A Model and Sensitivity Analysis of the Quality Economics.docx
《A Model and Sensitivity Analysis of the Quality Economics.docx》由会员分享,可在线阅读,更多相关《A Model and Sensitivity Analysis of the Quality Economics.docx(47页珍藏版)》请在冰豆网上搜索。
AModelandSensitivityAnalysisoftheQualityEconomics
AModelandSensitivityAnalysisoftheQualityEconomics
ofDefectDetection
Techniques
StefanWagner
Institutf╱rInformatik
TechnischeUniversit╝tM╱nchen
Boltzmannstr.3,D85748
Garchingb.M╱nchen,Germany
wagnerst@in.tum.de
ABSTRACT
Oneofthemaincostfactorsinsoftwaredevelopmentis
thedetectionandremovalofdefects.However,therelationships
andinfluencingfactorsofthecostsandrevenues
ofdefect-detectiontechniquesarestillnotwellunderstood.
Thispaperproposesananalytical,stochasticmodelofthe
economicsofdefectdetectionandremovaltoimprovethis
understanding.Themodelisabletoincorporatedynamic
aswellasstatictechniquesincontrasttomostothermodels
ofthatkind.Weespeciallyanalysethemodelwithstate-ofthe-
artsensitivityanalysismethodsto
(1)identifythemost
relevantfactorsformodelsimplificationand
(2)prioritise
thefactorstoguidefurtherresearchandmeasurements.
CategoriesandSubjectDescriptors
D.2.9[SoftwareEngineering]:
Management;D.2.8[Soft-
wareEngineering]:
Metrics;D.2.5[SoftwareEngineer-
ing]:
TestingandDebugging
GeneralTerms
Economics,Verification,Reliability
Keywords
Softwarequalityeconomics,qualitycosts,cost/benefit,defectdetection
techniques,sensitivityanalysis
1.INTRODUCTION
Thequalityofasoftwaresystemcanbedescribedusing
differentattributessuchasreliability,maintainabilityetc.
Therearealsovariousapproachestoimprovethequalityof
softwarewithdifferingemphasisontheseattributes.Constructive
methodscompriseonegroupthattriestoimprove
theoveralldevelopmentprocessinordertopreventtheintroduction
offaults.However,theprevalentapproachisstill
touseanalyticalmethods,alsocalleddefect-detectiontech-
Permissiontomakedigitalorhardcopiesofallorpartofthisworkfor
personalorclassroomuseisgrantedwithoutfeeprovidedthatcopiesare
notmadeordistributedforprofitorcommercialadvantageandthatcopies
bearthisnoticeandthefullcitationonthefirstpage.Tocopyotherwise,to
republish,topostonserversortoredistributetolists,requirespriorspecific
permissionand/orafee.
ISSTA?
6,July17?
0,2006,Portland,Maine,USA.
Copyright2006ACM1595932631/
06/0007...$5.00.
niques,tofindandremovefaults.Themainrepresentatives
ofthisapproacharetestsandreviews.
Anoftencitedestimate[26]relates50%oftheoveralldevelopment
coststotesting.Jones[14]stillassigns30?
0%to
qualityassuranceanddefectremoval.Hence,defect-detection
techniquesareapromisingfieldforcostoptimisations.
However,tobeabletooptimisetheusageofdefect-detection
techniques,weneedasuitableeconomicalmodelfirst.There
aresomeapproachesthatmodelsoftwarequalitycostsbut
mostlyonahighlevelofabstraction.Theeffectsofindividual
faultsandtheeffectivenessofdifferentdefect-detection
techniquesregardingthesefaultsarenottakenintoaccount.
Alsoin[28]itisdiscussedthat揷ostisclearlyacentralfactor
inanyrealisticcomparisonbutitishardtomeasure,
dataarenoteasytoobtain,andlittlehasbeendonetodeal
withit.?
Raietal.identifyin[31]mathematicalmodels
oftheeconomicsofsoftwarequalityassuranceasanimportant
researcharea.揂betterunderstandingofthecostsand
benefitsofSQAandimprovementstoexistingquantitative
modelsshouldbeusefultodecision-makers.?
1.1Problem
Theunderlyingquestionishowwecanoptimallyusedefect-
detectiontechniquestoimprovethequalityofsoftware.
Inparticular,weinvestigateinthispaperhowtheeconomical
relationshipsofdefect-detectiontechniquesandquality
canbemodelledandtheimportanceofthefactorsinterms
oftheinfluenceontheoutputandespeciallyitsvariance.
1.2Contribution
Weproposeananalyticalmodeloftheeconomicsofdefect-
detectiontechniquesincorporatingdifferenttypesofdefect
costs,thedifficultyoffindingafaultofdifferenttechniques
andtheprobabilityoffailureforafault.Thisallows
anevaluationofdifferenttechniquesandgivesabetterunderstanding
oftherelationships.Theusedinputfactorsare
prioritisedtosimplifythemodelandtoidentifythefactors
thataremostbeneficialtobefurtherinvestigated.Furthermore,
amodelbasedondefecttypesisderivedtoallow
asimplerapplicationonrealworldprojects.Thismodel
couldbeusedtopredictoptimalusageofdefect-detection
techniquesinthefuturebasedonoldprojectdata.
1.3Outline
Westartbydescribingsoftwarequalitycostsingeneral
andourunderstandingofthevariouscostfactorsinSec.2.
Sec.3proposesananalyticalmodeloftheeconomicsofdefect-
detectiontechniquesthatcontainsthecostsassociated
73
witheachfault.Thismodelissubjecttoasensitivityanalysis
basedonthedatafromanolderstudyinSec.4.Forthe
practicalapplicationofthemodelasimplifiedversionbased
ondefectclassesisderivedinSec.5.Sec.6givesrelated
workandfinalconclusionscanbefoundinSec.7.
2.SOFTWAREQUALITYCOSTS
Qualitycostsarethecostsassociatedwithpreventing,
finding,andcorrectingdefectivework.Basedonexperience
fromthemanufacturingarea[15,5]similarqualitycostmodels
havebeendevelopedexplicitlyforsoftware[18,35,19].
Thesecostsaredividedintoconformanceandnonconfor-
mancecosts,alsocalledcontrolcostsandfailureofcontrol
costs.Theformercomprisesallcoststhatneedtobespent
tobuildthesoftwareinawaythatitconformstoitsquality
requirements.Thiscanbefurtherbrokendowntopreven-
tionandappraisalcosts.Preventioncostsareforexample
developertraining,toolcosts,orqualityaudits,i.e.costs
formeanstopreventtheinjectionoffaults.Theappraisal
costsarecausedbytheusageofvarioustypesoftestsand
reviews.
Thenonconformancecostscomeintoplaywhenthesoftware
doesnotconformtothequalityrequirements.These
costsaredividedintointernalfailurecostsandexternalfail-
urecosts.Theformercontainscostscausedbyfailuresthat
occurduringdevelopment,thelatterdescribescoststhat
resultfromfailuresattheclient.Agraphicaloverviewis
giveninFig.1.Becauseofthedistinctionbetweenprevention,
appraisal,andfailurecoststhisisoftencalledPAF
model.
costofquality
preventioncostsappraisalcostsexternalfailure
conformancenonconformance
internalfailure
setupexecutionfaultremovaleffect
Figure1:
Overviewoverthecostsrelatedtoquality
WeaddfurtherdetailtothePAFmodelbyintroducing
themaintypesofconcretecoststhatareimportantfor
defect-detectiontechniques.Notethattherearemoretypes
thatcouldbeincluded,forexample,maintenancecosts.
However,weconcentrateonamorereliability-orientedview.
Theappraisalcostsaredetailedtothesetupandexecution
costs.Theformerconstitutingallinitialcostsforbuying
testtools,configuringthetestenvironment,andsoon.The
lattermeansallthecoststhatareconnectedtoactualtest
executionsorreviewmeetings,mainlypersonnelcosts.
Onthenonconformanceside,wehavefaultremovalcosts
thatcanbeattributedtotheinternalfailurecostsaswell
astheexternalfailurecosts.Thisisbecauseifwefounda
faultandwanttoremoveit,itwouldalwaysresultincosts
nomatterwhethercausedinaninternalorexternalfailure.
Actually,theredoesnothavetobeafailureatall.
Consideringcodeinspections,faultsarefoundandremoved
thathavenevercausedafailureduringtesting.Itisalso
agoodexamplethattheremovalcostscanbequitedifferent
regardingdifferenttechniques.Whenatestidentifiesa
failure,thereneedstobeconsiderableeffortspenttofind
thecorrespondingfault.Duringaninspection,faultsare
founddirectly.Faultremovalcostsalsocontainthecosts
fornecessaryre-testingandre-inspections.
Externalfailuresalsocauseeffectcosts.Theseareall
furthercostswiththefailureapartfromtheremovalcosts.
Forexample,compensationcostscouldbepartoftheeffect
costs,ifthefailurecausedsomekindofdamageatthe
customersite.Wemightalsoincludefurthercostssuchas
lossofsalesbecauseofbadreputationintheeffectcostsbut
donotconsideritexplicitlybecauseitsoutofscopeofthis
paper.
3.ANALYTICALMODEL
Wedescribeageneral,analyticalmodelofdefect-detection
techniquesinthefollowing.Itisgeneralwithrespect
tothevarioustypesoftechniquesitisabletoanalyse.We
mainlyanalysedifferenttypesoftestingwhichessentially
detectfailuresandstaticanalysistechniquesthatreveal
faultsinthecodeorotherdocuments.Amodelthatincorporates
allimportantfactorsforthesedifferingtechniques
needstousetheuniversalunitofmoney,i.e.,unitssuchas
euroordollar.Wefirstdescribethemodelanditsassumptions
ingeneral,andthengiveequationsforeachcomponent
ofthemodelforasingletechniqueandforthecombination
ofseveraltechniques.
3.1General
Inthissection,weconcentrateonanidealmodelofquality
economicsinthesensethatwedonotconsiderthepractical
useofthemodelbutwanttomirrortheactualrelationships
asfaithfullyaspossible.
3.1.1Components
Wedividethemodelinthreemaincomponents:
?
DirectcostsdA
?
FuturecoststA
?
Revenues/savedcostsrA
Thedirectcostsarecharacterisedbycontainingonlycosts
thatcanbedirectlymeasuredduringtheexecutionofthe
technique.Thefuturecostsandrevenuesarebothconcerned
withthe(potential)costsinthefieldbutcanbedistinguished
becausethefuturecostscontainthecoststhatare