DTIC ADA461697: A Traffic Engineering Approach based on Minimum-Delay Routing

7 Pages·0.1 MB·English

by  Defense Technical Information Center

A Traffic Engineering Approach based on
Minimum-Delay Routing
SrinivasVutukury J.J.Garcia-Luna-Aceves
ComputerScience Department ComputerEngineeringDepartment
Baskin School ofEngineering
UniversityofCalifornia, SantaCruz, CA 95064
vutukury,jj @cse.ucsc.edu
f g
Abstract—Single-path routing provided by today’s widely used IGP’s theout-of-orderpacketdeliveryproblem.
suchasRIPmakeextremelyinefficientusageofnetworkbandwidth,andis Recently,TrafficEngineering(TE),hasreceivedtremendous
evidentinthelargeend-to-enddelaysflowsexperienceinsingle-pathrout-
attentionintheInternetcommunityandseveralIETFdraftshave
ingascomparedtominimum-delayrouting.EnhancementtoOSPFsuchas
optimizedmultipathhavenotprovedtobeadequatetobridgethislargede- appeared [2], [18], [12]. Typically in a TE approach, for a
laygap.Practicalimplementationofminimum-delayrouting,ontheother givenoptimizationcriteriaandinputtrafficmatrixspecifiedover
hand,havebeenlargelyunsuccessfulforreasonssuchasscalability, slow
medium-to-large time scales, the goal is to determine a set of
convergenceandout-of-orderpacketdelivery. Thispaperproposesatraf-
flows with associated paths and bandwidths that meet the op-
ficengineeringsolutionthatforagivenlong-termtrafficmatrixadaptsthe
minimum-delayroutingtothebackbonenetworkswhichispracticalandis timization criteria. TE solutions are most effective in a net-
suitabletoimplementinaDifferentialServicesframework. Asimplescal- workunderasingleadministrativedomainsuchasISPs,where
ablepacketforwardingtechniqueisintroducedthatdistinguishesbetween
knowledgeofthelinkcharacteristicsandinputtrafficmatrixcan
datagram and traffic that requires in-order delivery and forwards them
accordinglyandefficiently. Usingsimulationsweshowthatthedelaysob- beobtained.TEisusuallyemployedinoperationalnetworksaf-
tainedarecomparabletominimumdelaysandfarbetterthansingle-path tertopologyandcapacityplanninghavetakenplace. Trafficen-
routing.
gineeringhasawidescopeandcoversdiverseformsofrouting
thataddresssurvivability,QoS,policy-basedrouting,apartfrom
I. INTRODUCTION congestionmanagementandbandwidthutilization. Bandwidth
utilizationandreductionofdelaysduetocongestion,however,
The shortest-path routing protocols such as RIP[10],
are of pressing importance and generally more difficult to ad-
EIGRP[1]andOSPF[14]thatarewidelyusedasInteriorGate-
dress,andthusisthefocusofthispaper.
way Protocols (IGPs) in today’s Internet are highly inefficient
Despitetheinitialfailureofdirectintroductionofminimum-
withrespecttobandwidthusage.Toimprovebandwidthutiliza-
delay routing in networks, a traffic engineering approach to
tion and reduce delays, severalimprovementto the basic rout-
minimum-delay routing seems to have great potential for suc-
ingprotocolshavebeenproposed[15],[19],[21],[6]. However,
cess. We explore this potential in this paper. To the best of
there has been no theoretical basis for such improvement and
ourknowledge,thisisthefirstattempttoconstructaTEsystem
have remained adhoc. For example, in ECMP[15] load is dis-
basedonminimum-delayroutingformulatedin[9].
tributed equally over multiple equal-cost paths typically using
There are several algorithms to solve the minimum-delay
simple round-robindistributionor usinghashingon the packet
routing problem[3], [17]. As in most optimization problems,
headers. To make optimal use of network resources and min-
a solution to the minimum-delay routing requires splitting of
imize delays, traffic between source-destination pairs may of-
traffic between a source-destination pair along multiple paths.
ten have to be split and routed along multiple paths in propor-
Setting up explicitpathsfrom the source to the destination us-
tionsthatarenotnecessarilyequal[9].ThoughOSPF-OMP[21]
ingMPLSleadstocomplexityatthesource.Ontheotherhand,
suggests using unequal traffic distritbution on multiple paths,
if splitting of flows is not allowed, in order to simplify MPLS
it provides no concrete method to find a distribution that con-
implementation,thesolutionsobtainedaresub-optimal[23].
formswiththeoreticallyoptimaldistribution. Onepracticalim-
To obtain an accurate implementation of the solution that
plementationofoptimalrouting,namelyCodex[11],usesvirtual
is practical and scalable we use adopt ideas from Differential
circuitstosetupupflowsbetweensource-destinationtorealize
Services model[7] and OSPF-OMP[21]. The solution to the
optimal distribution, but the architecture introduces unaccept-
minimum-delayproblemisobtainedusingoff-linealgorithmsin
ablecomplexityattheingressandcoreroutersandisnotscale-
theformofroutingparameterswhicharethendownloadedinto
ble. In[22],weshowedhowminimum-delayroutingcanbeap-
therouters,manuallyorusingautomatedtools. Thepacketfor-
proximatedandadaptedtohighlydynamictrafficonshorttime-
wardingmoduletoforwardpacketsaccordingtotheroutingpa-
scales. The main drawback of that approach was that packets
rameters. Attheingressrouter,akeyisinsertedineachpacket.
may arrive out-of-orderat destination effecting protocols such
In the intermediaterouterspacketsare forwardedbased on the
as TCP. This paper is an extension of that work and addresses
keyandtheroutingparameters.Thismethodisbetterthanhash-
ing on source-destination pair[21] as it distinguishes between
ThisworkwassupportedinpartbytheDefenseAdvancedResearchProjects
Agency(DARPA)undergrantsF30602-97-1-0291andF19628-96-C-0038. manyconnectionsbetweena source-destinationpair. Also dif-
Report Documentation Page Form Approved
OMB No. 0704-0188
Public reporting burden for the collection of information is estimated to average 1 hour per response, including the time for reviewing instructions, searching existing data sources, gathering and
maintaining the data needed, and completing and reviewing the collection of information. Send comments regarding this burden estimate or any other aspect of this collection of information,
including suggestions for reducing this burden, to Washington Headquarters Services, Directorate for Information Operations and Reports, 1215 Jefferson Davis Highway, Suite 1204, Arlington
VA 22202-4302. Respondents should be aware that notwithstanding any other provision of law, no person shall be subject to a penalty for failing to comply with a collection of information if it

does not display a currently valid OMB control number.
1. REPORT DATE 3. DATES COVERED
2000 2. REPORT TYPE 00-00-2000 to 00-00-2000
4. TITLE AND SUBTITLE 5a. CONTRACT NUMBER
A Traffic Engineering Approach based on Minimum-Delay Routing
5b. GRANT NUMBER
5c. PROGRAM ELEMENT NUMBER
6. AUTHOR(S) 5d. PROJECT NUMBER
5e. TASK NUMBER
5f. WORK UNIT NUMBER
7. PERFORMING ORGANIZATION NAME(S) AND ADDRESS(ES) 8. PERFORMING ORGANIZATION
University of California at Santa Cruz,Department of Computer REPORT NUMBER
Engineering,Santa Cruz,CA,95064
9. SPONSORING/MONITORING AGENCY NAME(S) AND ADDRESS(ES) 10. SPONSOR/MONITOR’S ACRONYM(S)
11. SPONSOR/MONITOR’S REPORT
NUMBER(S)
12. DISTRIBUTION/AVAILABILITY STATEMENT
Approved for public release; distribution unlimited
13. SUPPLEMENTARY NOTES
14. ABSTRACT
15. SUBJECT TERMS
16. SECURITY CLASSIFICATION OF: 17. LIMITATION OF 18. NUMBER 19a. NAME OF
ABSTRACT OF PAGES RESPONSIBLE PERSON

a. REPORT b. ABSTRACT c. THIS PAGE 6
unclassified unclassified unclassified
Standard Form 298 (Rev. 8-98)
Prescribed by ANSI Std Z39-18
ferentialtreatmentisgiventodatagramandtrafficthatrequires themcanbeusedtoobtainthesolution. Ifnecessary,afastbut
in-orderdelivery. Theeffectisthatamoreaccuratedistribution suboptimal method described in [22] can also be used. Once
ofactualtrafficaccordingtoroutingparametersisachievedand thesolutionisobtained,mechanismssuchMPLS’sLSP,virtual-
consequently the end-to-end delays are closer to the optimal. circuits or routing parameters are setup in the network to for-
The architecture is practical and can be implemented in con- wardtrafficalongpathsspecifiedbythesolution.Ourproposed
juction with other per-hop behaviors the Differential Services implementationisbasedonroutingparametersandisdescribed
architecture. inthenextsection,butbeforewemoveontothenextsectionlet
The paper is organized as follows. Section II describes the usseehowtheabovesolutioncanbeimplementedusingMPLS
minimum-delay routing problem and the challenges involved andvirtual-circuitsandwhatthedrawbacksare.
in implementing it. Section III describes our implementation We assume the TE solution is implemented in a single con-
strategy. InSectionIV,throughsimulationsweevaluatetheef- tiguousdomainwherethetrafficmatrixandthelinkcharacter-
fectivenessoftheTEsysteminreducingend-to-enddelaysand istics is knownapriori. The minimum-delayroutingalgorithm
congestion.SectionVconcludesthepaper. is first appliedoffline on the giventraffic matrix(the set )
i
andthegivennetworkusinganyofthealgorithmsinthefrerfjegr-
II. MINIMUM-DELAY ROUTING encesmentionedabovetoobtaintheflowset .
Weusetheminimumdelayroutingoroptimalroutingformu- Foreachflow ,anMPLSlabelandtheassofcxipat>ed0bjapn2dwWidtgh
lated in Bertsekas and Gallager [4] as the basis for our traffic is obtained anxdpa corresponding LSP (label-switched path) is
engineeringtechnique. Theformulationisreproducedherefor setupinthenetwork.Theendresultis,ateachsourcenodeaset
convenience.Inthenextsectionwewilldescribeourimplemen- of labels with correspondingbandwidthsare obtainedfor each
tation. destination. Each source node then distributes the traffic orig-
Let a computer network be represented as a graph inating at the node for each destinationaccordingto the band-
where is the set of routers and is the set ofGlink=s widthandassignslabelstothepackets. Themaindrawbackof
b(Netw;Le)en them.N is the neighbor set of rLouter . For , the abovetechnique is that at each source for each destination
i
let be thNe expected input traffic, measuried in biits6=pejr there can be flows. Therefore there can be potentially
secornjid(cid:21),en0teringthenetworkatrouter anddestinedforrouter flows as inpu0t(tLo)the weighted round-robin (WRR) procedurLe
. Let bethe setof alldirectedpatihsconnecting and in thatdistributestrafficforathesamedestination. Thisquadratic
i
tjhenetwPojrk,andlet . For ,let i bejthe complexitycanrestrictthescalabilityofMPLSmethod. Inthe
trafficrateonpath .WIF= SisPjtiheexppec2teWdtrafficx,pm(cid:21)ea0suredin interiorrouterstoo, the numberof lablesincreasesrapidly, but
bitspersecond,onplink fik ,where and is thiscanbecontrolledtocertainextentifmultipoint-to-pointag-
thecapacityoflink (i;ink)bitspers0ec(cid:20)onfdi,kf(cid:20)romCickonservCaitkion gregationof LSPs is used. A virtualcircuitimplementationof
oftrafficwehave (i;k) thesolution,suchasCodex[11],hasthesamescalabilityprob-
lem.
(1) III. PROPOSED IMPLEMENTATION
i
rj = X xp The complexity in the MPLS and virtual-circuit implemen-
p2Pji tationcanbeovercomeifthe flowsare setupusingroutingpa-
(2)
rameters.Theroutingparameter specifiesthefractionofthe
fik = X xp: i
traffic that receivesfordestinat(cid:30)iojkn thatit hasto forwardon
(i;k)2p;p2W
link ,aindisdefinedasfollows[4j]:
Let bedefinedastheexpectednumberofmessagesor
(i;k)
packetDspike(r:s)econdtransmittedonlink timestheexpected
delay per message or packet, includin(gi;tkh)e queuing delays at
the link. We assume dependsonly on flow through i fik(j) (4)
lliinnkk(cia;pka)caitnyd. WlinekacshsauDrmaicket(e:r)isticssucishaascpornotipnaugoautisofaninkddelcaoynavnedx (cid:30)jk = Pm2Nifim(j)
where istherateoftrafficdestinedfor thatflowsonlink
functionthattendstoinfiniDtyika(sfik)approaches . Thisisthe .fik(j) j
caseforexamplewhenthelinkifsimk odeledasanCMik/M/1queue. (i;Nke)ighborsfor whichthe routingparametersare greaterthan
The total expected delay per message times the total expected
zeroaresuccessors,andthesetisdefinedas .
number of message arrivals per second is denoted by and Alsonotethat and S.jiT=hefnkej(cid:30)xtij-kho>p0sget
definedas DT i i
represent a(cid:30)djikre(cid:21)cte0d-acycPlick2gNraip(cid:30)hj.kO=bs1erve that if ’s are
i
rSejstricted to 0 or 1, the routingproblemreducesto sing(cid:30)le-path
(3) routing.
DT = X Dik(fik) Once the routing parameters are obtained, they are down-
(i;k)2L loaded into the routers. The packet forwarding mechanism of
The minimum-delay routing problem is to minimize . the routing architecture then ensures that traffic is forwarded
This is a non-linearprogrammingproblem, and the solutioDnTis according those parameters. The WRR handles atmost in-
i
thesetofflows: . Therearemanyalgorithmstosolve putsforeachdestinationinthiscase,whichisfarmorescNalable
thisproblem[3],f[x5]p,>[8]0,g[9],[13],[20],[16],[17],andanyof than . This task is non-trivial for two reasons: non-zero
O(L)
packet-sizeandtherequirementofin-orderdeliveryofpackets. node first checks which neighbor has a deficit and forwards
TheWRR distributorisinadequatetohandlethissituationand it to that neighbor, after which k and are updated ac-
i i
a more sophisticated mechanism is needed. In the rest of the cordingly. The procedureallowsWatjmostWjbkits in excess than
section we describethis mechanism. We assume thatthereare whattheroutingparametersallow.ThisisLthefinestdistribution
two types of traffic: one tolerates out-of-orderpacket delivery onecanobtainwithoutbreakingpacketsintosmallerunits. The
(e.g.,UDP)andtheotherdoesnot(e.g.,TCP).(Notethatwewill proofisprovidedintheappendix.
denotepacketsthatdonotrequirein-orderdeliveryusingUDP
andthosethatrequirein-orderdeliverywithTCP)BecauseTCP B. ForwardingTCPtraffic
traffic must be delivered in-order, granularity of allocation for ForTCPtraffic,becausethegranularityisatflowlevel,such
TCPtrafficisattheflowlevel,butforUDPtrafficitisatpacket fine distribution of traffic as in UDP is very difficult, but by
level. In the presence of only UDP traffic, achieving accurate making some reasonable assumptions, a fairly accurate distri-
traffic distribution according routing parameters can be easily butioncanbe achieved. We assumethereare sufficientlylarge
doneusingWRR.TodeliveryTCPtrafficin-order,astraightfor- number of flows passing through a router and the number of
wardmethodthatisoftensuggestedisusingahashingfunction flowsis severalorderof magnitudegreaterthan the numberof
onthepacket’ssource-destinationaddresstodeterminethenext- next-hopchoices.ThedurationofTCPconnections,theaverage
hop[21], [19]. In OSPF-OMP[21], only the source-destination ratesofTCPconnectionsandthepacketsizesarealluniformly
pairisusedintheCRC16 hashingfunctionwhichgivesonlya distributedsothatbandwidthofagroupofTCPconnectionsis
coarse distribution. By using TCP portnumbersin addition to proportional to number of connections in the group. In back-
source-destinationaddressesinthehashfunction,finerdistribu- bone networks where there are large number of flows, we be-
tion can be achieved. But this can be expensive to implement lievethisassumptionisquitereasonable. Whenthenumberof
becauseunpackingofthepacketsisrequiredateachhop. Also flows is low, however, the distribution is relatively imprecise.
in OSPF-OMP, both UDP and TCP traffic are handled identi- Butthisshouldbeacceptablebecausewhennetworkloadislow
cally. We show that by treating UDP and TCP traffic differ- delaysduetocongestionnotaseriousproblem.Wewillproceed
ently and accounting for differentpacket sizes, greater fidelity withtheseassumptions,andpresentaprocedureforTCPpacket
between routing parameters and actual traffic distribution can forwarding. In section IV, we will givesome experimentalre-
achieved. In our approach, the hashing is decoupled from the sultstovalidatesomeoftheseassumptions. Itshouldbenoted
source-destination addresses, and hybrid packet forwarding is thatunderheavyloadconditiontheperformanceoftherouting
usedtoimprovegranularityindistribution. schemeintheworstcaseonlydropstothatofsingle-pathrout-
ing.
A. ForwardingDatagramtraffic
As mentioned earlier, we use an architecture similar to the
BeforeproceedingtodescribeTCPtrafficandhybridpacket- Diffservmodelandobviousadvantageisitcanbeincorporated
forwarding,wewillshowthat,ifonlyUDPtrafficispresentand along with other differentiated services. At the ingress router,
isthemaximumpacketsize,trafficcanbeforwardedsuchthat for each TCP connection a randomly generated key is associ-
aLtmost amountoftrafficmorethanthetrafficthatwouldbeal- ated. The ingress router maintains this per-connection table.
lowedbLytheroutingparameters.Thislimitationimposedbythe When a packet of that connection arrives, the key is inserted
packet-sizeisfundamentalbecausepackettransmissionisnon- into the packet. Effectively the computation of hash is decou-
preemptive.Fig.1belowdescribestheprocedureforforwarding pledfromthesourceanddestinationaddresses. Thecodepoint
whenonlyUDPtrafficispresent. (TOS or DSFIELD field) of the packet indicates, among other
things, thatthepacketis ofTCP-type. Within thecorethekey
——————————————————————– is used to hash into the next-hop. The 13-bit fragment offset
proceduredatagram-forwarding(P) is used to hold the hash key. To prevent fragmentation in the
Executedat onarrivalofthepacketPfor .
fbegin i jg domain, the DF field is set. This is possible because this so-
Forsome ,let ; lution is applied in a single domain and it is feasible to know
i i i i
ForwardPkto2nSeijghborWj;k(cid:20)(cid:30)jkWj the MTUs of all the links. At the ingress node, use of the 13-
i i k ; bitoffsetfieldcanthusbeprevented. Theper-hopbehaviorre-
Wjik Wijk+sizeof(P;) latedtoqueuemanagementcanstillbeimplementedinconjunc-
endWj Wj +sizeof(P)
tionwithminimum-delayrouting. Bydecouplingthekeyfrom
———————————————————————
thesource-destination,betterrandomnessinthekeydistribution
Fig.1. Packetforwardingwhenout-of-orderpacketdeliveryisacceptable. canbe achievedandtheper-packetprocessingateachhopcan
bereduced. We mustmentionthatper-connectiontablecanbe
Itcanbeviewedasageneralizationofsingle-pathrouting;in eliminatedandCRC16hashbeusedoneachenteringpacketin-
single-path routing all packets are forwarded to a single next- stead. ButrememberthatDiffservarchitecturemaintainssome
hop, whereas here, packets are forwarded to each of the sev- per-connectioninformationforprofilingandotherpurposes.
eral next-hopsin proportionsspecified by the routing parame- In OSPF-OMP, boundary values are associated with each
ters. Thisprocedureisextendedlatertohandlehybridtraffic.In next-hop. We propose a different method which uses more
Fig.1, is the total traffic that node receives and forwards memory, but is faster. With each destination at node , a
i
for , aWnjd is the portion of that tiraffic that is forwarded hash table, denoted by , is associated. Thejtable is a isin-
i i
tonjeighborWj.kWhenapacketfordestination isreceived,the glecolumntablewith HTejntries. Witheachentryofthetable
i
k j Mj
anext-hop isassociated. Thenext-hopsaredistributed existingnetworks. Thoughhashingisnotnew,whencombined
i
randomly okve2r tShej range, and the fraction of entries that point withDiffservframeworkandhybridpacketforwardingcangive
to a particular next-hopwill be proportionalto the routing pa- significantbenefits.
rameters.Thatis,foreach , entriesof ,chosen The architecture described above can be implemented with
i i i
randomly,arefilledwiththkev2alSuej m,wj;khere HTj and currentIPforwardingtechnologieswithsmallmodificationsand
i i i
. Inotherwordks, eachemntjr;yk =in(cid:30)thje;khMasjh ta- in the Diffserv framework without needing other forwarding
i i
bMlejp=oinPtskt2oSsjiommej;kneighborinthesuccessorsetandthenumber technologiessuchasMPLS. Itcanbeusedtoimplementother
traffic engineering approach using other optimization criteria
of entries filled with a successor is proportionalto the routing
[23]. The only requirementis that the solution should be rep-
parameters. ComparisonswithboundaryvaluesinOSPF-OMP
resentableusingroutingparameters.
has complexity.
i
WOh(eNna)TCPpacketfor arrivesat ,themodfunctiononthe IV. EXPERIMENTALRESULTS
keyisusedtoindexintothejtable itoobtainthenext-hop.If
ischosensuchthatitisapowHerTojiftwo,thelower We tested the effectivenessof our TE scheme through a se-
i i ries of simulation experiments. In each of the experiments,
Mbitjsofthekeycanbeusedtoindexintothehashtablel.oBge2c(Maujse)
for the same given input traffic matrix and network configu-
of the assumptions made, this should result in each next-hop
ration, the routing parameters are first computed using an off-
receiving the amount of traffic in accordance with the routing
lineminimum-delayroutingalgorithmanddownloadedintothe
parameters.
routertables. Afterthatpacketsareforwardedaccordingtothe
routing parameters. The end-to-end average delays are mea-
C. Hybridpacketforwarding
suredforeachflow. Comparisonsaremadebetweendelaysob-
Becauseoflackofcompletecontrolonthepacketforwarding tainedforsingle-pathrouting,differentvolumesofTCPflows,
of TCP traffic, the actual traffic forwarded can deviate signif- anddifferentproportionsofTCPandUDPtraffic.
icantly from the amounts specified by the routing parameters. The network used in the simulation is shown in Fig.3. The
The skew in the distribution introduced by the hashing mech- networkisacontrivednetworkwithnodedegreehighenoughto
anism, can be ironed out to some extent if there is some UDP provideseveralnext-hopchoices, butlow enoughso thatthere
trafficalsopresent. TheUDPpacketsthencanbeforwardedto are not too many one-hop paths. The links have bandwidth
neighbors that received too little traffic compared to what the 5MB, propagationdelay of 100 microseconds and packets are
routingparametersallow. OSPF-OMP does notmake this dis- ofsize1000bytes.
tinction.Themodifiedpacketforwardingprocedureisasshown
intheFig.(2).
8 7
——————————————————————– 9 6
procedurehybrid-forwarding(P)
Executedat onarrivalofpacketPfor .
fbegin i jg 0 5
if(PisaUDPpacket)then
endiFforsomek2Sji,letWjik (cid:20)(cid:30)ijkWji; 1 4
if(PisaTCPpacket)then 2 3
LetP’sheadermaptonext-hop ;
endif k Fig.3. Topologyusedinsimulations
ForwardPtoneighbor ;
k ;
i i Experiment1:ThedelaysofSParecomparedwiththedelays
Wjk Wjk+sizeof(P;)
endWji Wji+sizeof(P) obtained in our TE scheme. The input traffic consists of 500
——————————————————————— identical TCP flows; 50 of them originating from each node.
Fig.4(a) makes the delay versus load comparison. The x-axis
Fig.2. PacketforwardinginthepresenceofbothUDPandTCPpackets. denotes the load and the y-axis the delay. The average delays
of single-pathroutingare denotedby SP and the delaysofour
ThehybridprocedureissimilartotheUDP-onlyforwarding schemearedenotedbyTE.Observethatforagivenaveragede-
proceduredescribedinFig.1,exceptthattheskewindistribution lay, the load that the network can carry is much greater in TE
created by TCP traffic is mitigated by UDP traffic; greater the scheme. AtverylowloadsSPperformedbetterbecauseofthe
share of UDP traffic in the totaltraffic, finer is the distribution tendency of TE to route along longer than shortest paths, un-
ofthetrafficaccordingtotheroutingparameters.Inthediffserv der low loads. Fig. 4(b)-4(d) show the comparison of delays
model, out-of-orderprofile packets can be treated as datagram ofindividualflowsforbothschemesunderthreedifferentload
traffic. InsectionIV,wewillgivesomeperformancefiguresre- conditions. Thex-axisinthiscasedenotesflowIDsandthey-
gardingthis. Theamountofextraspacerequiredintherouting axistheaveragepacketdelaysfortheflows.Notethattoremove
table is of the order of . The processing time re- clutterandtheplotsclearer,theflowsarefirstsortedinascend-
i
quiredisoftheorder O(NMN. N)oCRC16hashateachhop ingorderofdelaysof single-pathroutingandthenplotted. As
i
isusedandnocompariOso(nlowg(itNhb)oundaryvaluesisdone.Webe- canbeseen,theproposedTEschemesignificantlyoutperforms
lievethattheproposedTEarchitecturecanbeeasilydeployedin thesingle-pathroutingastheloadinthenetworkincreases. In
Shortest-path vs TE Shortest-path vs TE
12 5
’TE’ ’SP’
’SP’ ’TE’
11 4.5
4
10
3.5
ms) 9 ms)
Avg. Delays ( 78 Avg. Delays ( 2.53
2
6
1.5
5 1
4 0.5
60 70 80 90 100 110 0 10 20 30 40 50 60
Load (Mb) Flow IDs
(a) (b)
Shortest-path vs TE Shortest-path vs TE
7 80
’SP’ ’SP’
’TE’ ’TE’
6 70
60
5
Avg. Delays (ms) 34 Avg. Delays (ms) 345000
2
20
1 10
0 0
0 10 20 30 40 50 60 0 10 20 30 40 50 60
Flow IDs Flow IDs
(c) (d)
Fig.4. ComparisonofSPFandTEscheme
Fig. 4(b),forlowloads,theyareverycloseandoftenSPisbet- ing an offline algorithmand are then established using routing
ter. Underhighloadstherecanseverecongestioninsingle-path parameters. Packets are then forwarded according to routing
routing which can be avoided in multipath routing as seen in parameters using hash techniques similar to OSPF-OMP. Our
Fig. 4(d). Because of the use of multipathsin the TE scheme approach differs from OSPF-OMP in several significant ways.
congestionand,therefore,thedelaysarereduced. Firstly,weuseIETF’sDifferentialServicesmodeltoimplement
Experiment2: We testourassumptionthatwhenlargenum- theTEsystem. Thehashcomputationforapacketisdoneonce
ber of TCP flows are present, delaysclose to minimum-delays attheingressnodeandinsertedintothepacket. Withinthenet-
can be obtained. TCP-10 indicatesthere are 10 flows between work the hash key is directly used to determine the next hop
eachsource-destinationpair.CurveUDPrepresentstheoptimal withoutfurtherhashcomputation.Tofurtherspeedupforward-
routing. Observe that in Fig. 5 as the total number of flows ingthenext-hopstructureusesatablewithnext-hopentriesin-
increasesfrom90to900, thedelaysdecreaseto levelscompa- steadofboundaryvaluesasinOSPF-OMP.UnlikeOSPF-OMP,
rableto theoptimal. Thisisbecauseofthe finergranularityin oursystemtreatsdatagramandTCPtrafficdifferentlyandtakes
distribution. into account the packet sizes giving finer granularity of traffic
Experiment3: Trafficcanbeforwardedmoreaccuratelyac- distribution. Theuse ofDiffservmodelenablescomplexoper-
cording to routing parameters, if the presence of UDP traffic ations, such as peeking into the TCP header, to be done only
andpacketsizesareconsideredduringpacketforwarding. This once at the ingress node. Also finer granularityof distribution
experiment tests this effect. We run the experiment for traffic can beachievedby decouplinghashcomputationfromsource-
that is split 60-40 between TCP and UDP. As the fraction of destinationaddress.
UDPtrafficincreases,theaveragedelaysimprove(UDP-40).As
Once problem that we have not addressed in this paper is
mentionedearlier,thisisduetoreducingtheskewintroducedby
adaptationtotrafficfluctuationsandadjacencylinkfailures.We
TCPpacketforwardingwhichOSPF-OMPdoesnotdo.
can adapt similar techniques in OSPF-OMP, but this problem
isinherentlydifficultbecauseourgoalistoachieveminimum-
V. CONCLUSIONS
delay routingrather than mere adjustmentto congestion. Also
We proposed a traffic engineering (TE) system based on the TE is described as applied to an autonomous system and
minimum-delay routing. For a given traffic matrix, multiple doesnotaddressinter-domainrouting. Thisistopicforfurther
flows between each source-destinationpair are determined us- research.
35 35
30 ’UDP’ 30 ’UDP-40’
’TCP-100’ ’TCP-only’
’TCP-10’
25 25
c Avg. Delays 1250 Avg. Delays 1250
10 10
5 5
0 0
0 10 20 30 40 50 60 70 80 0 10 20 30 40 50 60 70 80
Flow IDs Flow IDs
Fig.5. (a)Delaysasafunctionofvolumeofflows(b)Delaysinpresenceofhybridtraffic
Our packet forwarding mechanism is quite general and can APPENDIX
beusedforotheroptimizationcriteriaasin[23]. Thesystemis
A. PropertiesofDatagramForwardingAlgorithm
practicalandcanbeimplementedinconjuctionwiththeemerg-
ing DifferentialServicesarchitecture. Itofferssignificantben- Foragivenroutingparameterswehavetoshowthatthefor-
efits in termsof delayandthroughputperformanceoversingle warding algorithm(Fig. 1) does notforward more than bits
toanyofthenext-hop,andhencethetrafficdistributionisLfairly
pathrouting.
accuratefor UDP traffic evenon a small scale. And the maxi-
REFERENCES mumdeficitis .
[1] R. Albrightson, J.J.Garcia-Luna-Aceves, andJ. Boyle. EIGRP-AFast Forsimplicit(yk,w(cid:0)e1s)lLightlymodifythenotationasalldistribu-
RoutingProtocolBasedonDistanceVectors. Proc.Networld/Interop94,
May1994. torsarethesame.Assumeastreamisdividedinto streamsac-
[2] D. Awduche and et al. A Framework for Internet Traffic Engineering. cordingtheroutingparameters. Let betheamokuntoftraffic
InternetDraft,URL:draft-ietf-tewg-framework-00.txt,January2000. thatarrivedsofarand ,theamounWtoftrafficthatisforwarded
[3] D.BersekasandR.Gallager. SecondDerivativeAlgorithmforMinimum
DelayDistributedRoutinginNetworks. IEEETrans.Commun.,32:911– tostream and betWhekcorrespondingroutingparameter.Also
919,1984. denotkesthe(cid:30)vkalueof whenthe packetarrives.
[4] D.BersekasandR.Gallager.DataNetworks.2ndEd.Prentice-Hall,1992. WistpheamountofinputtraWfficthatarrivpetshintheinterval A(ta;n(cid:28)d)
[5] D.BertsekasandE.Gafni. ProjectedNetworkMethodsandOptimization
of Multicommodity Fflows. IEEETrans. Automatic Control, 28:1090– lettheamountoftrafficthatstream receivesinthesam[t;e(cid:28)i]nter-
1096,1983. valbe . Notethatattime , k willbeequl .
[6] D.Katzetal.TrafficEngineeringExtensionstoOSPF.¡draft-katz-yeung- TheoAre(tm;(cid:28)1): Inthe algorithmitnWFig.1(a)foreacAh[0s;utb]stream
ospf-traffic-01.txt¿,Oct.1999.
[7] Y.Bernetetal.AnFrameworkforDifferentiatedServices.InternetDraft, , .
May1998. k (KP(cid:0)roo1f):LI(cid:20)faApakc(kt;e(cid:28)tc)o(cid:20)ulLdn+ot(cid:30)bke(cid:26)s(cth(cid:0)ed(cid:28)u)ledbythealgorithm,
[8] L.Fratt,M.Gerla,andL.Klienrock.TheFlowDeviationMethod:AnAp-
then . Thisimplies whichisim-
proachtoStoreandForwardCommunicationNetworkDesign. Netorks,
pages97–133,1973. possi8blke;.WThkis>pr(cid:30)okvWesthateverypackPetiksWsukcc>essWfullyscheduled
[9] R.G.Gallager. AMinimumDelayRoutingAlgorithmUsingDistributed bythealgorithm.
Computation. IEEETrans.Commun.,25:73–84,January1977.
Wefirstshow .Letitbetrueuptoprocessing
[10] C.Hendrick. RoutingInformationProtocol. RFC,1058,june1988.
[11] P. Humblet and S. Soloway. Algorithm for Data Communication Net- of packets.WThke(cid:20)n,fLor+a(cid:30)llkW, . Letthe
p(cid:0)1 p(cid:0)1
works:PartIandII.CodexCorporation,1986. newp(cid:0)pa1cket beassignedtoquieuWek,we(cid:20)haLve+(cid:30)kW .
[12] W.Lai. Capacityengineeringofip-basednetworkswithmpls. Internet p p(cid:0)1
Draft,URL:draft-wlai-tewg-cap-eng-00.txt,March2000. Thisimpliesp kbecause Wk (cid:20)Wk +L.
[13] P.M.MerlinandA.Segall.AFailsafeDistributedRoutingProtocol.IEEE Substituting Wkp (cid:20) L+(cid:30)kW,pw(cid:0)e1get Wkp(cid:0)1 (cid:20) (cid:30)kWp(cid:0)1.
Trans.Commun.,27:1280–1287,September1979. p p(cid:0)1 p p
[14] J.Moy.OSPFVersion2.RFC,2328,1998. Therefore Wp =W +Lp. BecausWekat(cid:20)iniLti+ali(cid:30)zkaWtion(cid:0)L(cid:30)k
[15] J.Moy.OSPFVersion2.RFC,2328,1998. , frWomk i(cid:20)nduLct+ion(cid:30)kitWfollows . BeWcakus=e
[16] A.Segall. TheModeling ofAdaptive Routing inDataCommunication W = 0 and Wk (cid:20), Lwe+h(cid:30)avkeW
Networks. IEEETrans.Commun.,25:85–95,January1977.
[17] A.SegallandM.Sidi.AFailsafeDistributedProtocolforMinimumDelay Wk = Ak.(I0t;fot)llowstWhat= A(0;t) (cid:20) (cid:26)t andAk(0;t) (cid:20)
Routing. IEEETrans.Commun.,29:689–695,May1981. L+(cid:30)k(cid:26)t for Ak.(T0;hte)re(cid:20)fo(cid:14)r1e,+(cid:30)k(cid:26)t Ak(0;(cid:28))(cid:20)
[18] K.Takashimaandetal.ConceptofIPTrafficEngineering.InternetDraft, (cid:14)2+(cid:30)k(cid:26)(cid:28) an(cid:14)d1;(cid:14)2 2[0;L] . That Ak(t;(cid:28))(cid:20)((cid:14)1(cid:0)(cid:14)2)+
URL:draft-takashima-te-concept-00.txt,October1999.
[19] D.Thaler. Multipath issuesinunicastandmulticastnext-hopselection. d(cid:30)ikr(cid:26)ec(ttly(cid:0)fo(cid:28)l)lows.((cid:14)1 (cid:0)(cid:14)2) (cid:20) L (K (cid:0)1)L (cid:20) Ak(t;(cid:28))
InternetDraft,URL:draft-thaler-multipath-05.txt,Feb2000.
[20] J.TsitsiklisandD.Bertsekas.DistributedAsynchronousOptimalRouting
inDataNetworks. IEEETrans.AutomaticControl,31:325–331,1986.
[21] C.Villamizar.OSPFOptimizedMultipath.InternetDraft,URL:draft-ietf-
ospf-omp-00.txt,March1998.
[22] S.Vutukury andJ.J.Garcia-Luna-Aceves. ASimple Approximation to
MinimumDelayRouting.Proc.ofACMSIGCOMM,Sept.1999.
[23] Y.WangandZ.Wang. ExplicitRoutingAlgorithmsforInternetTraffic
Engineerin. Proc.ofICCCN,pages582–588,1999.