Routing on the Internet - Lunds tekniska högskolaBuilding a Shortest Path Tree 2012-10-30 ETSF10 - Internet Protocols 31 • After flooding • Take: A ARPANET Routing Strategies

RoutingontheInternet

ETSF05/ETSF10– InternetProtocols

Howdoesroutingwork?

ETSF05/ETSF10- InternetProtocols 3

Choosinganoptimalpath• Accordingtoacostmetric• Decentralisedforwarding– eachrouterhasfull/necessaryinformation

Packet-switchedRouting

4ETSF05/ETSF10- InternetProtocols

Routing inPacketSwitchingNetworks

• Selectrouteacrossnetworkbetweenendnodes• Characteristicsrequired:– Correctness– Simplicity– RobustnessvsStability– FairnessvsOptimality– Efficiency


SA DA Data SA DA Data

Router

• Internetworkingdevice– Passesdatapacketsbetweennetworks– ChecksNetworkLayeraddresses– UsesRouting/forwardingtables

Two functions:❶ Routing❷ Forwarding


RouterArchitectureOverview


InputPort

Decentralizedswitching:• Givendestination,lookupoutputportusingroutingtable

• Goal:completeinputportprocessingat‘linespeed’

Physicallayer:bit-levelreception

Datalinklayer:e.g.,Ethernet


InputPortQueuing

• Fabricslowerthatsumofinputportsà queuing• Delayandloss duetoinputbufferoverflow• Head-of-the-Line(HOL)blocking:Datagramatfrontofqueuepreventsothersinqueuefromproceeding


OutputPort

PriorityScheduling:• Schedulingdisciplinemaychooseamongqueueddatagramsfortransmission


OutputPortQueuing

• Datagrams’arrivalratethroughtheswitchexceedsthetransmissionrateoftheoutputlineà buffering

• Delayandlossduetooutputportbufferoverflow


SwitchingFabrics


ForwardingTable

OSPFDomain

RIPDomain

BGP

OSkernel

OSPFProcess

OSPFRoutingtables

RIPProcess

RIPRoutingtables

BGPProcess

BGPRoutingtables

ForwardingTableManager

Staticroutingtable

RoutingTablesandForwardingTable


Routercache

• Savenexthopforpackettype(e.g.addr andTOS)– Keeppacketswithinasessiononthesamepath– Prohibitsreordering– decreasesdelayvariations

• Worksinbothdirections– Replytakethesamepathasrequest

• Drawback:forlongsessions(e.g.video)sessioncontinuitymightbebrokeniflink fails (e.g.mobility)

• Typicalforusernetworks


PerformanceCriteria

• Usedforselectionofroute• Simplestistochoose“minimumhop”• Canbegeneralizedas“leastcost”routing• Because“leastcost”ismoreflexibleitismorecommonthan“minimumhop”


Flooding

• InFlooding anincomming packetisretransmittedonalloutgoing links.Ahopcounter isused toprevent loops

• What are thealternativestofind theleast costpath.

B

BestPath:DecisionTimeandPlace

Decisiontime(when?)• Packetorvirtualcircuit(session)basis• Fixedordynamicallychanging

Decisionplace(where?)• Distributed- madebyeachnode• Morecomplex,butmorerobust

• Centralized– madebyadesignatednode• Source– madebysourcestation


NetworkInformationSourceandUpdateTiming• Routingdecisionsusuallybasedonknowledgeofnetwork,trafficload,andlinkcost– Distributedrouting

• Usinglocalknowledge, information fromadjacentnodes,information fromallnodesonapotential route

– Centralrouting• Collectinformation fromallnodes

Issueofupdatetiming

• Dependsonroutingstrategy• Fixed- neverupdated• Adaptive- regularupdates


RoutingStrategies- FixedRouting

• Useasinglepermanentrouteforeachsourcetodestinationpairofnodes

• Determinedusingaleastcostalgorithm• Routeisfixed– Untilachangeinnetworktopology– Basedonexpectedtrafficorcapacity

• Advantageissimplicity• Disadvantageislackofflexibility– Doesnotreacttonetworkfailureorcongestion


RoutingStrategies- AdaptiveRouting

• Usedbyalmostallpacketswitchingnetworks• Routingdecisionschangeasconditionsonthenetworkchangeduetofailureorcongestion

• RequiresinformationaboutnetworkDisadvantages• Morecomplex• Tradeoffbetweenqualityandoverhead• Tooquickupdatesmayleadtooscillations• Tooslowupdatesmayleadtooutdatesinformation


Linkcost

• Acost function describes thecost fortransmittingapacketoveralink

• Thelink cost can depend one.g.– Datarate– Load– Length– Transmissionmedia– etc

Graf

N={A,B,C,D,E}

Anetworkcanbedescribedbyagraph,consistingofnodes(N)andadges (E)withweightsw(e),i.e.costs.Example (undirectedgraph)

E w(e)AB 3

AD 1

BC 1

BD 1

BE 1

CE 3

DE 3

ARPANETRoutingStrategies1stGeneration

DistanceVectorRouting

• 1969• Distributedadaptiveusingestimateddelay

– Queuelengthusedasestimateofdelay• VersionofBellman-Ford algorithm• Nodeexchangesdelayvectorwithneighbors• Updateroutingtablebasedonincominginformation• Doesn'tconsiderlinespeed,justqueuelengthandrespondsslowlytocongestion


Leastcostalg 1Bellman-Ford• Find theshortes path fromone sourcenode s totheothers.• Let !(#) bethecost from% to#

• Addition:Keep track of thepath!!

Init:!(%) = 0! # = ∞, # ≠ %

fori = 1./ 0 − 1foreach # ∈ 0

! # = min5∈6

{! 8 +:(8, #)} //Findtheshortestpathfrom//nodeutonodeninonestep

Example Bellman-Ford

Nod A B C D E

init 0 ∞ ∞ ∞ ∞i=1 0 3 ∞ 1 ∞i=2 0 2 4 1 4

i=3 0 2 3 1 3

i=4 0 2 3 1 3

A

D

B

E

C1

11 1

Netgraph asatree

Nod Dist

A 0

B 2

C 3

D 1

E 3

Distantvector forAwhen thealgorithmconverged

Bellman-Fordsalgoritmgrafiskt

<=> = min ?=@ + <@> , ?=A + <A> , ?=B + <B> …

<=> = DE# <=>, ?=F + <F>

Jmf Stallings kap 19.2

Distance Vector Routing

• Best path info shared locally– Periodically– Upon any change

• Routing tables updated for– New entries– Cost changes

2012-10-30 15ETSF10 - Internet Protocols

Updating a Routing Table

2012-10-30 ETSF10 - Internet Protocols 16

to C (+2)

Updating Algorithm (Bellman-Ford)if (advertised destination not in table)

{

add new entry // rule #1

}

else if (adv. next hop = next hop in table)

{

update cost // rule #2

}

else if (adv. cost < cost in table)

{

replace old entry // rule #3

}2012-10-30 17ETSF10 - Internet Protocols

Completed Routing Tables


Problem: Count to Infinity (1)

I can reach A at cost 1A I can reach

A at cost 2I can reach A at cost 3

I can reach A at cost 3

I don’t know A, so I will not tellA I can reach

A at cost 2

2012-10-30 19

1 2 3

1 2 3

ETSF10 - Internet Protocols

Problem: Count to Infinity (2)

A I can reach A at cost 5

2012-10-30 20







1 2 3

1 2 3

1 2 3


Solution: Split Horizon

A I can’t reach A

I can reach A at cost 2 but I won’t tell 1

A I can’t reach A

I can’t reach A

A I can reach A at cost 2 but I won’t tell 1


2012-10-30 21

1 2 3

1 2 3

1 2 3


ARPANETRoutingStrategies2ndGeneration

Link-StateRouting• 1979• Distributedadaptiveusingdelay criterion– Usingtimestampsofarrival,departureandACKtimes

• Re-computesaveragedelaysevery10seconds• Anychangesarefloodedtoallothernodes• Re-computesroutingusingDijkstra’s algorithm• Goodunderlightandmediumloads• Underheavyloads,littlecorrelationbetweenreporteddelaysandthoseexperienced


Leastcostalg 2Dijkstra

Init:!(%) = 0! # = ∞, # ≠ %G = ∅ //Visited nodes

whileG ⊂ J8 = KLMmin

5∉O!(8) //Find least weight tounvisted node

G = G⋃8 //Add u tovisited nodesfor# ∉ G

! # = min{! # , ! 8 +:(8,#)} //Lesscost togoton viau?

• Findtheshortes path fromone sourcenode s totheothers.• Let !(#) bethecost from% to#

• Addition:Keep track of thepath!!

SPFexempelmedgrafA

D

B

E

C3

1

A

D

B

E

C3

12

4

A

D

B

E

C

12

4

3

3

A

D

B

E

C

12

3

3 6

A

D

B

E

C

12

3

3

Rotnod

Permanent nod

Preliminär nod

Potentiell väg

Permanent väg

Dijkstra tabell

Besökt L(A) L(B) L(C) L(D) L(E)Q 0 ∞ ∞ ∞ ∞{A} 3:A ∞ 1:A ∞{A,D} 2:D ∞ 4:D

{A,D,B} 3:B 3:B

{A,D,B,C} 3:B

{A,D,B,C,E}

Link State Routing

• Local topology info flooded globally– Periodically– Upon any change

• Routing tables updated for– Link state changes– Cost changes


Initial Link State Knowledge


Tree Generation Algorithm (Dijkstra)put yourself to tentative list

while tentative list not empty

{

pick node which can be reached

with least cumulative cost

add it to your tree*

put its neighbours to tentative list**

with cumulative costs to reach them

}*(a.k.a. permanent list)**(if not already there)


Building a Shortest Path Tree


• After flooding• Take: A

ARPANETRoutingStrategies3rdGeneration• 1987• Linkcostcalculationchanged– Dampenroutingoscillations– Reduceroutingoverhead

• Measureaveragedelayoverlast10secondsandtransformintolinkutilizationestimate

• Calculateaverageutilizationbasedoncurrentvalueandpreviousaverage

R # +1 = STU # + S

T R(#)• Useaslinkcostafunctionbasedntheaverageutilization


AutonomousSystems(AS)

• Exhibitsthefollowingcharacteristics:– Isasetofroutersandnetworksmanagedbyasingleorganization

– Consistsofagroupofroutersexchanginginformationviaacommonroutingprotocol

– Exceptintimesoffailure,isconnected(inagraph-theoreticsense);thereisapathbetweenanypairofnodes


InteriorRouterProtocol(IRP)InteriorGatewayProtocol(IGP)

• SharedroutingprotocolspassesroutinginformationbetweenrouterswithinanAS

• Customtailoredtospecificapplicationsandrequirements

• Examples:– RoutingInformationProtocol(RIP)– OpenShortestPathFirst(OSPF)


ExteriorRouterProtocol(ERP)ExteriorGatewayProtocol(EGP)

• ProtocolusedtopassroutinginformationbetweenroutersindifferentASs

• WillneedtopasslessinformationthananIRP– TotransmitadatagramfromahostinoneAStoahostinanotherAS,arouterinthefirstsystemneedonlydeterminethetargetASanddevisearoutetogetintoit

– OncethedatagramentersthetargetAS,therouterswithinthatsystemcancooperatetodeliverthedatagram

– TheERPisnotconcernedwithdetailsoftheroute• Examples:– BorderGatewayProtocol(BGP)– OpenShortestPathFirst(OSPF)


Figure 19.9 Application of Exterior and Interior Routing Protocols

Autonomous System 1Autonomous System 2

Subnetwork1.4

Subnetwork1.3

Subnetwork1.2

Subnetwork1.1

Subnetwork2.3

Subnetwork2.2

Subnetwork2.4

Subnetwork2.1

Interior router protocolExterior router protocol

R4R1

R5

R8

R7

R6R2R3


RoutingAlgorithmsandProtocols


Distance-VectorRouting• Requiresthateachnodeexchangeinformationwithitsneighboringnodes– Twonodesaresaidtobeneighborsiftheyarebothdirectlyconnectedtothesamenetwork

• Usedinthefirst-generationroutingalgorithmforARPANET

• Eachnodemaintainsavectoroflinkcostsforeachdirectlyattachednetworkanddistanceandnext-hopvectorsforeachdestination

• RoutingInformationProtocol(RIP)usesthisapproach


RIP(RoutingInformationProtocol)

• IncludedinBSD-UNIXDistributionin1982• Distancemetric:– #ofhops(max15)todestinationnetwork

• Distancevectors:– exchangedamongneighboursevery30secondviaResponseMessage(advertisement)

• Implementation:– Applicationlayerprotocol,usesUDP/IP


RIPupdatemessage

• Contains thewhole forwarding table• Actiononreception:– Add 1to cost inreceivedmessage– Changenext hopto sending router– Apply RIPupdating algorithm

• Received update msgs identify neighbours!


RIP:LinkFailureandRecovery

• Ifnoadvertisementheardafter180seconds– Neighbour/linkdeclareddead– Routesvianeighbourinvalidated(infinitedistance=16hops)

– Newadvertisementssenttoneighbours(triggeringachainreactioniftableschanged)

– “Poisonreverse”usedtopreventcounttoinfinityloops

– “Goodnewstravelfast,badnewstravelslow”


RoutingAlgorithmsandProtocols


Link-StateRouting• Whenarouterisinitialized,itdeterminesthelinkcostoneachofitsnetworkinterfaces

• Therouterthenadvertisesthissetoflinkcoststoallotherroutersintheinternettopology,notjustneighboringrouters

• Fromthenon,theroutermonitorsitslinkcosts• Whenthereisasignificantchange,therouteradvertisesitslinkcoststoallotherrouters

• TheOSPFprotocolisanexample• Thesecond-generationroutingalgorithmforARPANETalsousesthisapproach


OpenShortestPathFirst(OSPF)Protocol• RFC2328(RequestForComments)• UsedastheinteriorrouterprotocolinTCP/IPnetworks

• Computesaroutethatincurstheleastcostbasedonauser-configurablemetric

• Isabletobalanceloadsovermultipleequal-costpaths


OSPF(OpenShortestPathFirst)

• Dividesdomainintoareas– Limitsfloodingforefficiency– One”backbone”areaconnectsall

• Distancemetric:– Costtodestinationnetwork


Areas,RouterandLinkTypes


Graph

Networktopologyexpressedasagraph• Routers• Networks– Transit,passingdatathrough– Stub,nottransit

• Edges– Direct,routertorouter– Indirect,routertonetwork


Figure 19.11 A Sample Autonomous System

3 1

13

1

2

1 8 8

86

6

8 88

N13 N14N12

7 6

67

5

1N15

N12

1

3

1

4

2

9

1 21

3

1210

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12H1

N10

N7

N8

N9

N11

N4

N6

N3

N1

N2


Figure 19.12 Directed Graph of Autonomous System of Figure 19.11

31

1

3

2

1

8

8

6

6

88 8

N14N13N12

7

6

6

7

5

1 N15

N12

1

3 1

4

29

1

21

3

1

210

1

7

N1

N2

R1

R2

N3

8R3

N4

R4R5

R6

R7

N6

R8

R10

N8R11

N9

R9

N11

H1

R12

N10 N7


Transitnetwork

Stubnetwork

Direct edgePoint to point

Indirect edge

LinkStateAdvertisements

• Whattoadvertise?– Differententitiesasnodes– Differentlinktypesasconnections– Differenttypesofcost


RouterLinkAdvertisement


NetworkLinkAdvertisement

• Networkisapassiveentity– Itcannotadvertiseitself


SummaryLinktoNetwork

• Donebyareaborderrouters– Goesthroughthebackbone


ExternalLinkAdvertisement

• Linktoasinglenetworkoutsidethedomain


Hellomessage

• Find neighbours• Keep contact with neighbours:Iam stillalive!• Sentout periodically,typically every 10second• Ifnohellos received during holdtime (typically30seconds),neighbour declared dead.

• Compare RIPupdate messages


Routing Algorithms andProtocols

• InteriorandExteriorRouterProtocols• Distancevector– Bellman-Ford– Announcewholetabletoneighbors– RIP

• LinkState– Dijkstra– Announceneighborconnectionstowholenetwork– OSPF


Documents

Routing on the Internet - Lunds tekniska högskolaBuilding a Shortest Path Tree 2012-10-30 ETSF10 - Internet Protocols 31 • After flooding • Take: A ARPANET Routing Strategies