Practical and Representative Faultloads for Large-Scale Software Systems

Authors

Abstract

Thefaultloadisoneofthemostcriticalelementsofexperimentaldependabilityevaluation.Itshouldembodyarepeatable,portable,representativeandgenerallyacceptedfaultset.Concerningsoftwarefaults,thedefinitionofthatkindoffaultloadsisparticularlydifficult,asitrequiresamuchmorecomplexemulationmethodthanthetraditionalstuck-atorbit-flipusedforhardwarefaults.Althoughfaultloadsbasedonsoftwarefaultshavealreadybeenproposed,thechoiceofadequatefaultinjectiontargets(i.e.,actualsoftwarecomponentswherethefaultsareinjected)isstillanopenandcrucialissue.Furthermore,knowingthatthenumberofpossiblesoftwarefaultsthatcanbeinjectedinagivensystemispotentiallyverylarge,theproblemofdefiningafaultloadmadeofasmallnumberofrepresentativefaultsisofutmostimportance.Thispaperpresentsacomprehensivefaultinjectionstudyandproposesastrategytoguidethefaultinjectiontargetselectiontoreducethenumberoffaultsrequiredforthefaultloadandexemplifiestheproposedapproachwitharealweb-serverdependabilitybenchmarkandalarge-scaleintegervectorsortapplication.

Keywords

Experimental dependability evaluation, Dependability benchmarking, Injection of softwarefaults

Subject

Fault injection

Journal

Journal of Systems and Software, Vol. 103, pp. 182-197, Elsevier, February 2015

DOI