564CHAPTER12DISTRIBUTEDFILESYSTEMS12.5TowhatextentdoesSunNFSdeviatefromone-copyfileupdatesernafitip~?onstructascenarioinwhichtwouser-levelprocessessharingafilewouldoperatecorrectlyinasingleUNIXhostbutwouldobserveinconsistencies whenrunningin differenthosts.~p#ge54212.6SunNFSaims tosupportheterogeneousdistributedsystemsbytheprovisionofanoperatingsystem—independentfileservice.WhatarethekeydecisionsthattheimplementerofanNFSserverforanoperatingsystemotherthanUNIX~,,’vouldhavetake?WhatconstraintsshouldanunderlyingfilingsystemobeytobesuitablefortheimplementationofNFSservers?page53612.7WhatdatamusttheNFSclientmoduleholdonbehalfofeachuser-levelprocess?~pa~es5i36—537/12.8OutlineclientmoduleimplementationsfortheUNIXopen()andre~9systemcatusingtheNFSRPCcallsofFigure12.9,(i) withoutand(ii)withaclient~ache.~~ages5~8,54213.112.9ExplainwhytheRPC interfacetoearlyimplementationsofNFSispote~tiallyinsecure.13.2Thesecurity loopholehasbeenclosedinNFS3bytheuseofencr~tion.:owis~e13.3encryptionkeykeptsecret?Is thesecurityofthekeyadequate?1~j~ges~~13412.10AfterthetimeoutofanRPCcalltoaccessafileonahard-mountedfil~‘stemtheNFS135clientmoduledoesnotreturncontroltotheuser-levelprocessthatoriginatedthecall.Why?pag’53913.612.11HowdoestheNFS automounterhelp toimprovetheperformance~lscalabilityofNFS?page541ThisE12.12Howmanylookupcallsareneededtoresolveafive-partpathnañie~forexampleprOce~/usr/users/jim/code/xyz.c)forafilethatisstored onanNFSserver?W~iaiisthereasontheirforperformingthetranslationstep-by-step?page540differe12.13Whatconditionmustbefulfilledbytheconfigurationofthemounttai{ièsattheclientcomputersforaccesstransparencytobeachievedinanNFS-basedfihiii~gsystem?~omepage54012.14HowdoesAFSgaincontrolwhenanopenorclosesystemcallreferri~igtoatileinthespsharedfilespace isissuedbyaclient?~‘‘page549suppo12.15ComparetheupdatesemanticsofUNIXwhenaccessinglocalfiles~ith~thoseofNFSSystenandAFS.Underwhatcircumstancesmightclientsbecomeawareofthedifferences?fpages542,554navigadataar12.16HowdoesAFSdealwiththeriskthatcallbackmessagesmaybelost?page552112.17WhichfeaturesoftheAFSdesign makeitmorescalablethanNFS?W~arethelimitsX.500onitsscalability,assumingthatserverscan beaddedasrequited?Whicidevelopmentsoffergreaterscalability?pa~es54~5,556,561qg:i•~672CHAPTER15COORDINATIONANDAGREEMENT15.5Adaptthecentralserveralgorithmformutual exclusionto handletcrashfailureofL15.17Showticlient(inanystate),assumingthattheserveriscorrectandgiy~nareliablefailFIFO-ordetector.Commentonwhethertheresultantsystemisfault-to1~rant.What w.thecasehappenifaclientthatpossessesthetokeniswronglysuspectedtó~havêfailed?“.causally63615.8Suggest15.6Giveanexampleexecutionofthering-basedalgorithmtoshowthatprocessesaregroups.necessarily grantedentrytothecriticalsectioninhappened-before>6rder.pag~‘~15.19Indiscu15.7Inacertainsystem,eachprocesstypicallyusesacriticalsec~iok;inanytimesbefsubsetsj4;~..1~anotherprocessrequiresit.ExplainwhyRicartandAgrawalasmulticast-basedmumujilcasexclusionalgorithmisinefficientforthiscase,anddescribe ,l~ó*tojmproveperformance.Doesyouradaptationsatisfylivenesscondition~page6315.0Construc15.8Inthebullyalgorithm,arecoveringprocessstarts anelectionanE1~villbecometheiarehablccoordinatorifithasahigheridentifierthan thecurrentincumbeiit~Is thisanec...15.2Wegavefeatureofthealgorithm?pageswhichin15.9Suggesthowtoadapt thebullyalgorithmtodealwithtemporâ~’networkp~itiWhy,in.reliableh(slow communication)andslowprocesses...pageifwecou15.10DeviseaprotocolforbasicmulticastoverIPmulticast.page’consider15.11How,ifatall,shouldthedefinitionsofintegrity,agreementai~d~alidityorrelia15.22Considermulticastchangeforthecaseofopengroups?.~;page’whichus’proposed15.12Explainwhyreversingtheorderofthelines‘R-delivermai1~~f#≠~)then~valueNcmulticast(g,m);end~finFigure15.9makesthealgorithmno1oñg~rsatis~junifresult~eagreement. DoesthereliablemulticastalgorithmbasedonIPmu1~ic~ssatis1~’unifsysten’s~agreement?~.pagethusto th15.13ExplainwhetherthealgorithmforreliablemulticastoverIPmultistworksforopen15.23Showthawellasclosed groups.Givenanyalgorithmforclosedgroups, how,simply,canfaulty,ifderiveanalgorithmforopengroups?page’15.14ExplainhowtoadaptthealgorithmforreliablemulticastoverIP.multicasttoelimithehold-backqueue—sothatareceivedmessagethatisnotadupii~àtecanbedelivimmediately,butwithoutanyorderingguarantees.Hint:usesetsdf~equencenumtorepresentthemessagesthathavebeendeliveredsofar.,‘.:page15.15Considerhowtoaddresstheimpracticalassumptionswemadeiinordertomeetvalidityandagreementpropertiesforthereliable multicastprdt~ôco1basedonI?multicast.Hint:addarulefordeletingretainedmessageswhenthey~havebeedeliveverywhere,andconsideraddingadummy‘heartbeat’rncss~ge: whichisndeliveredtotheapplication,butwhichtheprotocolsendsiftheàjiplicationhasmessagetosend.~•.pag15.16ShowthattheFIFO-ordered multicast algorithmdoesnotworkforoyevlappingi~’byconsideringtwomessagessentfromthesamesourcetotwocw~1appinggroupsconsideringaprocessintheintersectionofthosegroups.Adapttheprotocoltoworkthiscase.Hint:processesshouldincludewiththeirmessagesthelatestsequnumbersofmessagessenttoallgroups.~.-page’EXERCISES673ishfái1u’i~e~fany15.17Showthat,ifthebasicmulticastthat weuseinthealgorithmofFigure15.13isalsotreliablefailure FIFO-ordered,thentheresultanttotally-ordered multicastisalsocausallyordered.Isitnt.What.~vouldthecasethatanymulticastthatisbothFIFO-orderedandtotallyorderedistherebyfailed?••Qagecausallyordered?page65515.18Suggesthowtoadapt thecausallyorderedmulticastprotocoltohandleoverlappingrocessesarenotgroups.page 657er.page’63715.19IndiscussingMaekawa’smutual exclusionalgorithm,wegaveanexampleofthreenytimesbeforesubsetsofasetofthreeprocessesthatcouldleadtoadeadlock.Usethesesubsetsasist-based~niutuamulticastgroupsto showhowapairwise total orderingisnotnecessarilyacyclic.toimproye’itspage658page63915.20Constructasolutiontoreliable,totallyorderedmulticastinasynchronoussystem,usingareliable multicastandasolutiontotheconsensusproblem. page659becomethenewthisane~essai~y15.21Wegaveasolutiontoconsensusfromasolutiontoreliableandtotallyorderedmulticast,page644which involvedselectingthefirstvaluetobedelivered.Explainfromfirstprincipleswhy,inanasynchronoussystemwecouldnotinsteadderiveasolutionbyusingatworkpartitionsreliablebut
View Full Document