Distributed file systems dfs

Distributed file systems dfs

1. Distributed File Systems

2. Adistributedfilesystemisaresourcemanagement
componentofadistributedoperatingsystem.Itimplementsa
commonfilesystemthatcanbesharedbyalltheautonomous
computersinthesystem.
DISTRIBUTED FILE SYSTEM
Twoimportantgoals:
1.Networktransparency–toaccessfilesdistributedovera
network.Ideally,usersdonothavetobeawareofthe
locationoffilestoaccessthem.
2.HighAvailability-toprovidehighavailability.Usersshould
havethesameeasyaccesstofiles,irrespectiveoftheir
physicallocation.

3. ARCHITECTURE
Inadistributedfilesystem,filescanbestoredatany
machineandthecomputationcanbeperformedatany
machine.
Thetwomostimportantservicespresentinadistributedfile
systemarenameserverandcachemanager.
Anameserverisaprocessthatmapsnamesspecifiedby
clientstostoredobjectssuchasfilesanddirectories.
Themapping(alsoreferredtoasnameresolution)
Acachemanagerisaprocessthatimplementsfilecaching.
Infilecaching,acopyofdatastoredataremotefileserveris
broughttotheclient’smachinewhenreferencedbythe
client.
DISTRIBUTED FILE SYSTEM

4. Cachemanagerscanbepresentatbothclientsandfileservers.
Cachemanagersattheserverscachefilesinthemainmemory
toreducedelaysduetodisklatency.
Ifmultipleclientsareallowedtocacheafileandmodifyit,
thecopiescanbecomeinconsistent.
Toavoidthisinconsistencyproblem,cachemanagersatboth
serversandclientscoordinatetoperformdatastorageand
retrievaloperations.

5. Architecture of a Distributed File System

6. DISTRIBUTED FILE SYSTEM
Arequestbyaprocesstoaccessadatablockispresentedto
thelocalcache(clientcache)ofthemachine(client)on
whichtheprocessisrunning.
Iftheblockisnotinthecache,thenthelocaldisk,if
present,ischeckedforthepresenceofthedatablock.
Iftheblockispresent,thentherequestissatisfiedandthe
blockisloadedintotheclientcache.
Iftheblockisnotstoredlocally,thentherequestispassed
ontotheappropriatefileserver
Theserverchecksitsowncacheforthepresenceofthedata
blockbeforeissuingadiskI/Orequest.
Thedatablockistransferredtotheclientcacheinanycase
andloadedtotheservercacheifitwasmissingintheserver
cache.

7. MECHANISMS FOR BUILDING DISTRIBUTED
FILE SYSTEM
Mounting
Amountmechanismallowsbindingtogetherofdifferent
filenamespacestoformasinglehierarchicallystructuredname
space.
Two approaches to maintain the mount information:
Mount information can be maintained at clients, in which
case each client has to individually mount every required file
system. This approach is employed in the Sun network file
system. Since each client can mount a file system at any node
in the name space tree, every client need not necessarily see
an identical filename space.

8. Mount information can be maintained at servers, in which
case it is possible that every client sees an identical filename
space. If files are moved to a different server, then mount
information need only be updated at the servers. In the first
approach, every client needs to update its mount table.

9. Cachingiscommonlyemployedindistributedfilessystemsto
reducedelaysintheaccessingofdata.
Infilecaching,acopyofdatastoredataremotefileserveris
broughttotheclientwhenreferencedbytheclient.
Thetemporallocalityofreferencereferstothefactthatafile
recentlyaccessedislikelytobeaccessedagaininthenear
future.
Datacaneitherbecachedinthemainmemory(servercache)at
theserverstoreducediskaccesslatency.
Cachingimprovesfilessystemperformancebyreducingthe
delayinaccessingdata.
CACHING

10. Analternativeapproachistotreatcacheddataashints.
Inthiscase,cacheddataarenotexpectedtobecompletely
accurate.
However,validcacheentriesimproveperformance
substantiallywithoutincurringthecostofmaintaining
costconsistency.
Theclassofapplicationsthatcanutilizehintsarethose
whichcanrecoverafterdiscoveringthatthecacheddata
areinvalid.
HINTS

11. Transferringdatainbulkreducestheprotocolprocessingoverhead
atbothserversandclients.
Inbulkdatatransfer,multipleconsecutivedatablocksare
transferredfromserverstoclientsinsteadofjusttheblock
referencedbyclients.
Whilefilecachingamortizesthehighcostofaccessingremote
serversoverseverallocalreferencestothesameinformation
Bulktransferamortizestheprotocolprocessingoverheadanddisk
seektimeovermanyconsecutiveblocksofafile.
Bulktransfersreducefileaccessoverheadthroughobtaininga
multiplenumberofblockswithasingleseek;byformattingand
transmittingamultiplenumberoflargepacketsinasinglecontext
switch;andbyreducingthenumberofacknowledgementsthat
needtobesent.
BULK DATA TRANSFER

12. Encryptionisusedforenforcingsecurityindistributed
systems.
TheworkofNeedhamandSchroederisthebasisformostof
thecurrentsecuritymechanismsindistributedsystems.
Intheirscheme,twoentitieswishingtocommunicatewith
eachotherestablishakeyforconversationwiththehelpof
anauthenticationserver.
Itisimportanttonotethattheconversationkeyis
determinedbytheauthenticationserver,butisneverspent
inplain(unencrypted)texttoeitheroftheentities.
ENCRYPTION