distrubuted system bca tu distrubuted system bca tudistrubuted system bca tu

distrubuted system bca tudistrubuted system bca tudistrubuted system bca tudistrubuted system bca tudistrubuted system bca tu

1. UNIT: 2
Architecture
9/27/2020
BCA (T.U.) Distributed System
By: Er. Nabin Adhikari
1

2. •DistributedSystemsarecomposedofvarioushardwareand
software(collectivelycalledcomponents)thatcommunicate
witheachotheronlybytransferofmessages.These
componentsareplacedinsideasinglenetwork.
•So,whenwethinkaboutthearchitecturestylesfor
distributedcomputingthemaintaskwouldbetoensurethat
thesecancommunicate witheachotherovera
communicationnetwork.
9/27/2020
BCA (T.U.) Distributed System
By: Er. Nabin Adhikari
2

3. 2.1 Architectural Styles
•Westartourdiscussiononarchitecturesbyfirstconsideringthelogical
organizationofdistributedsystemsintosoftwarecomponents,alsoreferred
toassoftwarearchitecture.
•Thenotationofanarchitecturalstyleisimportant.Suchastyleis
formulatedintermsofcomponents,thewaythatcomponentsareconnected
toeachother,thedataexchangedbetweencomponentsandfinallyhow
theseelementsarejointlyconfiguredintoasystem.
•Acomponentisamodularunitwithwell-definedrequiredandprovided
interfacesthatisreplaceablewithinitsenvironment.
•Usingcomponentsandconnectors,wecancometovariousconfigurations,
which,intumhavebeenclassifiedintoarchitecturalstyles.
9/27/2020
BCA (T.U.) Distributed System
By: Er. Nabin Adhikari
3

4. •Severalstyleshavebynowbeenidentified,ofwhichthemost
importantonesfordistributedsystemsare:
1.Layeredarchitectures
2.Object-basedarchitectures
3.Data-centeredarchitectures
4.Event-basedarchitectures
9/27/2020
BCA (T.U.) Distributed System
By: Er. Nabin Adhikari
4

5. •Thebasicideaforthelayeredstyleissimple:componentsareorganizedinalayered
fashionwhereacomponentatlayerL;isallowedtocallcomponentsattheunderlying
layerLi:«,butnottheotherwayaround,asshowninFig(a).
•Ankeyobservationisthatcontrolgenerallyflowsfromlayertolayer:requestsgodown
thehierarchywhereastheresultsflowupward.
•Thelayeredarchitectureseparatesthelayersofcomponentsfromeachother,givingita
muchmodularapproach.
•AwellknownexampleforthisistheOSIreferencemodelthatincorporatesthelayered
architecturewheninteractingwitheachofthecomponents.
•Theadvantagesofusingthisapproachisthatthecallalwaysfollowpre-definedpathand
thateachlayercanbeeasilyreplacedormodifiedwithoutaffectingtheentirearchitecture.
9/27/2020
BCA (T.U.) Distributed System
By: Er. Nabin Adhikari
5

6. 9/27/2020
BCA (T.U.) Distributed System
By: Er. Nabin Adhikari
6

7. •Object-basedarchitectures,areillustratedinFig.(b).
•Thisarchitecturedoesnothaveasequentialsetofstepsasinlayered.
•Eachcomponentsarereferredasobject,whereeachobjectscaninteract
withotherobjectthroughagivenconnectororinterface.
•Thesearemuchmoredirectwhereallthecomponentscaninteractwith
eachotherthroughadirectcallmethod.
•Notsurprisingly,thissoftwarearchitecturematchestheclient-serversystem
architecture.
•Thelayeredandobjectbasedarchitecturesstillformthemostimportant
stylesforlargesoftwaresystems(BassetaI.,2003).
9/27/2020
BCA (T.U.) Distributed System
By: Er. Nabin Adhikari
7

8. •Some popular examples are java
RMI(Remote method invocation)
, web Services and REST
API Calls. This things has
following properties.
❖The architecture style is less
structured.
❖Components=Object
❖Connector=RPC or RMI
9/27/2020
BCA (T.U.) Distributed System
By: Er. Nabin Adhikari
8

9. •Data-centeredarchitecturesevolvearoundtheideathatprocessescommunicatethrough
acommon(passiveoractive)repository.
•Itcanbearguedthatfordistributedsystemsthesearchitecturesareasimportantasthe
layeredandobjectbasedarchitectures.
•Thisismorelikeproducerconsumerproblems.The
•Forexample,awealthofnetworkedapplicationshavebeendevelopedthatrelyona
shareddistributedfilesysteminwhichvirtuallyallcommunicationtakesplacethrough
files.Likewise,Web-baseddistributedsystems,arelargelydata-centric:processes
communicatethroughtheuseofsharedWeb-baseddataservices.
9/27/2020
BCA (T.U.) Distributed System
By: Er. Nabin Adhikari
9

10. 9/27/2020
BCA (T.U.) Distributed System
By: Er. Nabin Adhikari
10

11. •Inevent-basedarchitectures,processesessentiallycommunicate
throughthepropagationofevents,whichoptionallyalsocarrydata.
•Whenaneventisgenerated,itwillbesentthroughthebusand
respectivenodecanpulltheeventanduse.
•SometimethiseventcouldbedataorevenURLtoresource.
•Itsadvantagesisthatthecomponentscanbeeasilyadd,removeor
modified.Theheterogenouscomponentscancommunicatethrough
anycommunicationprotocol.
•Someexamplesare:publisher-subscribesystem,broadcast,pointto
point,enterpriseservicebus(ESB),etc.
9/27/2020
BCA (T.U.) Distributed System
By: Er. Nabin Adhikari
11

12. 9/27/2020
BCA (T.U.) Distributed System
By: Er. Nabin Adhikari
12

13. 2.2 System Architecture
•Nowthatwehavebrieflydiscussedsomecommon
architecturalstyles,letustakealookathowmanydistributed
systemsareactuallyorganizedbyconsideringwheresoftware
componentsareplaced.
•Decidingonsoftwarecomponents,theirinteraction,andtheir
placementleadsaninstanceofasoftwarearchitecture,also
calledasystemarchitecture(BassetaI.,2003).
9/27/2020
BCA (T.U.) Distributed System
By: Er. Nabin Adhikari
13

14. 2.2.1 Client server architecture
•Theclientserverarchitecturehastwomajorcomponents-clientand
server.
•Theserveriswherealltheprocessing,computinganddatahandlingis
happeningandtheclientiswheretheusercanaccesstheservice
andresourcesgivenbyserver.
•Theclientmakerequestandtheserverreply.
•Asonecommondesignfeature,thisarchitecturehascentralized
securitydatabase.Thisdatabasecontainsthesecuritydetails,access
details.
9/27/2020
BCA (T.U.) Distributed System
By: Er. Nabin Adhikari
14

15. AdvantagesOfClient/Servernetwork:
•AClient/Servernetworkcontainsthecentralized
system.Thereforewecanbackupthedataeasily.
•AClient/Servernetworkhasadedicatedserverthat
improvestheoverallperformanceofthewholesystem.
•SecurityisbetterinClient/Servernetworkasasingle
serveradministersthesharedresources.
•Italsoincreasesthespeedofthesharingresources.
DisadvantagesOfClient/Servernetwork:
•Client/Servernetworkisexpensiveasitrequiresthe
serverwithlargememory.
•AserverhasaNetworkOperatingSystem(NOS) to
providetheresourcestotheclients,butthecostofNOS
isveryhigh.
•Itrequiresadedicatednetworkadministratorto
managealltheresources.
•Ifservercomputerisdownthenwholenetworkwill
affected.
9/27/2020
BCA (T.U.) Distributed System
By: Er. Nabin Adhikari
15

16. 2.2.2 Peer to peer network architecture
•Peer-To-Peernetworkisanetworkinwhichallthe
computersarelinkedtogetherwithequalprivilegeand
responsibilitiesforprocessingthedata.
•Peer-To-Peernetworkisusefulforsmallenvironments,
usuallyupto10computers.
•Peer-To-Peernetworkhasnodedicatedserver.
•Specialpermissionsareassignedtoeachcomputerfor
sharingtheresources,butthiscanleadtoaproblemif
thecomputerwiththeresourceisdown.
9/27/2020
BCA (T.U.) Distributed System
By: Er. Nabin Adhikari
16

17. AdvantagesOfPeer-To-PeerNetwork:
•Ithassimplearchitectureandeasytoinstall.
•Itislesscostlyasitdoesnotcontainanydedicated
server.
•Ifissuitableforsmallsizednetwork.
•Itiseasytosetupandmaintainaseachcomputer
managesitself.
DisadvantagesOfPeer-To-PeerNetwork:
•InthecaseofPeer-To-Peernetwork,itdoesnotcontain
thecentralizedsystem.Therefore,itcannotbackupthe
dataasthedataisdifferentindifferentlocations.
•Ithasasecurityissueasthedeviceismanageditself.
•Itisnotsuitableforlargesizenetwork.
9/27/2020
BCA (T.U.) Distributed System
By: Er. Nabin Adhikari
17

18. 9/27/2020
BCA (T.U.) Distributed System
By: Er. Nabin Adhikari
18

19. 2.2.3 Centralized Architecture
•Despitethelackofconsensusonmanydistributedsystemsissues,thereisone
issuethatmanyresearchersandpractitionersagreeupon:thinkingintermsof
clientsthatrequestservicesfromservershelpsusunderstandandmanagethe
complexityofdistributedsystemsandthatisagoodthing.
•Inthebasicclient-servermodel,processesinadistributedsystemaredividedinto
two(possiblyoverlapping)groups.
•Aserverisaprocessimplementingaspecificservice,forexample,afilesystem
serviceoradatabaseservice.
•Aclientisaprocessthatrequestsaservicefromaserverbysendingitarequest
andsubsequentlywaitingfortheserver'sreply.
•Thisclient-serverinteraction,alsoknownasrequest-replybehaviorisshownin
Fig.
9/27/2020
BCA (T.U.) Distributed System
By: Er. Nabin Adhikari
19

20. Two tier architecture
9/27/2020
BCA (T.U.) Distributed System
By: Er. Nabin Adhikari
20

21. •Communicationbetweenaclientandaservercanbeimplementedbymeansofasimple
connectionlessprotocolwhentheunderlyingnetworkisfairlyreliableasinmanylocal-
areanetworks.
•Inthesecases,whenaclientrequestsaservice,itsimplypackagesamessageforthe
server,identifyingtheserviceitwants,alongwiththenecessaryinputdata.Themessage
isthensenttotheserver.
•Thelatter,inturn,willalwayswaitforanincomingrequest,subsequentlyprocessit,and
packagetheresultsinareplymessagethatisthensenttotheclient.
•Usingaconnectionlessprotocolhastheobviousadvantageofbeingefficient.
•Aslongasmessagesdonotgetlostorcorrupted,therequest/replyprotocoljustsketched
worksfine.
•Unfortunately,makingtheprotocolresistanttooccasionaltransmissionfailuresisnot
trivial.Theonlythingwecandoispossiblylettheclientresendtherequestwhenno
replymessagecomesin.
9/27/2020
BCA (T.U.) Distributed System
By: Er. Nabin Adhikari
21

22. •However,consideringthatmanyclient-serverapplicationsaretargetedtoward
supportinguseraccesstodatabases,manypeoplehaveadvocatedadistinction
betweenthefollowingthreelevels,essentiallyfollowingthelayeredarchitectural
stylewediscussedpreviously:
1.Theuser-interfacelevel
2.Theprocessinglevel
3.Thedatalevel
•Theuser-interfacelevelcontainsallthatisnecessarytodirectlyinterfacewiththe
user,suchasdisplaymanagement.
•Theprocessingleveltypicallycontainstheapplications.
•Thedatalevelmanagestheactualdatathatisbeingactedon.
9/27/2020
BCA (T.U.) Distributed System
By: Er. Nabin Adhikari
22

23. •Asafirstexample,consideranInternetsearchengine.Ignoringalltheanimated
banners,images,andotherfancywindowdressing,theuserinterfaceofasearch
engineisverysimple:ausertypesinastringofkeywordsandissubsequently
presentedwithalistoftitlesofWebpages.
•ThebackendisformedbyahugedatabaseofWebpagesthathavebeen
prefetchedandindexed.Thecoreofthesearchengineisaprogramthattransforms
theuser'sstringofkeywordsintooneormoredatabasequeries.
•Itsubsequentlyrankstheresultsintoalist,andtransformsthatlistintoaseriesof
HTMLpages.
•Withintheclient-servermodel,thisinformationretrievalpartistypicallyplacedat
theprocessinglevel.Fig.belowshowsthisorganization.
9/27/2020
BCA (T.U.) Distributed System
By: Er. Nabin Adhikari
23

24. Three tier
9/27/2020
BCA (T.U.) Distributed System
By: Er. Nabin Adhikari
24

25. •Thedistinctionintothreelogicallevelsasdiscussedsofar,suggestsanumberof
possibilitiesforphysicallydistributingaclient-serverapplicationacrossseveral
machines.Thesimplestorganizationistohaveonlytwotypesofmachines:
1.Aclientmachinecontainingonlytheprogramsimplementing(partof)theuser-
interfacelevel
2.Aservermachinecontainingtherest,thatistheprogramsimplementingthe
processinganddatalevel
•Inthisorganizationeverythingishandledbytheserverwhiletheclientis
essentiallynomorethanadumbterminal,possiblywithaprettygraphical
interface.Therearemanyotherpossibilities,ofwhichweexploresomeofthe
morecommononesinthissection.
9/27/2020
BCA (T.U.) Distributed System
By: Er. Nabin Adhikari
25

26. 9/27/2020
BCA (T.U.) Distributed System
By: Er. Nabin Adhikari
26

27. •Onepossibleorganizationistohaveonlytheterminal-dependentpartoftheuserinterface
ontheclientmachine,asshowninFig.2-5(a),andgivetheapplicationsremotecontrol
overthepresentationoftheirdata.
•Analternativeistoplacetheentireuser-interfacesoftwareontheclientside,asshownin
Fig.2-5(b).Insuchcases,weessentiallydividetheapplicationintoagraphicalfrontend,
whichcommunicateswiththerestoftheapplication(residingattheserver)throughan
application-specificprotocol.Inthismodel,thefrontend(theclientsoftware)doesno
processingotherthannecessaryforpresentingtheapplication'sinterface.
•Continuingalongthislineofreasoning,wemayalsomovepartoftheapplicationtothe
frontend,asshowninFig.2-5(c).Anexamplewherethismakessenseiswherethe
applicationmakesuseofaformthatneedstobefilledinentirelybeforeitcanbe
processed.Thefrontendcanthencheckthecorrectnessandconsistencyoftheform,and
wherenecessaryinteractwiththeuser.
9/27/2020
BCA (T.U.) Distributed System
By: Er. Nabin Adhikari
27

28. •AnotherexampleoftheorganizationofFig.2:.5(c),isthatofawordprocessorin
whichthebasiceditingfunctionsexecuteontheclientsidewheretheyoperateon
locallycached,orin-memorydata.butwheretheadvancedsupporttoolssuchas
checkingthespellingandgrammarexecuteontheserverside.
•Inmanyclient-serverenvironments,theorganizationsshowninFig.2-5(d)andFig.
2-5(e)areparticularlypopular.Theseorganizationsareusedwherethefilesystemor
database.Essentially,mostoftheapplicationisrunningontheclientmachine,butall
operationsonfilesordatabaseentriesgototheserver.Forexample,manybanking
applicationsrunonanend-user'smachinewheretheuserpreparestransactionsand
such.Oncefinished,theapplicationcontactsthedatabaseonthebank'sserverand
uploadsthetransactionsforfurtherprocessing.
•Fig.2-5(e)representsthesituationwheretheclient'slocaldiskcontainspartofthe
data.Forexample,whenbrowsingtheWeb,aclientcangraduallybuildahugecache
onlocaldiskofmostrecentinspectedWebpages.
9/27/2020
BCA (T.U.) Distributed System
By: Er. Nabin Adhikari
28

29. 2.2.4 Decentralized Architecture
•Inmodernarchitectures(HorizontalDistribution)aclientorserver
maybephysicallysplitupintologicallyequivalentparts,buteachpart
isoperatingonitsownshareofthecompletedataset,thusbalancing
theload.
•Aclassofmodernsystemarchitecturesthatsupporthorizontal
distribution,knownaspeer-to-peersystems
9/27/2020
BCA (T.U.) Distributed System
By: Er. Nabin Adhikari
29

30. •Fromahigh-levelperspective,theprocessesthatconstituteapeer-to-peer
systemareallequal.Thismeansthatthefunctionsthatneedtobecarriedout
arerepresentedbyeveryprocessthatconstitutesthedistributedsystem.
•Asaconsequence,muchoftheinteractionbetweenprocessesissymmetric:each
processwillactasaclientandaserveratthesametime(whichisalsoreferredto
asactingasaservant).
•Giventhissymmetricbehavior,peer-to-peerarchitecturesevolvearoundthe
questionhowtoorganizetheprocessesinanoverlaynetwork,thatis,anetwork
inwhichthenodesareformedbytheprocessesandthelinksrepresentthe
possiblecommunicationchannels(whichareusuallyrealizedasTCPconnections).
•Ingeneral,aprocesscannotcommunicatedirectlywithanarbitraryother
process,butisrequiredtosendmessagesthroughtheavailablecommunication
channels.
•Twotypesofoverlaynetworksexist:thosethatarestructuredandunstructured.
9/27/2020
BCA (T.U.) Distributed System
By: Er. Nabin Adhikari
30

31. Structured p2p system
•The node has to follow the pre-defined structure.
•Everystructurednetworkinherentlysuffersfromthepoorscalabilitydueto
needforstructuremaintenance.
•Ingeneral,thenodesinastructuredoverlaynetworkareformedina
logicalringwithnodesbeingconnectedtothisring.Inthisring,thecertain
nodesareresponsibleforcertainservices.
•Acommonapproachthatcanbeusedtotacklethecoordinationbetween
nodes,istousedistributedhashtable(DHT).Atraditionalhashfunction
convertsauniquekeyintoahashvalues,thatwillrepresentanobjectin
thenetwork.
•Thehashfunctionvalueisusedtoinsertanobjectinthehashtableandto
retrieveit.
9/27/2020
BCA (T.U.) Distributed System
By: Er. Nabin Adhikari
31

32. •InaDHT,eachkeyisassignedtoa
uniquewheretherandomhasvalue
needtobeofaverylargeaddress
space,inordertoensureuniqueness.
•Hashfunction:takesakeyand
produceauniquehashfunction.
•Mappingfunction:mapthehashvalue
toaspecificnodeinthesystem.
•Lookuptable:returnthenetwork
addressofthenoderepresentedby
theuniquehashvalue.
9/27/2020
BCA (T.U.) Distributed System
By: Er. Nabin Adhikari
32

33. Unstructured p2p system
•Thereisnospecificstructure.
•Thescalabilityisveryhigh.
•Thesesystemrelyonrandomizedalgorithmsforconstructinganoverlay
network.
•therenofixedpathfornodetonodetransmission.
•Thebasicprincipleisthateachnodeisrequiredtorandomlyselectanother
nodeandcontactit.
❖Leteachpeermaintainapartialviewofthenetwork,consistingof‘n’
othernodes.
❖Eachnode‘p’periodicallyselectanode‘q’fromitspartialview.
❖Pandqexchangeinformationandexchangemembersfromtheir
respectivepartialviews.
9/27/2020
BCA (T.U.) Distributed System
By: Er. Nabin Adhikari
33

34. 2.2.5 Hybrid architecture
•Animportantclassofdistributedsystemsthatisorganizedaccordingtoahybrid
architectureisformedbyedge-serversystems.
•ThesesystemsaredeployedontheInternetwhereserversareplaced"attheedge"of
thenetwork.
•Thisedgeisformedbytheboundarybetweenenterprisenetworksandtheactual
Internet,forexample,asprovidedbyanInternetServiceProvider(ISP).
•Likewise,whereendusersathomeconnecttotheInternetthroughtheirISP,theISPcan
beconsideredasresidingattheedgeoftheInternet.
•Endusers,orclientsingeneral,connecttotheInternetbymeansofanedgeserver.The
edgeserver'smainpurposeistoservecontent,possiblyafterapplyingfilteringand
transcodingfunctions.
•Moreinterestingisthefactthatacollectionofedgeserverscanbeusedtooptimize
contentandapplicationdistribution.
•Thebasicmodelisthatforaspecificorganization,oneedgeserveractsasanorigin
serverfromwhichallcontentoriginates.Thatservercanuseotheredgeserversfor
replicatingWebpages
9/27/2020
BCA (T.U.) Distributed System
By: Er. Nabin Adhikari
34

35. Viewing the internet as consisting of a
collection of edge server
9/27/2020
BCA (T.U.) Distributed System
By: Er. Nabin Adhikari
35

36. Collaborative distributed systems
•Hybridstructuresarenotablydeployedincollaborativedistributedsystems.
•Themainissueinmanyofthesesystemsistofirstgetstarted,forwhichoftena
traditionalclient-serverschemeisdeployed.
•Onceanodehasjoinedthesystem,itcanuseafullydecentralizedschemefor
collaboration.
•BitTorrentisapeer-to-peerfiledownloadingsystem.
•ItsprincipalworkingisshowninFig.2-14.Thebasicideaisthatwhenanenduser
islookingforafile,hedownloadschunksofthefilefromotherusersuntilthe
downloadedchunkscanbeassembledtogetheryieldingthecompletefile.
•Animportantdesigngoalwastoensurecollaboration.Inmostfile-sharing
systems,asignificantfractionofparticipantsismerelydownloadfilesbut
otherwisecontributeclosetonothing.Tothisend,afilecanbedownloadedonly
whenthedownloadingclientisprovidingcontenttosomeoneelse.
9/27/2020
BCA (T.U.) Distributed System
By: Er. Nabin Adhikari
36

37. 9/27/2020
BCA (T.U.) Distributed System
By: Er. Nabin Adhikari
37

38. •Asanotherexample,considertheGlobulecollaborativeContent
DistributionNetwork(CDN).
•Globulestronglyresemblestheedge-serverarchitecturementioned
above.Inthiscase,insteadofedgeservers,endusers(butalso
organizations)voluntarilyprovideenhancedWebserversthatare
capableofcollaboratinginthereplicationofWebpages.
•Initssimplestform,eachsuchserverhasthefollowingcomponents:
•Acomponentthatcanredirectclientrequeststootherservers.
•Acomponentforanalyzingaccesspatterns.
•AcomponentformanagingthereplicationofWebpages.
9/27/2020
BCA (T.U.) Distributed System
By: Er. Nabin Adhikari
38

39. 2.3 Middleware Organizations
•Middlewareoffersgeneralservicesthatsupportdistributedexecutionof
applications.Thetermmiddlewaresuggeststhatitissoftwarepositioned
betweentheoperatingsystemandtheapplication.
•Wewilllimitthediscussiontothemiddlewaresupportofanobject-based
applicationbecausetheconceptsofanobjectmodelideallyreflectthe
characteristicsofdistributedsystems.
•Anobjectencapsulatesstateandbehaviorandcanonlybeaccessedviaawell-
definedinterface.
•Theinterfacehidesthedetailsthatarespecifictotheimplementation,thereby
helpingtoencapsulatedifferenttechnologies.
•Anobjectthereforebecomesaunitofdistribution.Recallthatobjects
communicatewitheachotherbyexchangingmessages.wedefinethefollowing
middlewaretasks:
9/27/2020
BCA (T.U.) Distributed System
By: Er. Nabin Adhikari
39

40. Tasksofmiddleware
❖Objectmodelsupport:Middlewareshouldoffermechanismstosupport
theconceptsincorporatedintheobjectmodel.
❖Operationalinteraction:Middlewareshouldallowtheoperational
interactionbetweentwoobjects.Themodelusedisthemethodinvocation
ofanobjectorientedprogramminglanguage.
❖Remoteinteraction:Middlewareshouldallowtheinteractionbetweentwo
objectslocatedindifferentaddressspaces.
❖Distributiontransparency:Fromthestandpointoftheprogram,interaction
betweenobjectsisidenticalforbothlocalandremoteinteractions.
❖Technologicalindependence:Themiddlewaresupportstheintegrationof
differenttechnologies.
9/27/2020
BCA (T.U.) Distributed System
By: Er. Nabin Adhikari
40

41. •Aswementionedearlier,middlewareisconceptuallylocatedbetweentheapplication
andtheoperatingsystem(seeFigure2.7).
•Becauseanobjectmodelservesastheunderlyingparadigm,theapplicationis
representedasasetofinteractingobjects.
•Eachobjectisexplicitlyallocatedtoahardwareplatform(i.e.,wedonotconsidercases
inwhichanobjectlogicallyextendsbeyondcomputerboundaries).
•Typesofthemiddlewarehidestheheterogeneitythatoccursinadistributedsystem.This
heterogeneityexistsatdifferentplaces:
❖Programminglanguages:Differentobjectscanbedevelopedindifferentprogramming
languages.
❖Operatingsystem:Operatingsystemshavedifferentcharacteristicsandcapabilities.
❖Computerarchitectures:Computersdifferintheirtechnicaldetails(e.g.,data
representations).
9/27/2020
BCA (T.U.) Distributed System
By: Er. Nabin Adhikari
41

42. 9/27/2020
BCA (T.U.) Distributed System
By: Er. Nabin Adhikari
42

43. •Middlewareovercomesthisheterogeneitybyofferingequal
functionalityatallaccesspoints.
•Applicationshaveaccesstothisfunctionalitythroughan
ApplicationProgrammingInterface(API).
•BecausetheAPIdependsontheprogramminglanguagein
whichtheapplicationiswritten,theAPIhastobeadaptedto
theconditionsofeachprogramminglanguagethatis
supportedbythemiddleware
9/27/2020
BCA (T.U.) Distributed System
By: Er. Nabin Adhikari
43

44. Architecture vs Middleware
•Whenconsideringthearchitecturalissueswehavediscussedsofar,aquestionthat
comestomindiswheremiddlewarefitsin,middlewareformsalayerbetween
applicationsanddistributedplatforms.
•Animportantpurposeistoprovideadegreeofdistributiontransparency,thatis,toa
certainextenthidingthedistributionofdata,processing,andcontrolfromapplications.
•Whatiscommonlyseeninpracticeisthatmiddlewaresystemsactuallyfollowaspecific
architecturalstyle.
•Forexample,manymiddlewaresolutionshaveadoptedanobject-basedarchitectural
style,suchasCORBA.Others,likeTIBCORendezvousprovidemiddlewarethatfollows
theevent-basedarchitecturalstyle.
•Advantage:Havingmiddlewaremoldedaccordingtoaspecificarchitecturalstylehasthe
benefitthatdesigningapplicationsmaybecomesimpler.
•Disadvantage:Themiddlewaremaynolongerbeoptimalforwhatanapplication
developerhadinmind.
9/27/2020
BCA (T.U.) Distributed System
By: Er. Nabin Adhikari
44

45. •Althoughmiddlewareismeanttoprovidedistributiontransparency,itis
generallyfeltthatspecificsolutionsshouldbeadaptabletoapplication
requirements.
•Onesolutiontothisproblemistomakeseveralversionsofamiddleware
system,whereeachversionistailoredtoaspecificclassofapplications.
•Anotherapproachthatisgenerallyconsideredbetteristomake
middlewaresystemssuchthattheyareeasytoconfigure,adapt,and
customizeasneededbyanapplication.
•Asaresult,systemsarenowbeingdevelopedinwhichastricterseparation
betweenpoliciesandmechanismsisbeingmade.
•Thishasledtoseveralmechanismsbywhichthebehaviorofmiddleware
canbemodified(SadjadiandMcKinley,2003).UsingInterceptorsand
Adaptivesoftwarearetwopoliciesinthesecondsolution.
9/27/2020
BCA (T.U.) Distributed System
By: Er. Nabin Adhikari
45

46. Interceptor
•Conceptually,aninterceptorisnothingbutasoftwareconstructthatwillbreaktheusual
flowofcontrolandallowother(applicationspecific)codetobeexecuted.
•Tomakemattersconcrete,considerinterceptionassupportedinmanyobjectbased
distributedsystems.
•Thebasicideaissimple:anobjectAcancallamethodthatbelongstoanobjectB,while
thelatterresidesonadifferentmachinethanA.sucharemote-objectinvocationis
carriedasathree-stepapproach:
1.ObjectAisofferedalocalinterfacethatisexactlythesameastheinterfaceofferedby
objectB.Asimplycallsthemethodavailableinthatinterface.
2.ThecallbyAistransformedintoagenericobjectinvocation,madepossiblethrougha
generalobject-invocationinterfaceofferedbythemiddlewareatthemachinewhereA
resides.
3.Finally,thegenericobjectinvocationistransformedintoamessagethatissent
throughthetransport-levelnetworkinterfaceasofferedbyA'slocaloperatingsystem.
9/27/2020
BCA (T.U.) Distributed System
By: Er. Nabin Adhikari
46

47. 9/27/2020
BCA (T.U.) Distributed System
By: Er. Nabin Adhikari
47

48. General approach to adaptive software
•Whatinterceptorsactuallyofferisameanstoadaptthemiddleware.Theneedfor
adaptationcomesfromthefactthattheenvironmentinwhichdistributedapplicationsare
executedchangescontinuously.
•Changesincludethoseresultingfrommobility,astrongvarianceinthequality-of-service
ofnetworks,failinghardware,andbatterydrainage,amongstothers.
•Thesestronginfluencesfromtheenvironmenthavebroughtmanydesignersof
middlewaretoconsidertheconstructionofadaptivesoftware.
•Threebasictechniquestocometosoftwareadaptation:
•1.Separationofconcerns
•2.Computationalreflection
•3.Component-baseddesign
9/27/2020
BCA (T.U.) Distributed System
By: Er. Nabin Adhikari
48

49. •Separatingconcernsrelatestothetraditionalwayofmodularizingsystems:separatethe
partsthatimplementfunctionalityfromthosethattakecareofotherthings(knownas
extrafunctionalities)suchasreliability,performance,security,etc.Onecanarguethat
developingmiddlewarefordistributedapplicationsislargelyabouthandlingextra
functionalitiesindependentfromapplications.
•Computationalreflectionreferstotheabilityofaprogramtoinspectitselfand,if
necessary,adaptitsbehavior.Reflectionhasbeenbuiltintoprogramminglanguages,
includingJava,andoffersapowerfulfacilityforruntimemodifications.
•Component-baseddesignsupportsadaptationthroughcomposition.Asystemmayeither
beconfiguredstaticallyatdesigntime,ordynamicallyatruntime.Thelatterrequires
supportforlatebinding,atechniquethathasbeensuccessfullyappliedinprogramming
languageenvironments,butalsoforoperatingsystemswheremodulescanbeloadedand
unloadedatwill.
9/27/2020
BCA (T.U.) Distributed System
By: Er. Nabin Adhikari
49

50. 2.4 Self management in distributed system
The Feedback Control Model
•Therearemanydifferentviewsonself-managingsystems,butwhatmost
haveincommon(eitherexplicitlyorimplicitly)istheassumptionthat
adaptationstakeplacebymeansofoneormorefeedbackcontrolloops.
•Accordingly,systemsthatareorganizedbymeansofsuchloopsare
referredtoasfeedbackControlsystems.
•Feedbackcontrolhassincelongbeenappliedinvariousengineering
fields,anditsmathematicalfoundationsaregraduallyalsofindingtheir
wayincomputingsystems.
•Forselfmanagingsystems,thearchitecturalissuesareinitiallythemost
interesting.Thebasicideabehindthisorganizationisquitesimple,as
showninFig.
9/27/2020
BCA (T.U.) Distributed System
By: Er. Nabin Adhikari
50

51. 9/27/2020
BCA (T.U.) Distributed System
By: Er. Nabin Adhikari
51

52. •Thecoreofafeedbackcontrolsystemisformedbythecomponents
thatneedtobemanaged.
•Thesecomponentsareassumedtobedriventhroughcontrollableinput
parameters,buttheirbehaviormaybeinfluencedbyallkindsof
uncontrollableinput,alsoknownasdisturbanceornoiseinput.
•Althoughdisturbancewilloftencomefromtheenvironmentinwhich
adistributedsystemisexecuting,itmaywellbethecasethat
unanticipatedcomponentinteractioncausesunexpectedbehavior.
9/27/2020
BCA (T.U.) Distributed System
By: Er. Nabin Adhikari
52

53. There are essentially three elements that form the feedback
control loop.
•First,thesystemitselfneedstobemonitored,whichrequiresthatvarious
aspectsofthesystemneedtobemeasured.
•Inmanycases,measuringbehavioriseasiersaidthandone.
•Forexample,round-tripdelaysintheInternetmayvarywildly,andalso
dependonwhatexactlyisbeingmeasured.
•Insuchcases,accuratelyestimatingadelaymaybedifficultindeed.
•MattersarefurthercomplicatedwhenanodeAneedstoestimatethelatency
betweentwoothercompletelydifferentnodesBandC,withoutbeingable
tointrudeoneithertwonodes.
•Forreasonsasthis,afeedbackcontrolloopgenerallycontainsalogical
metricestimationcomponent.
9/27/2020
BCA (T.U.) Distributed System
By: Er. Nabin Adhikari
53

54. •Anotherpartofthefeedbackcontrolloopanalyzesthemeasurementsand
comparesthesetoreferencevalues.
•Thisfeedbackanalysiscomponentformstheheartofthecontrolloop,asitwill
containthealgorithmsthatdecideonpossibleadaptations.
•Thelastgroupofcomponentsconsistofvariousmechanismstodirectlyinfluence
thebehaviorofthesystem.
•Therecanbemanydifferentmechanisms:placingreplicas,changingscheduling
priorities,switchingservices,movingdataforreasonsofavailability,redirecting
requeststodifferentservers,etc.
•Theanalysiscomponentwillneedtobeawareofthesemechanismsandtheir
(expected)effectonsystembehavior.Therefore,itwilltriggeroneorseveral
mechanisms,tosubsequentlylaterobservetheeffect.
9/27/2020
BCA (T.U.) Distributed System
By: Er. Nabin Adhikari
54

55. •Aninterestingobservationisthatthefeedbackcontrolloopalsofitsthe
manualmanagementofsystems.Themaindifferenceisthattheanalysis
componentisreplacedbyhumanadministrators.
•However,inordertoproperlymanageanydistributedsystem,these
administratorswillneeddecentmonitoringequipmentaswellasdecent
mechanismstocontrolthebehaviorofthesystem.
•Itshouldbeclearthatproperlyanalyzingmeasureddataandtriggeringthe
correctactionsmakesthedevelopmentofself-managingsystemsso
difficult.
9/27/2020
BCA (T.U.) Distributed System
By: Er. Nabin Adhikari
55

56. Example: System monitoring with Astrolabe
•weconsiderAstrolabe(VanRenesseetaI.,2003),whichisasystemthatcansupport
generalmonitoringofverylargedistributedsystems.
•Inthecontextofself-managingsystems,Astrolabeistobepositionedasageneraltoolfor
observingsystemsbehavior.Itsoutputcanbeusedtofeedintoananalysiscomponentfor
decidingoncorrectiveactions.
•Astrolabeorganizesalargecollectionofhostsintoahierarchyofzones.
•Thelowest-levelzonesconsistofjustasinglehost,whicharesubsequentlygroupedinto
zonesofincreasingsize.
•Thetop-levelzonecoversallhosts.EveryhostrunsanAstrolabeprocess,calledanagent,
thatcollectsinformationonthezonesinwhichthathostiscontained.
•Theagentalsocommunicateswithotheragentswiththeaimtospreadzoneinformation
acrosstheentiresystem.
9/27/2020
BCA (T.U.) Distributed System
By: Er. Nabin Adhikari
56

57. •Eachhostmaintainsasetofattributesforcollectinglocalinformation.Forexample,a
hostmaykeeptrackofspecificfilesitstores,itsresourceusage,andsoon.
•Onlytheattributesasmaintaineddirectlybyhosts,thatis,atthelowestlevelofthe
hierarchyarewritable.
•Eachzonecanalsohaveacollectionofattributes,butthevaluesoftheseattributesare
computedfromthevaluesoflowerlevelzones.
•ConsiderthefollowingsimpleexampleshowninFig.withthreehosts,A,B,andC
groupedintoazone.EachmachinekeepstrackofitsIPaddress,CPUload,availablefree
memory.andthenumberofactiveprocesses.
•Eachoftheseattributescanbedirectlywrittenusinglocalinformationfromeachhost.At
thezonelevel,onlyaggregatedinformationcanbecollected,suchastheaverageCPU
load,ortheaveragenumberofactiveprocesses.
9/27/2020
BCA (T.U.) Distributed System
By: Er. Nabin Adhikari
57

58. 9/27/2020
BCA (T.U.) Distributed System
By: Er. Nabin Adhikari
58