05 scalable statistical bug isolation

《05 scalable statistical bug isolation》由会员上传分享，免费在线阅读，更多相关内容在行业资料-天天文库。

1、ScalableStatisticalBugIsolationBenLiblitMayurNaikAliceX.ZhengComputerSciencesDepartmentComputerScienceDepartmentDepartmentofElectricalEngineeringUniversityofWisconsin-MadisonStanfordUniversityandComputerScienceUniversityofCalifornia,Berkeley

2、ley.edu>AlexAikenMichaelI.JordanComputerScienceDepartmentDepartmentofElectricalEngineeringStanfordUniversityandComputerScienceDepartmentofStatisticsUniversityofCalifornia,BerkeleyAbstract1.IntroductionWepresentastatisticaldebuggingalgorithmthatisolate

3、sbugsinThispaperisaboutstatisticaldebugging,adynamicanalysisforprogramscontainingmultipleundiagnosedbugs.Earlierstatisticalidentifyingthecausesofsoftwarefailures(i.e.,bugs).Instrumentedalgorithmsthatfocussolelyonidentifyingpredictorsthatcorre-programsmonitortheirownbehaviorandproducefeedbackre-lat

4、ewithprogramfailureperformpoorlywhentherearemultipleports.Theinstrumentationexaminesprogrambehaviorduringex-bugs.Ournewtechniqueseparatestheeffectsofdifferentbugsandecutionbysampling,socompleteinformationisneveravailableidentifiespredictorsthatareassociatedwithindividualbugs.Theseaboutanysinglerun.

5、However,monitoringisalsolightweightandpredictorsrevealboththecircumstancesunderwhichbugsoccurthereforepracticaltodeployinaproductiontestingenvironmentoraswellasthefrequenciesoffailuremodes,makingiteasiertopri-tolargeusercommunities,makingitpossibletogatherinforma-oritizedebuggingefforts.Ouralgorit

6、hmisvalidatedusingseveraltionaboutmanyruns.Thecollecteddatacanthenbeanalyzedforcasestudies,includingexamplesinwhichthealgorithmidentifiedinterestingtrendsacrossallofthemonitoredexecutions.previouslyunknown,significantcrashingbugsinwidelyusedsys-Inourapproach,instrumentationconsistsofpredicatestested

7、tems.atparticularprogrampoints;wedeferdiscussingwhichpredicatesarechosenforinstrumentationtoSection2.AgivenprogrampointCategoriesandSubjectDescriptorsD.2.4[SoftwareEngineer-mayhavemanypredicatesthataresampledinde