined test adequacy criteria in terms of this model

• Created an automated technique to guide test executions in a way that systematically explores many interesting context change scenarios

• Demonstrated the superiority of this technique in discovering concurrency faults

Future Work

• Investigate Additional Classes of Faults– User interface ‘faults’– Adaptation priority faults– Memory leaks

• ContextNotifier and TestingEmulator– Emulation infrastructure for testing– (Tools and libraries from vendors are pathetic!)

• Simulation-Driven Testing– Test case execution driven by mobility traces from

simulation runs

Questions?

http://www.ubival.org/