Faults and fault-tolerance.ppt

Faultsandfault-toleranceOneofthesellingpointsofadistributedsystemisthatthesystemwillcontinuetoperform(atsomelevel)ponents/processes/,andthentrytomaskit,(MeanTimeBetweenFailures)andMTTR(MeanTimeToRepair)monlyusedmetricsintheengineeringworldAclassificationoffailuresCrashfailureOmissionfailureTransientfailureSoftwarefailureSecurityfailureByzantinefailureTemporalfailureEnvironmentalperturbationsCrashfailuresCrashfailure=. Crashfailureisaformof“nice”,itcanbedetectedusingtimeout,butinaasynchronoussystem,estricky. --stopbehaviorhelpdetectwhichprocessorhasfailed. Ifasystemcannottoleratefail-stopfailure,,likeTransmittermalfunctionBufferoverflowCollisionsattheMAClayerReceiveroutofrangeTransientfailure(Hardware),weakbatteries,lightning,radio-frequencyinterferences,cosmicraysetc.(Software),%ofbugsinIBMDB2productioncodearenon-deterministicandtransient(JimGray)NotHeisenbergSoftwarefailuresCodingerrororhumanerror OnSeptember23,1999,NASAlostthe$125millionMarsorbiterspacecraftbecauseoneengineeringteamusedmetricunitswhileanotherusedEnglishunitsleadingtoanavigationfiasco, MarspathfindermissionlandedflawlesslyontheMartialsurfaceonJuly4,,municationfailedduetoadesignflawinthereal-,(continued)Memoryleak Operatingsystemsmaycrashw