PROJECTONCOMPUTERNETWORKING

1.Abstract:
Thisprojectconsistsofthreetasksintotal.Thefirsttaskintendstoderivethewayforgreateruseof
cloudinfrastructureandapplicationwithfilestoragefacility,replacingoldmachineswithold
operatingsystemswiththinclientsandincreasingtheavailabilityofinternetbythecompanyby
bifurcatingthemainIPaddressintotwoseparateIPaddresses,notonlyatthecompanylevelbutalso
atthedepartmentlevel.ThesecondtaskintendstoprovidesuitableSaaS,PaaSandIaaSsolutionsfor
thedepartmentlevelbifurcationsofIPwithimpositionofthinclients.Thethirdtaskintendsto
provideasubnetstructureofthecompanywithsuitablenetworktopologymentioningpropersubnet
address,subnetmaskandbroadcastaddress.Thisapart,thisreportalsoexplainsfunctionsoflayer2
and3ofOSImodel,theirmappingtoTCP/IPmodel,usageoftheseprotocolsbythecompany,
differencebetweenlevel2switches,level3switchesandtherouters,usageofVPNforfacilitating
companysfunctionalityingivendisposition.(Threepackettracerfileswith.PKTfileextensionare
attachedherewithforeachofthethreetasksintendedtopresentoverhere).
2.Introduction:
Asaresultofrapidtechnologicalprogress,theseareasarerapidlyconverginginthe21stcenturyand
thedifferencesbetweencollecting,transporting,storing,andprocessinginformationarequickly
disappearing.Organizationswithhundredsofofficesspreadoverawidegeographicalarearoutinely
expecttobeabletoexaminethecurrentstatusofeventheirmostremoteoutpostatthepushofa
button.Asourabilitytogather,process,anddistributeinformationgrows,thedemandforevermore
sophisticatedinformationprocessinggrowsevenfaster.
Althoughthecomputerindustryisstillyoungcomparedtootherindustries(e.g.,automobilesandair
transportation),computershavemadespectacularprogressinashorttime.Duringthefirsttwo
decadesoftheirexistence,computersystemswerehighlycentralized,usuallywithinasinglelarge
room.Notinfrequently,thisroomhadglasswalls,throughwhichvisitorscouldgawkatthegreat
electronicwonderinside.Amediumsizedcompanyoruniversitymighthavehadoneortwo
computers,whileverylargeinstitutionshadatmostafewdozen.Theideathatwithinfortyyears
vastlymorepowerfulcomputerssmallerthanpostagestampswouldbemassproducedbythebillions
waspuresciencefiction.
The merging of computers and communications has had a profound influence on the way
computersystemsareorganized.Theoncedominantconceptofthecomputercenterasaroomwith
alargecomputertowhichusersbringtheirworkforprocessingisnowtotallyobsolete(althoughdata
centersholdingthousandsofInternetserversarebecomingcommon).Theoldmodelofasingle
computerservingalloftheorganizationscomputationalneedshasbeenreplacedbyoneinwhicha
largenumberofseparatebutinterconnectedcomputersdothejob.Thesesystemsarecalledcomputer
networks.
Inacomputernetwork,thiscoherence,model,andsoftwareareabsent.Usersareexposedtothe
actualmachines,withoutanyattemptbythesystemtomakethemachineslookandactinacoherent
way.Ifthemachineshavedifferenthardwareanddifferentoperatingsystems,thatisfullyvisibleto
theusers.Ifauserwantstorunaprogramonaremotemachine,hehastologontothatmachineand
runitthere. Ineffect,adistributedsystemisasoftwaresystembuiltontopofanetwork.Thesoftware
givesitahighdegreeofcohesivenessandtransparency.Thus,thedistinctionbetweenanetworkand
adistributedsystemlieswiththesoftware(especiallytheoperatingsystem),ratherthanwiththe
hardware. Nevertheless,thereisconsiderableoverlapbetweenthetwosubjects.Forexample,both
distributedsystemsandcomputernetworksneedtomovefilesaround.Thedifferenceliesinwho
invokesthemovement,thesystemortheuser.

Mostcompanieshaveasubstantialnumberofcomputers.Forexample,acompanymayhavea
computerforeachworkerandusethemtodesignproducts,writebrochures,anddothepayroll.
Initially,someofthesecomputersmayhaveworkedinisolationfromtheothers,butatsomepoint,
managementmayhavedecidedtoconnectthemtobeabletodistributeinformationthroughoutthe
company. Putinslightlymoregeneralform,theissuehereisresourcesharing.Thegoalistomakeall
programs,equipment,andespeciallydataavailabletoanyoneonthenetworkwithoutregardtothe
physicallocationoftheresourceortheuser.Anobviousandwidespreadexampleishavingagroupof
officeworkersshareacommonprinter.Noneoftheindividualsreallyneedsaprivateprinter,anda
highvolumenetworkedprinterisoftencheaper,faster,andeasiertomaintainthanalargecollection
ofindividualprinters. However,probablyevenmoreimportantthansharingphysicalresourcessuch
asprinters,andtapebackupsystems,issharinginformation.Companiessmallandlargearevitally
dependentoncomputerizedinformation.Mostcompanieshavecustomerrecords,product
information,inventories,financialstatements,taxinformation,andmuchmoreonline.Ifallofits
computerssuddenlywentdown,abankcouldnotlastmorethanfiveminutes.Amodern
manufacturingplant,withacomputercontrolledassemblyline,wouldnotlasteven5seconds.Evena
smalltravelagencyorthreepersonlawfirmisnowhighlydependentoncomputernetworksfor
allowingemployeestoaccessrelevantinformationanddocumentsinstantly. Forsmallercompanies,
allthecomputersarelikelytobeinasingleofficeorperhapsasinglebuilding,butforlargerones,the
computersandemployeesmaybescatteredoverdozensofofficesandplantsinmanycountries.
Nevertheless,asalespersoninNewYorkmightsometimesneedaccesstoaproductinventory
databaseinSingapore.NetworkscalledVPNs(VirtualPrivateNetworks)maybeusedtojointhe
individualnetworksatdifferentsitesintooneextendednetwork.Inotherwords,themerefactthata
userhappenstobe15,000kmawayfromhisdatashouldnotpreventhimfromusingthedataas
thoughtheywerelocal.Thisgoalmaybesummarizedbysayingthatitisanattempttoendthe
tyrannyofgeography. Inthesimplestofterms,onecanimagineacompanysinformationsystemas
consistingofoneormoredatabaseswithcompanyinformationandsomenumberofemployeeswho
needtoaccessthemremotely.Inthismodel,thedataarestoredonpowerfulcomputerscalledservers.
Oftenthesearecentrallyhousedandmaintainedbyasystemadministrator.Incontrast,theemployees
havesimplermachines,calledclients,ontheirdesks,withwhichtheyaccessremotedata,for
example,toincludeinspreadsheetstheyareconstructing.(Sometimeswewillrefertothehumanuser
oftheclientmachineastheclient,butitshouldbeclearfromthecontextwhetherwemeanthe
computeroritsuser.)Theclientandservermachinesareconnectedbyanetwork.Notethatwehave
shownthenetworkasasimpleoval,withoutanydetail.Wewillusethisformwhenwemeana
networkinthemostabstractsense.Whenmoredetailisrequired,itwillbeprovided.
3.ProjectTask#1:
Thedatalinklayerusestheservicesofthephysicallayertosendandreceivebitsovercommunication
channels.Ithasanumberoffunctions,including:
1. Providingawelldefinedserviceinterfacetothenetworklayer.
2. Dealingwithtransmissionerrors.
3. Regulatingtheflowofdatasothatslowreceiversarenotswampedbyfastsenders.
Toaccomplishthesegoals,thedatalinklayertakesthepacketsitgetsfromthenetworklayerand
encapsulatesthemintoframesfortransmission.Eachframecontainsaframeheader,apayloadfield
forholdingthepacket,andaframetrailer,asillustratedinFig.Framemanagementformstheheartof
whatthedatalinklayerdoes.Inthefollowingsectionswewillexaminealltheabovementioned
issuesindetail(Tan,K,2006).

Sending machine

Receiving machine

Packet

Packet
Frame

Header

Payload field

Trailer

Header

Payload field

Trailer

Figure1.Packetsandframes

Toprovideservicetothenetworklayer,thedatalinklayermustusetheserviceprovidedtoitby
thephysicallayer.Whatthephysicallayerdoesisacceptarawbitstreamandattempttodeliveritto
thedestination.Ifthechannelisnoisy,asitisformostwirelessandsomewiredlinks,thephysical
layerwilladdsomeredundancytoitssignalstoreducethebiterrorratetoatolerablelevel.However,
thebitstreamreceivedbythedatalinklayerisnotguaranteedtobeerrorfree.Somebitsmayhave
differentvaluesandthenumberofbitsreceivedmaybelessthan,equalto,ormorethanthenumber
ofbitstransmitted.Itisuptothedatalinklayertodetectand,ifnecessary,correcterrors.
Theusualapproachisforthedatalinklayertobreakupthebitstreamintodiscreteframes,
computeashorttokencalledachecksumforeachframe,andincludethechecksumintheframewhen
itistransmitted.(Sanchez,Ruiz,2001))Whenaframearrivesatthedestination,thechecksumis
recomputed.Ifthenewlycomputedchecksumisdifferentfromtheonecontainedintheframe,the
datalinklayerknowsthatanerrorhasoccurredandtakesstepstodealwithit(e.g.,discardingthebad
frameandpossiblyalsosendingbackanerrorreport).
Breakingupthebitstreamintoframesismoredifficultthanitatfirstappears.Agooddesign
mustmakeiteasyforareceivertofindthestartofnewframeswhileusinglittleofthechannel
bandwidth.Wewilllookatfourmethods:
1. Bytecount.
2. Flagbyteswithbytestuffing.
3. Flagbitswithbitstuffing.
4. Physicallayercodingviolations.

One byte

Byte count
5

1 2

Frame 1
5 bytes

6 7

9 8

0 1

Frame 2
5 bytes

3 4

5 6

7 8

Frame 3
8 bytes

0 1 2

Frame 4
8 bytes

(a)
Error
5

1 2

Frame 1

6 7

Frame 2
(Wrong)

0 1

4 5

Now a byte
count

7 8

0 1 2

(b)

Figure2.Abytestream.(a)Withouterrors.(b)Withoneerror.

Thesecondframingmethodgetsaroundtheproblemofresynchronizationafteranerrorbyhaving
eachframestartandendwithspecialbytes.Oftenthesamebyte,calledaflagbyte,isusedasboththe
startingandendingdelimiter.ThisbyteisshowninFig.asFLAG.Twoconsecutiveflagbytes
indicatetheendofoneframeandthestartofthenext.Thus,ifthereceivereverlosessyn
chronizationitcanjustsearchfortwoflagbytestofindtheendofthecurrentframeandthestartof
thenextframe(Leland,W,1994).
However,thereisastillaproblemwehavetosolve.Itmayhappenthattheflagbyteoccursinthe
data,especiallywhenbinarydatasuchasphotographsorsongsarebeingtransmitted.Thissituation
wouldinterferewiththeframing.Onewaytosolvethisproblemistohavethesendersdatalinklayer
insertaspecialescapebyte(ESC)justbeforeeachaccidentalflagbyteinthedata.Thus,aframing
flagbytecanbedistinguishedfromoneinthedatabytheabsenceorpresenceofanescapebyte
beforeit.Thedatalinklayeronthereceivingendremovestheescapebytesbeforegivingthedatato
thenetworklayer.Thistechniqueiscalledbytestuffing. Ofcourse,thenextquestionis:whathappens
ifanescapebyteoccursinthemiddleofthedata?Theansweristhatit,too,isstuffedwithanescape
byte.Atthereceiver,thefirstescapebyteisremoved,leavingthedatabytethatfollowsit(which
mightbeanotherescapebyteortheflagbyte).SomeexamplesareshowninFig.Inallcases,thebyte
sequencedeliveredafterdestuffingisexactlythesameastheoriginalbytesequence.Wecanstill
searchforaframeboundarybylookingfortwoflagbytesinarow,withoutbotheringtoundo
escapes. ThebytestuffingschemedepictedinFig.isaslightsimplificationoftheoneusedinPPP
(PointtoPointProtocol),whichisusedtocarrypacketsovercommunicationslinks(Wetherall,D.J.,
1990).
TosenddataoveranATMnetwork,itneedstobemappedintoasequenceofcells.Thismappingis
donewithanATMadaptationlayerinaprocesscalledsegmentationandreassembly.Several
adaptationlayershavebeendefinedfordifferentservices,rangingfromperiodicvoicesamplesto
packetdata.ThemainoneusedforpacketdataisAAL5(ATMAdaptationLayer5).
AnAAL5frameisshowninFig.Insteadofaheader,ithasatrailerthatgivesthelengthandhasa
4byteCRCforerrordetection.Naturally,theCRCisthesameoneusedforPPPandIEEE802LANs
likeEthernet.WangandCrowcroft(1992)haveshownthatitisstrongenoughtodetectnontraditional
errorssuchascellreordering.Aswellasapayload,theAAL5framehaspadding.Thisroundsoutthe
overalllengthtobeamultipleof48bytessothattheframecanbeevenlydividedintocells.No
addressesareneededontheframeasthevirtualcircuitidentifiercarriedineachcellwillgetittothe
rightdestination(Halperin,D,2010).
Bytes

1 or 2

Variable

0 to 47

PPP protocol

PPP payload

Pad

2
Unused

AAL5 payload

2
Length

4
CRC

AAL5 trailer

Figure3.AAL5framecarryingPPPdata.

NowthatwehavedescribedATM,wehaveonlytodescribehowPPPmakesuseofATMinthe
caseofADSL.ItisdonewithyetanotherstandardcalledPPPoA(PPPoverATM).Thisstandardis
notreallyaprotocol(soitdoesnotappearinFig.)butmoreaspecificationofhowtoworkwithboth
PPPandAAL5frames.ItisdescribedinRFC2364(Grossetal.,1998).

OnlythePPPprotocolandpayloadfieldsareplacedintheAAL5payload,asshowninFig.327.
TheprotocolfieldindicatestotheDSLAMatthefarendwhetherthepayloadisanIPpacketora
packetfromanotherprotocolsuchasLCP.ThefarendknowsthatthecellscontainPPPinformation
becauseanATMvirtualcircuitissetupforthispurpose.
WithintheAAL5frame,PPPframingisnotneededasitwouldservenopurpose;ATMand
AAL5alreadyprovidetheframing.Moreframingwouldbeworthless.ThePPPCRCisalsonot
needed because AAL5 already includes the very same CRC. This error detection mechanism
supplementstheADSLphysicallayercodingofaReedSolomoncodeforerrorcorrectionanda1
byteCRCforthedetectionofanyremainingerrorsnototherwisecaught.Thisschemehasamuch
moresophisticatederrorrecoverymechanismthanwhenpacketsaresentoveraSONETlinebecause
ADSLisamuchnoisierchannel(Held,G,2010).
PersistentandnonpersistentCSMAprotocolsaredefinitelyanimprovementoverALOHAbecause
theyensurethatnostationbeginstotransmitwhilethechannelisbusy.However,iftwostationssense
thechanneltobeidleandbegintransmittingsimultaneously,theirsignalswillstillcollide.Another
improvementisforthestationstoquicklydetectthecollisionandabruptlystoptransmitting,(rather
thanfinishingthem)sincetheyareirretrievablygarbledanyway.Thisstrategysavestimeand
bandwidth(Hoe,J,1996).
Thisprotocol,knownasCSMA/CD(CSMAwithCollisionDetection),isthebasisoftheclassic
EthernetLAN,soitisworthdevotingsometimetolookingatitindetail.Itisimportanttorealizethat
collisiondetectionisananalogprocess.Thestationshardwaremustlistentothechannelwhileitis
transmitting.Ifthesignalitreadsbackisdifferentfromthesignalitisputtingout,itknowsthata
collisionisoccurring.Theimplicationsarethatareceivedsignalmustnotbetinycomparedtothe
transmittedsignal(whichisdifficultforwireless,asreceivedsignalsmaybe1,000,000timesweaker
thantransmittedsignals)andthatthemodulationmustbechosentoallowcollisionstobedetected
(e.g.,acollisionoftwo0voltsignalsmaywellbeimpossibletodetect)(Fuller,V,2006).
CSMA/CD,aswellasmanyotherLANprotocols,usestheconceptualmodelofFig.Atthepoint
markedt0,astationhasfinishedtransmittingitsframe.Anyotherstationhavingaframetosendmay
nowattempttodoso.Iftwoormorestationsdecidetotransmitsimultaneously,therewillbea
collision.Ifastationdetectsacollision,itabortsitstransmission,waitsarandomperiodoftime,and
thentriesagain(assumingthatnootherstationhasstartedtransmitting)(Hu,Y,2001).
Thetaskofthedatalinklayeristoconverttherawbitstreamofferedbythephysicallayerintoa
streamofframesforusebythenetworklayer.Thelinklayercanpresentthisstreamwithvarying
levelsofreliability,rangingfromconnectionless,unacknowledgedservicetoreliable,connection
orientedservice(Gross,G,1998).
Variousframingmethodsareused,includingbytecount,bytestuffing,andbitstuffing.Datalink
protocolscanprovideerrorcontroltodetectorcorrectdamagedframesandtoretransmitlostframes.
Topreventafastsenderfromoverrunningaslowreceiver,thedatalinkprotocolcanalsoprovide
flowcontrol.Theslidingwindowmechanismiswidelyusedtointegrateerrorcontrolandflow
controlinasimpleway.Whenthewindowsizeis1packet,theprotocolisstopandwait(Halperin,
D,2010).
Codesforerrorcorrectionanddetectionaddredundantinformationtomessagesbyusinga
varietyofmathematicaltechniques.ConvolutionalcodesandReedSolomoncodesarewidely
deployedforerrorcorrection,withlowdensityparitycheckcodesincreasinginpopularity.Thecodes
forerrordetectionthatareusedinpracticeincludecyclicredundancychecksandchecksums.All
thesecodescanbeappliedatthelinklayer,aswellasatthephysicallayerandhigherlayers
(Sanchez,Ruiz,1994).

Weexaminedaseriesofprotocolsthatprovideareliablelinklayerusingacknowledgementsand
retransmissions,orARQ(AutomaticRepeatreQuest),undermorerealisticassumptions.Startingfrom
anerrorfreeenvironmentinwhichthereceivercanhandleanyframesenttoit,weintroducedflow
control,followedbyerrorcontrolwithsequencenumbersandthestopandwaitalgorithm.Thenwe
usedtheslidingwindowalgorithmtoallowbidirectionalcommunicationandintroducetheconceptof
piggybacking.Thelasttwoprotocolspipelinethetransmissionofmultipleframestopreventthe
senderfromblockingonalinkwithalongpropagationdelay.Thereceivercaneitherdiscardall
framesotherthanthenextoneinsequence,orbufferoutoforderframesandsendnegativeacknowl
edgementsforgreaterbandwidthefficiency.Theformerstrategyisagobacknprotocol,andthe
latterstrategyisaselectiverepeatprotocol(Hull,B,2006).
TheInternetusesPPPasthemaindatalinkprotocoloverpointtopointlines.Itprovidesa
connectionlessunacknowledgedservice,usingflagbytestodelimitframesandaCRCforerror
detection.Itisusedtocarrypacketsacrossarangeoflinks,includingSONETlinksinwidearea
networksandADSLlinksforthehome(Garfinkel,S,2002).
Insidethenetwork,severaltradeoffsexistbetweenvirtualcircuitsanddatagrams.Onetradeoffis
setuptimeversusaddressparsingtime.Usingvirtualcircuitsrequiresasetupphase,whichtakestime
andconsumesresources.However,oncethispriceispaid,figuringoutwhattodowithadatapacket
inavirtualcircuitnetworkiseasy:therouterjustusesthecircuitnumbertoindexintoatabletofind
outwherethepacketgoes.Inadatagramnetwork,nosetupisneededbutamorecomplicatedlookup
procedureisrequiredtolocatetheentryforthedestination(Garfinkel,S,2002)..
Arelatedissueisthatthedestinationaddressesusedindatagramnetworksarelongerthancircuit
numbersusedinvirtualcircuitnetworksbecausetheyhaveaglobalmeaning.Ifthepacketstendtobe
fairlyshort,includingafulldestinationaddressineverypacketmayrepresentasignificantamountof
overhead,andhenceawasteofbandwidth(Hull,B,2006).
Yetanotherissueistheamountoftablespacerequiredinroutermemory.Adatagramnetworkneeds
tohaveanentryforeverypossibledestination,whereasavirtualcircuitnetworkjustneedsanentry
foreachvirtualcircuit.However,thisadvantageissomewhatillusorysinceconnectionsetuppackets
havetoberoutedtoo,andtheyusedestinationaddresses,thesameasdatagramsdo.
Virtualcircuitshavesomeadvantagesinguaranteeingqualityofserviceandavoidingcongestion
withinthenetworkbecauseresources(e.g.,buffers,bandwidth,andCPUcycles)canbereservedin
advance,whentheconnectionisestablished.Oncethepacketsstartarriving,thenecessarybandwidth
androutercapacitywillbethere.Withadatagramnetwork,congestionavoidanceismoredifficult.
Fortransactionprocessingsystems(e.g.,storescallinguptoverifycreditcardpurchases),the
overheadrequiredtosetupandclearavirtualcircuitmayeasilydwarftheuseofthecircuit.Ifthe
majorityofthetrafficisexpectedtobeofthiskind,theuseofvirtualcircuitsinsidethenetwork
makeslittlesense.Ontheotherhand,forlongrunningusessuchasVPNtrafficbetweentwo
corporateoffices,permanentvirtualcircuits(thataresetupmanuallyandlastformonthsoryears)
maybeuseful(Jacobson,V,1990).
Virtualcircuitsalsohaveavulnerabilityproblem.Ifaroutercrashesandlosesitsmemory,evenif
itcomesbackupasecondlater,allthevirtualcircuitspassingthroughitwillhavetobeaborted.In
contrast,ifadatagramroutergoesdown,onlythoseuserswhosepacketswerequeuedintherouterat
thetimeneedsuffer(andprobablynoteventhensincethesenderislikelytoretransmitthemshortly).
Thelossofacommunicationlineisfataltovirtualcircuitsusingit,butcaneasilybecompensatedfor
ifdatagramsareused.Datagramsalsoallowtherouterstobalancethetrafficthroughoutthenetwork,
sinceroutescanbechangedpartwaythroughalongsequenceofpackettransmissions(Jacobson,V,
1990).

Theroutingalgorithm isthatpartofthenetworklayersoftwareresponsiblefordecidingwhich
outputlineanincomingpacketshouldbetransmittedon.Ifthenetworkusesdatagramsinternally,this
decisionmustbemadeanewforeveryarrivingdatapacketsincethebestroutemayhavechanged
sincelasttime.Ifthenetworkusesvirtualcircuitsinternally,routingdecisionsaremadeonlywhena
newvirtualcircuitisbeingsetup.Thereafter,datapacketsjustfollowthealreadyestablishedroute.
Thelattercaseissometimescalled sessionrouting becausearouteremainsinforceforanentire
session(e.g.,whileloggedinoveraVPN).
Itissometimesusefultomakeadistinctionbetweenrouting,whichismakingthedecisionwhich
routestouse,andforwarding,whichiswhathappenswhenapacketarrives.Onecanthinkofarouter
ashavingtwoprocessesinsideit.Oneofthemhandleseachpacketasitarrives,lookingupthe
outgoinglinetouseforitintheroutingtables.Thisprocessisforwarding.Theotherprocessis
responsibleforfillinginandupdatingtheroutingtables.Thatiswheretheroutingalgorithmcomes
intoplay(Joel,A,2002).
Regardlessofwhetherroutesarechosenindependentlyforeachpacketsentoronlywhennew
connectionsareestablished,certainpropertiesaredesirableinaroutingalgorithm:correctness,
simplicity,robustness,stability,fairness,andefficiency.Correctnessandsimplicityhardlyrequire
comment,buttheneedforrobustnessmaybelessobviousatfirst.Onceamajornetworkcomeson
theair,itmaybeexpectedtoruncontinuouslyforyearswithoutsystemwidefailures.Duringthat
periodtherewillbehardwareandsoftwarefailuresofallkinds.Hosts,routers,andlineswillfail
repeatedly,andthetopologywillchangemanytimes.Theroutingalgorithmshouldbeabletocope
withchangesinthetopologyandtrafficwithoutrequiringalljobsinallhoststobeaborted.Imagine
thehavocifthenetworkneededtoberebootedeverytimesomeroutercrashed(Gross,G,1994).
Stabilityisalsoanimportantgoalfortheroutingalgorithm.Thereexistroutingalgorithmsthat
neverconvergetoafixedsetofpaths,nomatterhowlongtheyrun.Astablealgorithmreaches
equilibriumandstaysthere.Itshouldconvergequicklytoo,sincecommunicationmaybedisrupted
untiltheroutingalgorithmhasreachedequilibrium.
Fairnessandefficiencymaysoundobvioussurelynoreasonablepersonwouldopposethem
butasitturnsout,theyareoftencontradictorygoals.Asasimpleexampleofthisconflict,lookatFig.
55.SupposethatthereisenoughtrafficbetweenAandA,betweenBandB,andbetweenCandC
tosaturatethehorizontallinks.Tomaximizethetotalflow,theXtoXtrafficshouldbeshutoff
altogether.Unfortunately,XandXmaynotseeitthatway.Evidently,somecompromisebetween
globalefficiencyandfairnesstoindividualconnectionsisneeded(Jacobson,A,2002).
4.ProjectTask#2:
Beforewecanevenattempttofindtradeoffsbetweenfairnessandefficiency,wemustdecide
whatitisweseektooptimize.Minimizingthemeanpacketdelayisanobviouscandidatetosend
trafficthroughthenetworkeffectively,butsoismaximizingtotalnetworkthroughput.Furthermore,
thesetwogoalsarealsoinconflict,sinceoperatinganyqueueingsystemnearcapacityimpliesalong
queueingdelay.Asacompromise,manynetworksattempttominimizethedistanceapacketmust
travel,orsimplyreducethenumberofhopsapacketmustmake.Eitherchoicetendstoimprovethe
delayandalsoreducetheamountofbandwidthconsumedperpacket,whichtendstoimprovethe
overallnetworkthroughputaswell(Kohler,E,2006).
Routingalgorithmscanbegroupedintotwomajorclasses:nonadaptiveandadaptive. Nonadaptive
algorithms do not base their routing decisions on any measurements or estimates of the current
topologyandtraffic.Instead,thechoiceoftheroutetousetogetfrom I to J (forall I and J)is
computed in advance, offline, anddownloaded to the routers whenthe networkis booted. This
procedureissometimescalledstaticrouting.Becauseitdoesnotrespondtofailures,staticroutingis
mostlyusefulforsituationsinwhichtheroutingchoiceisclear.Forexample,routerFinFig.should

send packets headed into the network to router E regardless of the ultimate destination.
Adaptivealgorithms,incontrast,changetheirroutingdecisionstoreflectchangesinthetopology,
andsometimeschangesinthetrafficaswell.Thesedynamicroutingalgorithmsdifferinwherethey
gettheirinformation(e.g.,locally,fromadjacentrouters,orfromallrouters),whentheychangethe
routes(e.g.,whenthetopologychanges,oreveryTsecondsastheloadchanges),andwhatmetricis
usedforoptimization(e.g.,distance,numberofhops,orestimatedtransittime).
Inthefollowingsections,wewilldiscussavarietyofroutingalgorithms.Thealgorithmscover
deliverymodelsbesidessendingapacketfromasourcetoadestination.Sometimesthegoalisto
sendthepackettomultiple,all,oroneofasetofdestinations.Alloftheroutingalgorithmswe
describeheremakedecisionsbasedonthetopology;wedeferthepossibilityofdecisionsbasedonthe
trafficlevelstoSection(Jacobson,A,2002).
Astreamofpacketsfromasourcetoadestinationiscalledaflow(Clark,1988).Aflowmightbeall
thepacketsofaconnectioninaconnectionorientednetwork,orallthepacketssentfromoneprocess
toanotherprocessinaconnectionlessnetwork.Theneedsofeachflowcanbecharacterizedbyfour
primaryparameters:bandwidth,delay,jitter,andloss.Together,thesedeterminethe QoS(Qualityof
Service)theflowrequires(Gross,G,1994).
SeveralcommonapplicationsandthestringencyoftheirnetworkrequirementsarelistedinFig.
Notethatnetworkrequirementsarelessdemandingthanapplicationrequirementsinthosecasesthat
theapplicationcanimproveontheserviceprovidedbythenetwork.Inparticular,networksdonot
needtobelosslessforreliablefiletransfer,andtheydonotneedtodeliverpacketswithidentical
delaysforaudioandvideoplayout.Someamountoflosscanberepairedwithretransmissions,and
someamountofjittercanbesmoothedbybufferingpacketsatthereceiver.However,thereisnothing
applicationscandotoremedythesituationifthenetworkprovidestoolittlebandwidthortoomuch
delay(Clark,1998).
FIFOschedulingissimpletoimplement,butitisnotsuitedtoprovidinggoodqualityofservice
becausewhentherearemultipleflows,oneflowcaneasilyaffecttheperformanceoftheotherflows.
If the first flow is aggressive and sends large bursts of packets, they will lodge in the queue.
Processingpacketsintheorderoftheirarrivalmeansthattheaggressivesendercanhogmostofthe
capacityoftheroutersitspacketstraverse,starvingtheotherflowsandreducingtheirqualityof
service.Toaddinsulttoinjury,thepacketsoftheotherflowsthatdogetthrougharelikelytobe
delayed because they had to sit in the queue behind many packets from the aggressive sender.
Manypacketschedulingalgorithmshavebeendevisedthatprovidestrongerisolationbetween
flowsandthwartattemptsatinterference(BhattiandCrowcroft,2000).Oneofthefirstoneswasthe
fairqueueingalgorithmdevisedbyNagle(1987).Theessenceofthisalgorithmisthatroutershave
separatequeues,oneforeachflowforagivenoutputline.Whenthelinebecomesidle,therouter
scansthequeuesroundrobin,asshowninFig.Itthentakesthefirstpacketonthenextqueue.Inthis
way,withnhostscompetingfortheoutputline,eachhostgetstosendoneoutofeverynpackets.Itis
fairinthesensethatallflowsgettosendpacketsatthesamerate.Sendingmorepacketswillnot
improvethisrate(Gross,G,1994).
Althoughastart,thealgorithmhasaflaw:itgivesmorebandwidthtohoststhatuselargepackets
thantohoststhatusesmallpackets.Demersetal.(1990)suggestedanimprovementinwhichthe
roundrobinisdoneinsuchawayasto

1
Round-robin

service

2 1

Output line

3
Input queues

Figure4.Roundrobinfairqueueing.

simulateabytebybyteroundrobin,insteadofapacketbypacketroundrobin.Thetrickisto
computeavirtualtimethatisthenumberoftheroundatwhicheachpacketwouldfinishbeingsent.
Eachrounddrainsabytefromallofthequeuesthathavedatatosend.Thepacketsarethensortedin
orderoftheirfinishingtimesandsentinthatorder. Thisalgorithmandanexampleoffinishtimesfor
packetsarrivinginthreeflowsareillustratedinFig.IfapackethaslengthL,theroundatwhichitwill
finishissimplyLroundsafterthestarttime.Thestarttimeiseitherthefinishtimeoftheprevious
packet,orthearrivaltimeofthepacket,ifthequeueisemptywhenitarrives(Crowcroft,2002).
CloudComputingisabroadtermthatdescribesabroadrangeofservices.Aswithothersignificant
developmentsintechnology,manyvendorshaveseizedthetermCloudandareusingitforproducts
thatsitoutsideofthecommondefinition.InordertotrulyunderstandhowtheCloudcanbeofvalue
toanorganization,itisfirstimportanttounderstandwhattheCloudreallyisanditsdifferent
components.SincetheCloudisabroadcollectionofservices,organizationscanchoosewhere,when,
andhowtheyuseCloudComputing.InthisreportwewillexplainthedifferenttypesofCloud
ComputingservicescommonlyreferredtoasSoftwareasaService(SaaS),PlatformasaService
(PaaS)andInfrastructureasaService(IaaS)andgivesomeexamplesandcasestudiestoillustrate
howtheyallwork.Wewillalsoprovidesomeguidanceonsituationswhereparticularflavorsof
CloudComputingarenotthebestoptionforanorganization(Labovitz,C,2009).
CloudComputingisoftendescribedasastack,asaresponsetothebroadrangeofservicesbuilton
topofoneanotherunderthemonikerCloud.ThegenerallyaccepteddefinitionofCloudComputing
comesfromtheNIST.TheNISTdefinitionrunstoseveralhundredwordsbutessentiallysaysthat;
Cloudcomputingisamodelforenablingconvenient,ondemandnetworkaccesstoasharedpoolof
configurablecomputingresources(e.g.,networks,servers,storage,applications,andservices)thatcan
berapidlyprovisionedandreleasedwithminimalmanagementeffortorserviceproviderinteraction.
Whatthismeansinplaintermsistheabilityforenduserstoutilizepartsofbulkresourcesandthat
theseresourcescanbeacquiredquicklyandeasily(Larmo,A,2010).
Morethanasemanticargumentaroundcategorization,webelievethatinordertomaximizethe
benefitsthatCloudComputingbringsasolutionneedstodemonstratetheseparticularcharacteristics.
Thisisespeciallytruesinceinrecentyearstherehasbeenamovebytraditionalsoftwarevendorsto
marketsolutionsasCloudComputingwhicharegenerallyacceptedtonotfallwithinthedefinition
oftrueCloudComputing,apracticeknownascloudwashing.Itisimportanttonotethatwhilefor
illustrationpurposesthiswhitepaperdrawsacleardistinctionbetweenSaaS,PaaSandIaaS,the
differencesbetweenthesecategoriesofcloudcomputing,especiallyPaaSandIaaS,haveblurredin
recentmonthsandwillcontinuetodoso.Nevertheless,withageneralunderstandingofhowthese
componentsinteractwitheachother,wewillturnourattentioninmoredetailtothetoplayerofthe
stack,SaaS(Labovitz,C,2009).
CharacteristicsofSaaS:LikeotherformsofCloudComputing,itisimportanttoensurethatsolutions
soldasSaaSinfactcomplywithgenerallyaccepteddefinitionsofCloudComputing.Somedefining
characteristicsofSaaSinclude:

Webaccesstocommercialsoftware
Softwareismanagedfromacentrallocation
Softwaredeliveredinaonetomanymodel
Usersnotrequiredtohandlesoftwareupgradesandpatches
ApplicationProgrammingInterfaces(APIs)allowforintegrationbetweendifferentpiecesof
software
CloudComputinggenerally,andSaaSinparticular,isarapidlygrowingmethodofdelivering
technology.Thatsaid,organizationsconsideringamovetothecloudwillwanttoconsiderwhich
applicationstheymovetoSaaS.Assuchthereareparticularsolutionsweconsiderprimecandidate
foraninitialmovetoSaaS;
CharacteristicsofPaaS:ThereareanumberofdifferenttakesonwhatconstitutesPaaSbutsome
basiccharacteristicsinclude:
Servicestodevelop,test,deploy,hostandmaintainapplicationsinthesameintegrated
developmentenvironment.Allthevaryingservicesneededtofulfilltheapplicationdevelopment
process
Webbaseduserinterfacecreationtoolshelptocreate,modify,testanddeploydifferentUI
scenarios
Multitenantarchitecturewheremultipleconcurrentusersutilizethesamedevelopment
application
Builtinscalabilityofdeployedsoftwareincludingloadbalancingandfailover
Integrationwithwebservicesanddatabasesviacommonstandards
SupportfordevelopmentteamcollaborationsomePaaSsolutionsincludeprojectplanningand
tools
Toolstohandlebillingandsubscriptionmanagement
PaaS,whichissimilarinmanywaystoInfrastructureasaServicethatwillbediscussedbelow,is
differentiatedfromIaaSbytheadditionofvalueaddedservicesandcomesintwodistinctflavors;
1.Acollaborativeplatformforsoftwaredevelopment,focusedonworkflowmanagementregardless
ofthedatasourcebeingusedfortheapplication.AnexampleofthisapproachwouldbeHeroku,a
PaaSthatutilizestheRubyonRailsdevelopmentlanguage.
2.Aplatformthatallowsforthecreationofsoftwareutilizingproprietarydatafromanapplication.
ThissortofPaaScanbeseenasamethodtocreateapplicationswithacommondataformortype.An
exampleofthissortofplatformwouldbetheForce.comPaaSfromSalesforce.comwhichisused
almostexclusivelytodevelopapplicationsthatworkwiththeSalesforce.comCRM.
CharacteristicsofIaaS:Aswiththetwoprevioussections,SaaSandPaaS,IaaSisarapidly
developingfield.ThatsaidtherearesomecorecharacteristicswhichdescribewhatIaaSis.IaaSis
generallyacceptedtocomplywiththefollowing;
Resourcesaredistributedasaservice
Allowsfordynamicscaling
Hasavariablecost,utilitypricingmodel
Generallyincludesmultipleusersonasinglepieceofhardware
ThereareaplethoraofIaaSprovidersouttherefromthelargestCloudplayerslikeAmazonWeb
ServicesandRackspacetomoreboutiqueregionalplayers.Asmentionedpreviously,thelinebetween
PaaSandIaaSisbecomingmoreblurredasvendorsintroducetoolsaspartofIaaSthathelpwith
deploymentincludingtheabilitytodeploymultipletypesofclouds.IaaSmakessenseinanumberof
situationsandthesearecloselyrelatedtothebenefitsthatCloudComputingbring.Situationsthatare
particularlysuitableforCloudinfrastructureinclude;

Wheredemandisveryvolatileanytimetherearesignificantspikesandtroughsintermsof
demandontheinfrastructure
Forneworganizationswithoutthecapitaltoinvestinhardware
Wheretheorganizationisgrowingrapidlyandscalinghardwarewouldbeproblematic
Wherethereispressureontheorganizationtolimitcapitalexpenditureandtomovetooperating
expenditure
Forspecificlineofbusiness,trialortemporaryinfrastructuralneeds
ThinClientDesignforthecompany:ThinClientComputingprovidesadvancedsystemdesignfor
bothsmallandlargeorganizationsbasedonmanyyearsofexperienceandbestpractices.Aneutral
adviserforawiderangeofproductsandtechnologies,ThinClientComputingmaintainsa
longstandingreputationformakingthebestinterestsofourclientsourcorefocus.Theideal
combinationofexpertdesignandbestofclassproductsarecombinedtoprovidethebestsystem
designforthelowestcost.
ThinClientImplementationforthecompany:ThinClientComputingworkscloselywithclientsto
planandexecutesystemdeploymentsinanorderlyandcarefulmanner.Properimplementation
requiresmorethanjusttechnicalexperienceitrequiresfullyunderstandingtheorganizationalneeds,
systeminterdependencies,thecorrectstepsandrighttiming.Carefulplanning,riskavoidanceand
parallelapproachestodeploymentareallbuiltintoourserviceofferings.Whetherisasingledayof
services,aweek,monthoryear,thesamecareandattentionisgiventoachievingexcellence.
5.ProjectTask#3:
ApartfromtheIPaddressesmentionedincompanysnetworkaddressing,therearealsoseveralother
addressesthathavespecialmeanings,asshowninFig.TheIPaddress0.0.0.0,thelowestaddress,is
usedbyhostswhentheyarebeingbooted.Itmeansthisnetworkorthishost.IPaddresseswith
0asthenetworknumberrefertothecurrentnetwork.Theseaddressesallowmachinestorefertotheir
ownnetworkwithoutknowingitsnumber(buttheyhavetoknowthenetworkmasktoknowhow
many0stoinclude).Theaddressconsistingofall1s,or255.255.255.255thehighestaddressis
usedtomeanallhostsontheindicatednetwork.Itallowsbroadcastingonthelocalnetwork,typically
aLAN.Theaddresseswithapropernetworknumberandall1sinthehostfieldallowmachinesto
sendbroadcastpacketstodistantLANsanywhereintheInternet.However,manynetwork
administratorsdisablethisfeatureasitismostlyasecurityhazard.Finally,alladdressesoftheform
127.xx.yy.zzarereservedforloopbacktesting.Packetssenttothataddressarenotputoutontothe
wire;theyareprocessedlocallyandtreatedasincomingpackets.Thisallowspacketstobesentto
thehostwithoutthesenderknowingitsnumber,whichisusefulfortesting(Jacobson,A,2002).
0 0

0000000000000

...

0 0

000000000000000

00

0 0 This host

Host

A host on this network

Broadcast on the

11111111111111111111111111111111
Network
127

...

1111
(Anything)

Figure5.SpecialIPaddresses.

local network
11 11

Broadcast on a
distant network
Loopback

NATNetworkAddress Translation: IPaddressesare scarce. AnISPmighthavea /16address,

giving it 65,534 usable host numbers. If it has more customers than that, it has a problem.
ThisscarcityhasledtotechniquestouseIPaddressessparingly.Oneapproachistodynamically
assignanIPaddresstoacomputerwhenitisonandusingthenetwork,andtotaketheIPaddress
backwhenthehostbecomesinactive.TheIPaddresscanthenbeassignedtoanothercomputerthat
becomesactive.Inthisway,asingle/16addresscanhandleupto65,534activeusers.
Thisstrategyworkswellinsomecases,forexample,fordialupnetworkingandmobileandother
computersthatmaybetemporarilyabsentorpoweredoff.However,itdoesnotworkverywellfor
businesscustomers.ManyPCsinbusinessesareexpectedtobeoncontinuously.Someareemployee
machines,backedupatnight,andsomeareserversthatmayhavetoservearemoterequestata
momentsnotice.Thesebusinesseshaveanaccesslinethatalwaysprovidesconnectivitytotherestof
theInternet(Gross,G,1994).
Increasingly,thissituationalsoappliestohomeuserssubscribingtoADSLorInternetovercable,
sincethereisnoconnectioncharge(justamonthlyflatratecharge).Manyoftheseusershavetwoor
morecomputersathome,oftenoneforeachfamilymember,andtheyallwanttobeonlineallthe
time.ThesolutionistoconnectallthecomputersintoahomenetworkviaaLANandputa(wireless)
routeronit.TherouterthenconnectstotheISP.FromtheISPspointofview,thefamilyisnowthe
sameasasmallbusinesswithahandfulofcomputers.Withthetechniqueswehaveseensofar,each
computermusthaveitsownIPaddressalldaylong.ForanISPwithmanythousandsofcustomers,
particularlybusinesscustomersandfamiliesthatarejustlikesmallbusinesses,thedemandforIP
addressescanquicklyexceedtheblockthatisavailable(Sanchez,Ruiz,2000).
TheproblemofrunningoutofIPaddressesisnotatheoreticalonethatmightoccuratsomepointin
thedistantfuture.Itishappeningrighthereandrightnow.Thelongtermsolutionisforthewhole
InternettomigratetoIPv6,whichhas128bitaddresses.Thistransitionisslowlyoccurring,butitwill
beyearsbeforetheprocessiscomplete.Togetbyinthemeantime,aquickfixwasneeded.Thequick
fixthatiswidelyusedtodaycameintheformofNAT(NetworkAddressTranslation),whichis
describedinRFC3022andwhichwewillsummarizebelow.(Dutcher,2001).
ThebasicideabehindNATisfortheISPtoassigneachhomeorbusinessasingleIPaddress(orat
most,asmallnumberofthem)forInternettraffic.Withinthecustomernetwork,everycomputergets
auniqueIPaddress,whichisusedforroutingintramuraltraffic.However,justbeforeapacketexits
thecustomernetworkandgoestotheISP,anaddresstranslationfromtheuniqueinternalIPaddress
tothesharedpublicIPaddresstakesplace.ThistranslationmakesuseofthreerangesofIPaddresses
thathavebeendeclaredasprivate.Networksmayusetheminternallyastheywish.Theonlyruleis
thatnopacketscontainingtheseaddressesmayappearontheInternetitself.Thethreereservedranges
are:
10.0.0.0
10.255.255.255/8
172.16.0.0 172.31.255.255/12
192.168.0.0 192.168.255.255/16

(16,777,216 hosts)
(1,048,576 hosts)
(65,536 hosts)

Thefirstrangeprovidesfor16,777,216addresses(exceptforall0sandall1s,asusual)andisthe
usualchoice,evenifthenetworkisnotlarge.
TheoperationofNATisshowninFig.Withinthecustomerpremises,everymachinehasaunique
address of the form 10.x.y.z. However, before a packet leaves the customer premises, it passes
through a NAT box that converts the internal IP source address, 10.0.0.1 in the figure, to the
customerstrueIPaddress,198.60.42.12inthisexample.TheNATboxisoftencombinedinasingle
devicewithafirewall,whichprovidessecuritybycarefullycontrollingwhatgoesintothecustomer
networkandwhatcomesoutofit.ItisalsopossibletointegratetheNATboxintoarouterorADSL
modem(Gross,G,1998).

Havingthenetworkmaintainconnectionstateisapropertyofconnectionorientednetworks,not
connectionlessones.IftheNATboxcrashesanditsmappingtableislost,allitsTCPconnectionsare
destroyed.IntheabsenceofNAT,aroutercancrashandrestartwithnolongtermeffectonTCP
connections.Thesendingprocessjusttimesoutwithinafewsecondsandretransmitsall
unacknowledgedpackets.WithNAT,theInternetbecomesasvulnerableasacircuitswitched
network.
Fourth,NATviolatesthemostfundamentalruleofprotocollayering:layerkmaynotmakeany
assumptionsaboutwhatlayerk1hasputintothepayloadfield.Thisbasicprincipleistheretokeep
thelayersindependent.IfTCPislaterupgradedtoTCP2,withadifferentheaderlayout(e.g.,32bit
ports),NATwillfail.Thewholeideaoflayeredprotocolsistoensurethatchangesinonelayerdonot
requirechangesinotherlayers.NATdestroysthisindependence(Clark,1994).
Fifth,processesontheInternetarenotrequiredtouseTCPorUDP.IfauseronmachineAdecidesto
usesomenewtransportprotocoltotalktoauseronmachineB(forexample,foramultimedia
application),introductionofaNATboxwillcausetheapplicationtofailbecausetheNATboxwill
notbeabletolocatetheTCPSourceportcorrectly. Asixthandrelatedproblemisthatsome
applicationsusemultipleTCP/IPconnectionsorUDPportsinprescribedways.Forexample,FTP,
thestandardFileTransferProtocol,insertsIPaddressesinthebodyofpacketforthereceiverto
extractanduse.SinceNATknowsnothingaboutthesearrangements,itcannotrewritetheIP
addressesorotherwiseaccountforthem.ThislackofunderstandingmeansthatFTPandother
applicationssuchastheH.323InternettelephonyprotocolwillfailinthepresenceofNATunless
specialprecautionsaretaken.ItisoftenpossibletopatchNATforthesecases,buthavingtopatchthe
codeintheNATboxeverytimeanewapplicationcomesalongisnotagoodidea.
Finally,sincetheTCPSourceportfieldis16bits,atmost65,536machinescanbemappedontoanIP
address.Actually,thenumberisslightlylessbecausethefirst4096portsarereservedforspecialuses.
However,ifmultipleIPaddressesareavailable,eachonecanhandleupto61,440machines. Aview
oftheseandotherproblemswithNATisgiveninRFC2993.Despitetheissues,NATiswidelyused
inpractice,especiallyforhomeandsmallbusinessnetworks,astheonlyexpedienttechniquetodeal
withtheIPaddressshortage.Ithasbecomewrappedupwithfirewallsandprivacybecauseitblocks
unsolicitedincomingpacketsbydefault.Forthisreason,itisunlikelytogoawayevenwhenIPv6is
widelydeployed(Kamoun,F,1995).

6. Project Task (Appexdix):

(i) Functions of Layer 2 and 3 of OSI Model:
OSI layer 2 is Datalink Layer and OSI layer 3 is Network Layer. Themaintaskofthedatalink

layeristotransformarawtransmissionfacilityintoalinethatappearsfreeofundetectedtransmission
errors.Itdoessobymaskingtherealerrorssothenetworklayerdoesnotseethem.Itaccomplishes
thistaskbyhavingthesenderbreakuptheinputdataintodataframes(typicallyafewhundredora
fewthousandbytes)andtransmittheframessequentially.Iftheserviceisreliable,thereceiver
confirmscorrectreceiptofeachframebysendingbackanacknowledgementframe. Anotherissue
thatarisesinthedatalinklayer(andmostofthehigherlayersaswell)ishowtokeepafast
transmitterfromdrowningaslowreceiverindata.Sometrafficregulationmechanismmaybeneeded
toletthetransmitterknowwhenthereceivercanacceptmoredata. Broadcastnetworkshavean
additionalissueinthedatalinklayer:howtocontrolaccesstothesharedchannel.Aspecialsublayer
ofthedatalinklayer,themediumaccesscontrolsublayer,dealswiththisproblem(Larmo,A,1998).
Thenetworklayercontrolstheoperationofthesubnet.Akeydesignissueisdetermininghowpackets
areroutedfromsourcetodestination.Routescanbebasedonstatictablesthatarewiredintothe
networkandrarelychanged,ormoreoftentheycanbeupdatedautomaticallytoavoidfailed

components.Theycanalsobedeterminedatthestartofeachconversation,forexample,aterminal
session,suchasalogintoaremotemachine.Finally,theycanbehighlydynamic,beingdetermined
anewforeachpackettoreflectthecurrentnetworkload. Iftoomanypacketsarepresentinthesubnet
atthesametime,theywillgetinoneanothersway,formingbottlenecks.Handlingcongestionisalso
aresponsibilityofthenetworklayer,inconjunctionwithhigherlayersthatadapttheloadtheyplace
onthenetwork.Moregenerally,thequalityofserviceprovided(delay,transittime,jitter,etc.)isalso
anetworklayerissue. Whenapackethastotravelfromonenetworktoanothertogettoitsdestina
tion,manyproblemscanarise.Theaddressingusedbythesecondnetworkmaybedifferentfromthat
usedbythefirstone.Thesecondonemaynotacceptthepacketatallbecauseitistoolarge.The
protocolsmaydiffer,andsoon.Itisuptothenetworklayertoovercomealltheseproblemstoallow
heterogeneousnetworkstobeinterconnected. Inbroadcastnetworks,theroutingproblemissimple,
sothenetworklayerisoftenthinorevennonexistent(Jacobson,A,2002).

(ii) Mapping on to TCP/IP Model:

Theinternetlayeristhelinchpinthatholdsthewholearchitecturetogether.ItisshowninFig.as
correspondingroughlytotheOSInetworklayer.Itsjobistopermithoststoinjectpacketsintoany
networkandhavethemtravelindependentlytothedestination(potentiallyonadifferentnetwork).
Theymayevenarriveinacompletelydifferentorderthantheyweresent,inwhichcaseitisthejobof
higherlayerstorearrangethem,ifinorderdeliveryisdesired.Notethatinternetisusedhereina
genericsense,eventhoughthislayerispresentintheInternet(Kent,C,1994).
OSI

TCP/IP

Application

Application

Presentation

Session

Transport

Transport

Network

Internet

Data link

Link

Physical

Not present
in the model

Figure6.TheTCP/IPreferencemodel.

(iii)Listofprotocolsthecompanyuseattheselevels:
HTTPHyperTextTransferProtocol
SMTPSimpleMailTransferProtocol
RTPRealTimeTransportProtocol
DNSDomainNameSystem
TCPTransmissionControlProtocol
UDPUserDatagramProtocol
IPInternetProtocol
ICMPInternetControlMessageProtocol

Application

HTTP

SMTP

RTP

DNS

Transport

TCP

UDP

Layers

Protocols
Internet

Link

IP

DSL

SONET

ICMP

802.11

Ethernet

Figure7.Protocolsusedbycompanyintheselevels.

(iv)DifferenceBetweenLevel2Switch/Level3Switch/Routers:
Anetworkswitchconnectsdevicestogetheronasinglecomputernetwork.Aswitchisalsocalled
switchinghub,bridginghub,orMACbridge.SwitchesuseMACaddressestoforwarddatatothe
correctdestination.AswitchisconsideredaLayer2device,operatingatthedatalinklayer;
switchesusepacketswitchingtoreceive,processandforwarddata(Massey,J.L.,1999).
Arouterisanetworkingdevicethatconnectscomputernetworks,forexample,connectinga
homenetworkwiththeInternet.Routersaretheworkhorsesthattransferpacketsofdatabetween
networkstoestablishandsustaincommunicationbetweentwonodesinaninternetwork.Routers
operateatLayer3(networklayer)oftheOSImodel;arouterusesthedestinationIPaddressina
datapackettodeterminewheretoforwardthepacket(Lubacz,J,2001).
Layer2switchesarefrequentlyinstalledintheenterpriseforhighspeedconnectivitybetween
endstationsatthedatalinklayer.Layer3switchesarearelativelynewphenomenon,made
popularby(amongothers)thetradepress.Thisarticledetailssomeoftheissuesintheevolution
ofLayer2andLayer3switches.Wehypothesizethatthatthetechnologyisevolutionaryandhas
itsoriginsinearlierproducts.Bridginginvolvessegmentationoflocalareanetworks(LANs)at
theLayer2level.AmultiportbridgetypicallylearnsabouttheMediaAccessControl(MAC)
addressesoneachofitsportsandtransparentlypassesMACframesdestinedtothoseports.These
bridgesalsoensurethatframesdestinedforMACaddressesthatlieonthesameportasthe
originatingstationarenotforwardedtotheotherports.Forthesakeofthisdiscussion,we
consideronlyEthernetLANs.Layer2switcheseffectivelyprovidethesamefunctionality.They
aresimilartomultiportbridgesinthattheylearnandforwardframesoneachport.Themajor
differenceistheinvolvementofhardwarethatensuresthatmultipleswitchingpathsinsidethe
switchcanbeactiveatthesametime(Nagle,J.,2004).
Layer3switchesarerouterswithfastforwardingdoneviahardware.IPforwardingtypically
involvesaroutelookup,decrementingtheTimeToLive(TTL)countandrecalculatingthe
checksum,andforwardingtheframewiththeappropriateMACheadertothecorrectoutputport.
Lookupscanbedoneinhardware,ascanthedecrementingoftheTTLandtherecalculationof
thechecksum.TheroutersrunroutingprotocolssuchasOpenShortestPathFirst(OSPF)or
RoutingInformationProtocol(RIP)tocommunicatewithotherLayer3switchesorroutersand
buildtheirroutingtables.Theseroutingtablesarelookeduptodeterminetherouteforan
incomingpacket(Nagle,J.,2004)

7.Conclusion:
ComputernetworkinghascomealongwayfromsimpleInternetProtocoltotodaysVirtual
PrivateNetwork.Whiletheageoldconceptofthenetworkisfoundationalinvirtuallyallareasof
society,ComputerNetworksandProtocolshaveforeverchangedthewayhumanswillwork,play,
andcommunicate.Forgingpowerfullyintoareasofourlivesthatnoonehadexpected,digital
networkingisfurtherempoweringusforthefuture.Newprotocolsandstandardswillemerge,
newapplicationswillbeconceived,andourliveswillbefurtherchangedandenhanced.While
thenewwillonlybebetter,themajorityofdigitalnetworking'scurrenttechnologiesarenot
cuttingedge,butratherareprotocolsandstandardsconceivedatthedawnofthedigital
networkingagethathavestoodsolidforovermanydecades.Cloudinfrastructurehasmademany
facilitationforfilestorageandVPNhasexploredallthewaytomakevirtualextensionofactual
network.Thisprojectsumsuponemodelcompanysnetworksetupandhasshownthe
enhancementsthatispossiblewithrealtimesimulationofpackettracingcoupledwithSaaS,
PaaS,IaaS.
8.References:
ALLMAN,M.,andPAXSON,V.:OnEstimatingEndtoEndNetworkPathProperties,Proc.SIGCOMM99

Conf.,ACM,pp.263274,1999.
ANDERSON,R.J.:SecurityEngineering:AGuidetoBuildingDependableDistributed Systems,2nded.,New

York:JohnWiley&Sons,2008b.

BELSNES,D.:FlowControlinthePacketSwitchingNetworks, CommunicationsNetworks,Uxbridge,

England:Online,pp.349361,1975.

BHATTI,S.N.,andCROWCROFT,J.:QoSSensitiveFlows:IssuesinIPPacketHandling,IEEEInternet

Computing,vol.4,pp.4857,JulyAug.2000.
BIRRELL, A.D., and NELSON, B.J.: ImplementingRemote Procedure Calls, ACM Trans.on Computer

Systems,vol.2,pp.3959,Feb.1984.
BRADEN,R.:RequirementsforInternetHostsCommunicationLayers,RFC1122, Oct.1989.
BRESLAU, L., CAO, P., FAN, L., PHILLIPS, G., and SHENKER, S.: Web Caching and Zipflike

Distributions:EvidenceandImplications,Proc.INFOCOMConf.,IEEE,pp.126134,1999.
CAPETANAKIS,J.I.:TreeAlgorithmsforPacketBroadcastChannels, IEEETrans.on InformationTheory,

vol.IT5,pp.505515,Sept.1979.
CERF,V.,andKAHN,R.:AProtocolforPacketNetworkInterconnection, IEEETrans. onCommun.,vol.

COM2,pp.637648,May1974.

CHANG, F., DEAN, J., GHEMAWAT, S., HSIEH, W., WALLACH, D., BURROWS, M., CHANDRA, T.,
FIKES,A.,andGRUBER,R.:Bigtable:ADistributedStorageSystemforStructuredData,Proc.OSDI2006

Symp.,USENIX,pp.1529,2006.
CHEN, S., and NAHRSTEDT, K.: AnOverview ofQoSRoutingforNextGeneration Networks, IEEE

NetworkMagazine,vol.12,pp.6469,Nov./Dec.1998.
CISCO: CiscoVisualNetworkingIndex:ForecastandMethodology,20092014,Cisco SystemsInc.,June

2010.

CLARK,D.D.:TheDesignPhilosophyoftheDARPAInternetProtocols, Proc. SIGCOMM88Conf.,ACM,

pp.106114,1988.

DALAL,Y.,andMETCLFE,R.:ReversePathForwardingofBroadcastPackets, Commun.oftheACM,vol.

21,pp.10401048,Dec.1978.

DAY,J.D.,andZIMMERMANN,H.:TheOSIReferenceModel, Proc.oftheIEEE,vol. 71,pp.13341340,

Dec.1983.
DEERING,S.E.:SIP:SimpleInternetProtocol,IEEENetworkMagazine,vol.7,pp.1628,May/June1993.
EDELMAN,B.,OSTROVSKY,M.,andSCHWARZ,M.: InternetAdvertisingandtheGeneralizedSecond

PriceAuction:SellingBillionsofDollarsWorthofKeywords,AmericanEconomicReview,vol.97,pp.242
259,Mar.2007.
FALL,K.: ADelayTolerantNetworkArchitectureforChallengedInternets, Proc. SIGCOMM2003Conf.,

ACM,pp.2734,Aug.2003.
FLANAGAN,D.:JavaScript:TheDefinitiveGuide,6thed.,Sebastopol,CA:OReilly,2010.

FLETCHER,J.:AnArithmeticChecksumforSerialTransmissions, IEEETrans.onCommun.,vol.COM0,

pp.247252,Jan.1982.
FLOYD,S.,HANDLEY,M.,PADHYE,J.,andWIDMER,J.: EquationBasedCongestion ControlforUnicast

Applications,Proc.SIGCOMM2000Conf.,ACM,pp.4356,Aug.2000.
GOODE,B.:VoiceoverInternetProtocol,Proc.oftheIEEE,vol.90,pp.14951517,Sept.2002.
GORALSKI,W.J.:SONET,2nded.,NewYork:McGrawHill,2002.
HIERTZ,G.,DENTENEER,D.,STIBOR,L.,ZANG,Y.,COSTA,X.,andWALKE,B.: The IEEE802.11

Universe,IEEECommun.Magazine,vol.48,pp.6270,Jan.2010.
HOE,J.:ImprovingtheStartupBehaviorofaCongestionControlSchemeforTCP, Proc.SIGCOMM96

Conf.,ACM,pp.270280,1996.
ITU (INTERNATIONAL TELECOMMUNICATION UNION): ITU Internet Reports 2005: The Internet of

Things,Geneva:ITU,Nov.2005.

JACOBSON,V.:CompressingTCP/IPHeadersforLowSpeedSerialLinks,RFC1144, Feb.1990.
JACOBSON,V.:CongestionAvoidanceandControl,Proc.SIGCOMM88Conf.,ACM,pp.314329,1988.
KLEINROCK,L.,andTOBAGI,F.:RandomAccessTechniquesforDataTransmission overPacketSwitched

RadioChannels,Proc.Nat.ComputerConf.,pp.187201,1975.

KOHLER,E.,HANDLEY,H.,andFLOYD,S.: DesigningDCCP:CongestionControl withoutReliability,

Proc.SIGCOMM2006Conf.,ACM,pp.2738,2006.
LARMO,A.,LINDSTROM,M.,MEYER,M.,PELLETIER,G.,TORSNER,J.,andWIEMANN,H.: TheLTE
LinkLayerDesign,IEEECommun.Magazine,vol.47,pp.5259,Apr.2009.

LEE,J.S.,andMILLER,L.E.:CDMASystemsEngineeringHandbook,London:ArtechHouse,1998.
LELAND,W.,TAQQU,M.,WILLINGER,W.,andWILSON,D.: OntheSelfSimilar NatureofEthernet

Traffic,IEEE/ACMTrans.onNetworking,vol.2,pp.115,Feb.1994.
MISHRA, P.P., KANAKIA, H., and TRIPATHI, S.: On Hop by Hop RateBased Congestion Control,

IEEE/ACMTrans.onNetworking,vol.4,pp.224239,Apr.1996.
NEUMAN,C.,andTSO,T.: Kerberos:AnAuthenticationServiceforComputerNetworks,IEEECommun.

Mag.,vol.32,pp.3338,Sept.1994.
PADHYE,J.,FIROIU,V.,TOWSLEY,D.,andKUROSE,J.:ModelingTCPThroughput:ASimpleModeland

ItsEmpiricalValidation,Proc.SIGCOMM98Conf.,ACM,pp.303314,1998.
RUIZSANCHEZ, M.A., BIERSACK, E.W., and DABBOUS, W.: Survey and Taxonomy of IP Address

LookupAlgorithms,IEEENetworkMagazine,vol.15,pp.823,Mar.Apr.2001.
SALTZER,J.H.,REED,D.P.,andCLARK,D.D.: EndtoEndArgumentsinSystemDesign,ACMTrans.on

ComputerSystems,vol.2,pp.277288,Nov.1984.
TAN,K.,SONG,J.,ZHANG,Q.,andSRIDHARN,M.: ACompoundTCPApproachfor HighSpeedandLong

DistanceNetworks,Proc.INFOCOMConf.,IEEE,pp.112,2006.
TURNER,J.S.:NewDirectionsinCommunications(orWhichWaytotheInformation Age),IEEECommun.

Magazine,vol.24,pp.815,Oct.1986.
WARNEKE,B.,LAST,M.,LIEBOWITZ,B.,andPISTER,K.S.J.:SmartDust:CommunicatingwithaCubic

MillimeterComputer,IEEEComputer,vol.34,pp.4451,Jan.2001.
WAYNER,P.: DisappearingCryptography:InformationHiding,Steganography,andWatermarking,3rded.,

SanFrancisco:MorganKaufmann,2008.

WYLIE, J., BIGRIGG, M.W., STRUNK, J.D., GANGER, G.R., KILICCOTE, H., and KHOSLA, P.K.:
SurvivableInformationStorageSystems,IEEEComputer,vol.33,pp.6168,Aug.2000.
YU,T.,HARTMAN,S.,andRAEBURN,K.:ThePerilsofUnauthenticatedEncryption: KerberosVersion4,

Proc.NDSSSymposium,InternetSociety,Feb.2004.
ZHAO,B.,LING,H.,STRIBLING,J.,RHEA,S.,JOSEPH,A.,andKUBIATOWICZ,J.:

Tapestry:AResilientGlobalScaleOverlayforServiceDeployment,IEEEJ.onSelectedAreasin
Commun.,vol.22,pp.4153,Jan.2004.
ZIV,J.,andLEMPEL,Z.:AUniversalAlgorithmforSequentialDataCompression,

IEEETrans.onInformationTheory,vol.IT3,pp.337343,May1977.

9.Annexure:
Packettracerfiles(i)3router.PKT,(ii)dynamic.PKT,(iii)nat_translation.PKTare
attachedherewithalongwiththisprojectfile.