搜索
StudiesinProcessSimplificationAshokDandekarDewayneE.PerryLawrenceG.VottaFujitsuNetworkSoftwareProductionResearchSoftwareProductionResearchCommunications,Inc.BellLaboratoriesBellLaboratoriesRichardsonTX75082MurrayHillNJ07974NapervilleIL60566A_Dandekar@fujitsu-fnc.comdep@research.bell-labs.comvotta@research.bell-labs.comAbstractOneofthemajorproblemswithsoftwaredevelopmentprocessesistheircomplexity.Hence,oneoftheprimarymotivationsinprocessimprovementisthesimplificationofthesecomplexprocesses.Wereportasetofstudiestoexplorevarioussimplificationapproachesandtechniques.Weusedtheavailableprocessdocumentation,questionnairesandinterviews,andasetofprocessvisualizationtoolfragments(pfv)togainanunderstandingoftheprocessunderexamination.Wethenusedthreebasicanalysistechniquestolocatecandidatesforsimplificationandimprovement:valueaddedanalysis,timeusageanalysis,andalternativesanalysis.Allthreeapproachesprovedeffectiveinisolatingproblemareasforimprovement.Theproposedsimplificationsresultedinasavingsof20%incost,20%inhumaneffort,40%inelapsetime,anda30%reductioninthenumberofactivities.1.IntroductionandOverviewTheprocesssystemthatisthecontextofourstudycameintobeingasaresultofamanagementedict(theyhadheardthatprocesseswereagoodthingandthattheyoughttobedefined).This‘‘goodthing’’beganwithvariouspeopleandgroupsdefiningtheprocesstheywereresponsibleformoreorlessinisolationfromalltheotherprocessesbeingdefined.Theseprocessesthengrewandevolvedinanunmanagedfashion,dueprimarilytothelackofeffectivemanagementoftheprocesssystemarchitecture.Thisunmanageabilitywasthenexacerbatedbythemergeroftwoprocesssystemsforsimilarproducts.In[6],[7]and[11]wereportedourexperienceintryingtounderstandthecurrentstateofthisprocesssystemanditsarchitecture.Oneoftheunderlyingrootcausesofthecurrentstateofboththesystemanditsarchitectureisthecomplexityoftheprocesses(boththeirinherentcomplexityandthecomplexityoftheirvisualizations).Therearetwofactorsthatcontributetosystemcomplexity:theexceedinglylargenumberofinteractionsamongtheprocessesandtheexceedinglylargenumberandwidevarietyofartifactsexchangedamongtheseprocesses.Factorsanalogoustothesecontributetothecomplexityofindividualprocesses:thenumberofinteractionsamongprocesselements,thenumberandvarietyofartifactsexchangedamongtheprocesselements,andthecontoursofthevariouspathsthroughtheprocess.Inthispaperwereportasetofstudiesinwhichourgoalistosimplifyaspecificprocess—thatis,tounderstandandresolvetheproblemsofinternalcomplexity.Wechosesimplificationratherthanre-engineeringasourapproachforseveralreasons.First,simplificationiseasiertodothanre-engineeringandismuchsmallerinitsscope.Second,wewantedtoseewhatcouldbegainedbythisapproachasanexamplefortheotherprocesssubsystems.Andthird,itwasourintuitionthattherewasmuchtobegainedintermsofcomplexity,costandintervalreductionwithoutgoingtothelengthsoffull-scalere-engineering.Ourintuitionshaveproventobecorrect.Ourcriteriaforchoosingaparticularprocesssubsystemwere•theprocessneededtobecustomerfocused,•theprocessneededtohaveknownproblems,and•thepeopleusingtheprocessesneededtobewillingtoparticipateinthestudy.Onthebasisofthesecriteria,wechosethecustomerdocumentationprocessesasthetrialstudyprocessesandwithinthatsetofprocessesthenewdocumentsprocessasthespecificfocus.Wethenspentabout4monthsmeetingweeklytodetermineappropriatesimplificationsandimprovements,andconcurrently,buildingthesupportingempiricalbaselineandtechnologyinfrastructure.Duringthisperiodweusedthreedifferentexperimentaltechniquesinouriterativeapproachtogaininginsightandunderstanding.Inthefollowingsectionsweintroduceourphilosophicalapproachtoprocessimprovement,delineatethescopeofthestudy,discussoureffortstounderstandtheprocesses,summarizeourstudiesandtheirresults,anddistilltheinsightsandlessonswelearned.2.TheApproachWeFoundUsefulWeusedabasicstraightforwardprocessimprovementparadigminwhichweiterativelybuildourunderstandingoftheprocessanditscontext,cost,qualityandintervalcharacteristics,thesupportingtechnology,andthevariousalternativeimprovements.Weclassifytheexperimentaltechniquesthatweusetobuildthisunderstandingintotwoclasses:retrospectiveandprospective.Retrospectivetechniquesusevariousexistinglegacydocumentsanddata,aswellasvariousformsofpersonalandgroupexperiencereports.Prospectivetechniquesusevariousformsofproactivedatagatheringandexperimentstogaindetailedunderstandingofcost,qualityandintervalfactors,aswellasvariousprocesscontrolfactors(thatis,causalconnectionsamongprocesselements).OnecommontechniqueforstudyingprocessesisGQM(Goal,Question,Metric)[2].Thisisprimarilyaformalizationoftheempiricalparadigmtailoredtounderstandingsoftwareproductsandprocesses.Assuchitisprimarilyprospectiveandnotretrospective.Sinceunderstandingtheprocessbeforeimprovingitisfundamental[12]andsinceretrospectivestudiescansupplyalargedegreeofunderstandingatverylowexperimentalcost,theGQMapproachdidnotseemappropriatefortheproblemsweaddressed.Thus,inouriterationsovervariousaspectsofthestudiedprocessweuseamixtureofretrospectiveandprospectivestudies.3.TheScopeoftheStudyThecustomerdoc