MPLS and Traffic Engineering in

IP Networks
Daniel 0.Awduche, UUNET (MCI Worldcorn)

ABSTRACT Rapid growth and increasing requirements for Service quality,

reliability,
. ’’ and cfficiency have made traffic engineering an cssen-
tial consideration in the design and operation of large public Internet backbone networks.
Internet traffic engineering addresses the issue of performance optimization of operational
networks. A paramount objective nf Internet traffic engineering is t o facilitate the trans-
port of IP traffic through a given network in the inost efficient, reliable, and expeditious
manner possible. Historically, traffic engineering in the Internet has been hampered by the
limited functional capabilities of conventional IP technologies. Recent developments in
multiprotoral label switching and differentiated services have opened up new possibilities ational iictwr,rks. I t cncoinpasscs l l i c
t o address some of the limitations of thc conventional technologies. This article discusses at~plication
.. o l tccliiiologq mil scientific
thc appiicatians of MPLS to traffic engineering in IP networks. principles t n tlic n i c ~ ~ ~ i ~ s e ~1iivdc1- i~eiit,
iog, ch;iraclcrizalioii, ;incl c i i i i t r n l v f
I i i t c r i i c l trtdfic 121. I t also iiicludcs thc
iipplicelion of knnwlcdgc aiid lccliiiiqucs to acliicve specilic
i i t c r i i c t g r v w t h i n rcccrit tiincs h a s Ihcc~ivery pcrfiirinaiicc ohjcctivcs, i n c l u d i n g reliable and c x p c d i t i i ~ u s
impressive. t\ r e p o r t from the U.S. Dcparlinciil inovciiiciit cif traffic llirougb the nctwork, cllicicnt iitiliz;ilion
‘cc 111 suggests that the rtitc at which the Interncl OS nctwiirk rcsourccs, and planning o f iictwork capzicity. U l t i -
Iias been adoptcd has surpassed all otlicr tecluiokigics prcccrl- mately, good traffic cnginccring incrc;iscs tlic value o l a iict-
iiig it, iocludiog rodio, television, tind tlic personal cornputcr. wosk t u both t h e s c r v i c c provider and t h e Iiiterott user
Today, thc Internet has Iiccomc a couvenicnl ;ind cost-cffcc- coinmiinily.
tive i n c i l i u m fnr ciillaliiiratioii, cducalioii, electronic cnm- Tiislorically, effective traffic eiiginccring h a s liceii dillicult
nicrcc, ;ind cntcrtainmcnl. A common coilsenstis i s that Llic ti1 acliicvc i n public 1P nclworks. ‘l’hc rciisnii lor t l i i s i s the
internet w i l l mclamorphosc into ii m e d i u m for the c o ~ i v c r - limitcd fiinctiimal capalilitics o f cv~ivciitioiialIP tcclinologics.
gcncc cif voicc, video, and data con~mi~nic;itiiins. Altliiiugli tlic O n e parliculxr sliorlcotning of convciitional I P syslcins i s tlic
long-term market 11chavivr cif tlic Internet i s iliflicult t i i fvrc- iiiadequ;icy of mcasurcmcnl functions. For cxainplc, ii lniffic
cast, liitcrnct trarfic i s clc;irly growing in a geomelric pmgrcs- matrix, wliicli i s a hasic il;ita set ncciled for traffic cnginccr-
s i o n Reported compounded traffic griiwtli ralcs raogc f r v m ing, i s d i f l i c u l l to cstiinatc f r o m i n l c r l a c c statistics 1111 11’
two to ten times pcr ~ i i n u ~ n . routers. T h e limi1;itions V C inLrxlom;iin routing control Iunc-
Largc Internet service providers (ISPs) IIWC rcspoiidcd io tioiis arc iiiivtlicr issue with coiivcntional Ir syslcnis. Interior
tlic cliallcngc cif intcriict growth Iiy cinploying tlircc cvmplr- galcwey IJKVtVL!Ok (IGPs), such ;is liitcrincdia~cSyshii-lnlcr-
incntarp technical instruments: nicdiatc System (1s-IS) and Open Shortest Path Pisst (OSI’F),
* Network architccturc ciimmonly used to roiitc traCfic williiii autoiio~iiniissystcnis in
* Capacity cxpaiisioii tlic Intcriict, arc Liil~ology-driveniind ciiiploy pcr-pickct pro-
* l‘raffic ciigiiiceriiig grcssivc ciiiiiicclion conlri,l. Each roulcr mikes inclcpcndcnt
N e t w o r k arcliileclurc dc;ils w i t h the abstract slrticture o l i c i i i s using a local iii~tiiiitiatioii01/I synclironizcd
tlic compiinciils or object ~ I R S S cifCtlic
S nelwork, routing iirca link s l a k d;it;ib;isc. R o u t e sclcctiun i s Iiascd i i i i
tlicir fuiiclions, and the rcletionships bctwccn tlicm. A good, shortcst p;ith cnnipuLations using simple ;idditivc link tnclrics.
sc;il;iblc network arcliitcctorc, premised 011 solinil architcc- This ;ippro;ich i s highly dislriliulcd i i n d scalal~lc,but Ilawcd.
Lural principles, is inipcrativc i o l h c quickly cvolving I n t c m c t The Slaw i s (het thcsc protocols do noi consiilcr the cliiiractcr-
ciiviroiiniciit. islics of ofFcrccl ir;iCTic ;incl nctwork c;ip;icity cotislraiiils wlicn
Tlic second instrumcnt cmploycd hy large ISPs to rcspvnd to making routing decisions. T h i s results in suhscls uf n c t w v r k
traffic griiwlli i s rapid expansion cif c a p x i t y and network iiilr:i- rcsoi~rccsbccoming congcsicd, wliilc i i ~ h c rcsvi~rccs r along
structure. In 1996 naisl large ISPs in Ihc Unitcd Skilcs iipcralcd sltclnalc ptitlia rcmain uitdcrutili.,,cd 121. ‘
backboncs with I)S3 (44.736 Mliis) linlcs. In I997 and 1998, O G tion tiroblcm is ii symptinii 01 [mir rcsoiir
12c (622 Mh/s) links bcc;iinc pcrvasivc. In 1999 ii iiiiiiihcr cif i i n issue t h a t trziffic e n g i n e e r i n g spccilic;illy ;ittcmpLs Lo
in;ijor lSPs upgraded to OC:-48c (2.488 Gbis) links. I3y the year rcdrcss.
21100, sonic lSPs cxpccl Lo hcgin clcploynicnt of IP liacklmnci l i c c c n t clcvclopinciils i n m i i l t i p r v t i i c ~ i l a l i c l swilcliing
willi OC-I92c (9.953 (;his) links, priivisiimcd ilircctly w c r dcnsc (MPLS) [2-8] iipcn iicw piissiibilitics to iiddrcss sonic 111 LIic
w;ivelcngtli-clivision mulliplexing (I)WI)M) fiicililics. limitatiiins vf 1P syslcms conccriiiiig traffic cngiiiccring. A
T l i c l h i r d i n s l r u m c n t c n i p l o y c d hy scrvicc p r v v i d c r s l r i framcwork for MI’LS i s prcscntcd i n IS] mid iiii arcliitcctorc
iiddrcss tlic lnlernct grnwth cliallciigc i s traffic ciiginccriiig. lor i t dcscribcd i n [ZJ. ‘ l l i c rcquirciiiciils lnr traffic coginccring
This suhjccl has a1tr;ictcd significaiil allention in rccciit times over MPLS wcrc iirticiilalcd iii 121.Although MPLS is ii rc1:i-
!2-01. A molivalirin for Intcrncl L d l i c engineering is tlic rcal- lively simple tcclinolngy (b;iscil mi Llic classic;il liilicl swapping
i z a l i o i i that a r c h i l c c l u r a l pzirticligms and s i m p l e capacity 1ixadigni), it ciiol~lesLlic introductiiin cif si~pliislicalcdcontrol
cxpansion arc ~~cccssary, hi11 not siifficicnt, to ilelivcr liigli- capabilities t l u t advance (lie traffic ciigiiiccriiig (unction in 11’
qmility Interncl scrvicc ~ i i i d c all r circumkinccs. Intcrnci tmI- nclworks [2-4, 6, 71. A particularly interesting u p c c t ol M1’1.S
l i e ciigiiiccriiig i s tlic q c c t cif l i i l c r n c l network cnginccring i s Ihal it cllicicntly supporls origination coiiiicccion control
that ;iddrcsscs the issue o f pcrlorinancc optiiniz;ition of upcr. ~ h r o u g hexplicit labcl-swilchcd paths. Whcn MI’1.S i s tun-
~~~ ~
 Formulation of
hincd witli dillcrciiti;ilcd scrviccs iind
i control volicv
ciriistl-aint-l,ase[l rouiiiig, lhcy bccunic
~"'"cr1u1 illld c~liilplcmclit;ily alrslrac-
lioiis l o r quality iilscrvicc (UoS) provi- Performance monitors
network state Fault monitors
sioiiiiig iii 11' iictworks. Analytical models
1
11' ~iclwiirka,focusing spccilic;illy o n
scrvicc Imividcr networks. 'I'lic h;isic
conccpts atid clidlciigcs o l traffic ciigi-
nccring iii tlic Internet arc intruduccd
firsl. Tlicsc conccpts iiiid cliallcnges arc ~ .~control
-~
~

followcil h y t11c cqxihilitics i h i l m;ikc

M P I S ;ipplical,lc to lrallic cnginccring
i n such cnvironmcnts. A icvicw of lhc
llvcrlay IIIclllll'llllogy tIl2lt w a s ,,sed lor
t n i l l i c ciiginccring iii classical 11' ICI-
works (prior io llic iiilvc111cif MI'IS) i s -
i i l s o p r o v i d c d . ' l l i i s article covcrs
A
iiitnidmiiiiiii trallic ciiginccriiig, i h t is, performance
t r a l l i c c i i g i n c c r i n g w i t l i i i i ii siiiglc
iiuti~iioiiio~ r y~ssl c i i i in the Iiitcriict.
The rciiiiiiiidcr 01this iiiticlc is org;i-
iiizcd iis lolluws. The frrllowitig sccliiiii
inlroduccs the Iiiisic coiicctiis iiiitl chiil-
Icngcs c i f trziflic c i i g i n c c r i n g i n i h c
Intcrnci. Wc tlicii dcscrihc tlic Ii~iiciiiiiiiilcapibililics m i k i n g applying c o n l r d x l i i i n s , iliicc iry. lo drive tlic iiclwork lo
h4I'I.S q q i l i c ~ i h kto traffic cngiiiccring iii II' networks. 'I'hc II ilcsircil slate ;icciirdiiip to Lhc i l r d policy. ~ ~ o n t r;iclions
ol
Iiisl section coiikiiiis tlic concluding rciiiarkb. m a y iiiviilvc m r d i l y i n g (11 Irclaxing network resource ciiii-
straints (e.g., augnicnting ciipxity), iii;iiiipulating triillic ~ i i i i i i -
;iqcinenl pir:inictcrs, or mirdifying llic p;il-;iiiiclcrs iissociiilcd
THE CONCEPTS AND CHALLENGE OF with riiutiiig through 21 ctinfigiiliitioii ciiiitrol systciii.
1r;iflic angioccriiig i s iiii ;id;iptivc priiccss. 1 1 1 ~ .riItlr stages
TRAFFICENGINEERING IN IP NETWORKS of ilic process iiiodcl dclincd iihovc arc iterated. 111 a n opcrn-
This scctiori iiilroduccs tlic conccpts ; i d practiciil functions of t i o n d conlcnl, i t i s hcst LO iiiiiiiiiiizc the lcvcl of manual intcr-
traffic cnginccriiig iii npcralioxil 11' iiclworks. Tlic cli;illciigc vcolbin ilivolvcd i n triillic cnginccring hy automaling the tasks
of traffic cnginccring iii i i u t m o m i i i s syslcnu within ihc Inter- wliciicvcr IJoSSiblC. Tlic l r i l l l i c cnginccring pr<>
iicl is liighliglitcd, ;ind a n overview ol the cl;issic;il 11' ~ i v c r illustratcil in Vig. I.
ATM iivcrlay iiiodcl is provided.
Traffic Engineering Objectives - A Iiractic;il luiicliiin o f
INTERNET TRAFFICENGINEERINGCONCEPTS l r x f f i c ciiginccring ill II' i i c t w m k s i s tlic tiiiipping of t r a l l i c
In conccpl, ii network coiisists of ii clcinatid system (tfiillic), a onto llic network iiiirastruclurc to iicliicvc specific perlor-
constraint system (intcrc~iniicctcdiiclwork clcniciits), mil ii iiiancc objcctivcs. H i g h scrvicc q d i t y , clliciancy, survivabili-
rcsponsc systciii (nclwnrk p r o l i ~ o l swid ~proccsscs).Traffic ty, imd cctiiuiiiiy arc crucial ohjcctivcs in loday's coinnicrcial,
lishcs the parameters and operating points compclilivc, i i i i d missioii-crilic;il Iiitcrnct. T r x l l i c cngiiiccr-
ts cif tlic nclwiirk in iin ,ipcr:itioiial contcxl. i i i g r r q u i r c s precise c ~ i i t r o over
l tlic r o u t i n g f u n c l i o n t o
(:oriscquently, Inlcrnct traffic cnginccring i s liiiidaniciitally ;I acliicvc tlic ubjcclivcs. Indccd, i i n csscnlial rcquircnicnt irir
contrr,l problcm 121. traffic ciigiiiccriiig i n IF networks i s tlic c;ipahility ttr coin-
piitc and c s l ; i h l i s l i ii f o r w a r d i n g p a t h f r o i n O I I C node t o
The Traffic Engineering Process Model - A tiumber or aiollicr. l'liis IxilIi i i i i i s i lullill smic rcqiiircincnts, wliilc iilsu
stiigcs can l i e idcntificd in the Iiitcrnct traflic cnginccring prii- satisfying nelwiirk c;ipacily alid policy coiistr;iints. (iciicrelly,
cess iiiiiilcl. 'The lirsl stage i s llic formulalim of a ctintrol poli- ,jcctivcs ciiii I i c l m J ~ i ~ ; i c o r i e r r t c;indlor
d
cy. The control p d i c y d c p c ~ i don ~ the iictwiirk coiilcxt, cost
struetiire, revciiiic or iiiilily iiiiidcl, operating constraints, iiiicl cd pcrlormiiocc oli.jcclivcs rclelc t o t h e
success criteria. l'hc secoiid s t q c i s tlic ihscrvation o l llic iniprovciriciit of the UuS prwisioocd 10 Internet lraflic. Traf-
network stale ilirough a set of m i n i l o r i n g functions. T h i s i s fic-oricntcd pcrloriiiaiicc mctrics include ~ i ~ i c kloss, c t delay,
the fccdlxick compoiiciil nf tlic traffic cnginccring process delay v;iriatiiin, and gocidpul. 'She efleclivcncss of tsalfic-iiri-
modcl. It may incliidc preprocessing activities siicli iis <lata cntcd prrlicics ciiii ; i h he iiiciisurcd i n terms of tlic relative
r c d u c t i o n xod data tr;inslormatioti. l'lic t h i r d stage i s the p r i i p o r l i i i i i cif i i l l c r c d traEiic ;ichicving tlicir pcrtormtincc
cliaractcri~.aliliiiof trnlfic a n d eniilysis of tlic inclwiirk slalc. rcquircnicnts. Wlicii scrvicc Icvcl ;igrcciiicnls (SI .As) a r c
Various qiialitativc iiiid qii;iotit;itivc tcchniqucs ciiii be applicd involvcd, prntccting trzillic streams LIIiit camp
i n llic c1i;iractcrizatiiiii ;ind analysis stage. llottlciiccks i i i i d SLAs from lllosc that arc iioncompliaiit I b c c ~ m
pathologies lliat impede (or potcnlidly iiilpcdc) Iictwork per- l m t kiclor i o the ziltainmcnl 01 lr;iffic-oriciitcd perkirmaiicc
furimiiicc iirc idcntificd. Tlic results iirc ~ i s c dfor network per- ubjcctivcs. l(csonrcc-oriciitcd pcrlorin;incc ohjcctivcs rcliitc ttr
liirniancc oplimizaliiiii, nclworli iipcr:itioiis cii~itrol,netwiirk tlic optitiiiztilioii cif tlic oli1iz;itiirn o l iictwork asscts. Ollicicnt
design, a n d capacity pl;ioiiiiig. Tlic fiiorlli stage i s tlic opti- rcsii~irccalloc;itioii i s l l i c hasic tipprmicli ti, scc~ircrcsuurcc-
mization cif iictwork pcrformaiicc. T h i s i s accoiiiplislied b y riricntcd pcrlnrinaocc objectives. A lI.alCic ciigiiiccriiig systcni
 is said to be “rational” if it addrcsscs traffic-oricntcd pcrfnr- rclativcly short intcrvals r i f tinic. This rqiid gr<iwtliis tlicii lol-
iniincc prohlcms wliilc s i m u l t a i i c o u s l y ulilixiiig n e t w o r k lowed b y inodcst griiwtli over rclativcly longer intcrv;ils vf
resources cfficicntly. time. Accuralc forcc;isting is thcrcBirc quite difficult. Furtlicr-
Minimizing congestion is a ccntral goal of traffic cnginccring. more, l i i t c r n e l traflic cxliihits very dynamic hchavior with
Congestion typically manifests iindcr the following sccnariiis: characlcrislics 1h;it iirc not yet well undcrslood. Traflic also
* When network rcsonrces arc insufficient or inadcquatc to tends to bc liiglily asyminctric.
liandlc olfcrcd load T h e operating covironnicot is also in a coiitiniuil state of
When trafric is incfficicnlly mappcd onlo rcsoiirccs, cBns- flux. New rcsnurccs iirc addcd const;intly. Resources also fail
iiig snlisets of resources to bccomc ovcrutilizcd wliilc regularly. New Inlcriict tippliceticrns with b;indwirltli require-
others rcmnin underutilized 121 inicnts wliicli may have signiiicant glah;il impact arc intro-
Congestion problems resulting f r n m insufficient o r in;iilc- iluccd all tlic time. Facility location is i i l s o iiii issue. Soinclinics
quatc rcsnurccs can he addressed Iiy: augmenting network nctw~iskrcsvurccs arc sited i n less th;m idc;il lncativiis duc to
e x p a c i t y , o r m o d u l a t i n g , c o n d i t i o n i n g , o r t h r o t t l i n g tlic f.. u.”
l i t y constr;iints. Additinnal coinplications iirc introducccl
dcmmd so th;it tlic traffic fits onto the available capicity (c.g., by intcrdomain traffic traversing atitotiiiinoiis systems’ boiinil-
using policing, flow c ~ i n t r o l ,rate shaping, link schciluling, arics. These environmcntel filetors result in tlic network topol-
qoeur inzinagcmcnt, tariSfs, et al.). C o n g e s t i o n problcms ogy not usually corrcl;iting with tlic lrallic inatria. Addressing
resulting from incfficicnt mapping of tnilfic o n h rcsonrccs thcsc issues rcquircs contino;il monitoring ;ind performance
can he addrcsscd by incrc;ising tlic cfficicncy of rcs~iiirccallo- o p t i m i d o n (IS pnhlic 11’ nctwvrks.
czitiiin. An cxemplc of this incrc;iscd cfficicncy of rcsoiircc
allvcatioii would l i e to route snnic traflic tiway from congested TRAFFICENGINEERINGWITH THE
rcsourccs to rclativcly undcrutilizcd ones. CLASSICAL OVERLAY MODEL
Reliable network opcr;ition is anotlicr impiirtant vlijcctivc
of Internet trafSic enginccring. Multiple fiiilurc rccovcry scc- Tlic overlay modcl is a tcchniquc t h a t was applied, prior to
narios must h e devised to ensure continuity of service follow- MPLS, t n circumvent sonic of the 1imit;itiiins o1 IP systcms
ing n c t w o r k i m p a i r m e n t s . A d c q u a t c c a p a c i t y f o r service regarding traflic cngioccring. Tlic basic idea is to introduce ;I
restoration mnst be provisioncd, thcrcfiirc, and tlic operational sccondary tcchnology, witli virtual circuit ;itid traffic manage-
capability must exist to cxpcditiously rcrootc traffic througli ment capaliilitics (e.g., ;isynclironoiis translcr ~norlc,ATM),
the rcdnndant capacity when faults occur. Rcoptiinization may into tlic II’ inlrastructurc i n an ovcr1;iy ciinfiguratioii. ‘Shc vir-
bc rcqiiircd fnllnwing restoration to inakc more effective iisc toal circuits d tlie sccoiichry tcchnology serve as pvint-lo-
of the residod p~ist-Eaultcapacity. It may Iic advantageous tii p i n t links hctwccn II’ riiulcrs. I’igore 2 illustr;itcs thc ovcrl;
utilize subsets o l the rci1und;int capacity t o impriivc nct-
work pcrforni;ince arid ellicicncy wlicn the nctwirk is fiiult-
rrcc.
T r a f f i c cnginccring liccomcs cvcii more criticzil in a
miilticlass service cnvironmcnt like the cincrgiiig rlillercnti-
atcd scrviccs Internet, whcrc traffic strc;ims with dilTcrciit
scrvicc rcqiiiremcnts a r e i n contention f o r n c t w o r k
resoiirccs. In tlicsc cnvironoiciits, triilCic enginccring Adaptation layer

.
lishcs tlic rcsn rcc sh;iring p;ir;inictcrs s o that tlic network ~~

ntiiil trc;itmcnt lo sonic service cliisscs in DWDM

accordancc with a utility modcl. ~ ~. network
(a) (b) (C)
THE CHALLENGE OF INTERNET TRAFFIC
ENGINEERING
Traffic cnginccring i n convcntional 1P nclworka is a c h a - Time
lcnging prohlcm. Singularitics and discontinuitics charac-
t c r i z c Iiiternct growth. Very r a p i d g r o w t h occnrs over
 __---
iiii,dcl \villi A'rM swilclics ill l l i c corc surrviindcd hy a n "pi-
dcrinis o l IF riiulcrs logically intcrcviincctcd by ATM pcriiiii-
iicul virlual coniicctiviis (PVCs).
The ~ivcrlay;ipprvach calcncls llic design spice and dlows Packet aver
SONET links
arbitrary viriuiil lnpolrigics to Ihc clcliiicd i i i i d supcrimposcd
onto the pliysicxl iictwvrk k i p l n g y . Tlic overlay iiiodcl iilso
pcrmils cstiiiiiitim of a rudiiiicnlirry t r a f k iiiiitrix frirm statis-
tics on tlic PVCs that intercvniicct r,iiitcrs. 'l'raffic ciiginccr-
ing can a l s o iiiiivc lraific lrviii overloaded links til rcl;itivcly I
undcrulilizcd links hy chiingiiig t h e dcsignalcd lraosit lisls
(VI'LS) nf il sullscl of I'VCS.
Tlicrc arc I'undii~iiciil;rldriiwl~;icksto the 1P over A I ' M
iivcrlay model. Perhaps t h e m o s t sigiiificaiil problem is the
iiccd to I)uild iiiid man;igc ~ w o iictwtrrhs willi dissimilar tccli- ( I S R ) . I:cwcr network clciiicnls arc rcqiiircd t h n willi over-
iii1ii101 a l s o incrciiscs tlic crmplcxiiy of lay iiltcrnativcs, rcliahility is increased, a n d opcriiliiig cnsts
iiiid iiclwork ilcsign. Ikliahilily is a l s o ii and qiicuiiig dclays arc ~cduccd.Addilionally, MI'I .S silnpli-
cniiccrii hecause rniirc iictwork c l c i i i c i i l s iiow exist i n scrics fics network arcliitcctiirc and iictwiirk design rcliitivc to the
nii the rviitcd p;illi. Scalability is anolhcr issiic Iicciiiisc ilic ovcrley iiiiiilcl. MI'LS couplcd willi diffcrcnlialcd scrviccs iiiiil
iiuiriber 11s adjaceiicics i n Lhc iivcrliiy graph gcocfiilly iiicrciiscs constraint-liascil routing lundiimciitelly chiiiigcs llrc way cure
qiiiidraticiilly with the iiiiinbcr ( 1 1 roulcrs, Ilicrcl)y inercuing I P networks arc designed i i n d Lriiflic cnginccrcd. Figure 4
the CPU a i i d network rcsoiircc ciiiisuinplioo i~ssiiciailcclwith depicts ii siinplificd topolvgy L l i a t results when I . S l l s that arc
rouling. Other issiics include llic qiiiiiitiaiition and c i i c q i s u l a - cqiiippcd with packcl ~ i v c rSONTI' intcrfaccs rcpl;icc llic
lion ovcrhciid iwocialcd with ATM, tlic lcchnical dil'ficullics iivcrlay network previously rhnwii iii I'ig. 2.
iiihcrcul in dcvclopiiig iiiicroclcctrmics Lhet perform scgincn- O n e l c n t u r c vf traffic ciigincering i i i II' n e t w o r k is tlic
Lalion aiid rc;mcinbly (SAR) 211 very high spccils (OC-48c inid triift'ic trunk concept. A triillic trunk is an aggrcgilinn 01 traf-
aliovc), iiiid t h e possiliility of routing iiislaliility in t h e IF (ic helonging 111 Lhc same class 121.11is cssciitially iiii abstract
dviiiiiiii iiiduccd by mulliplc PVCs hilurcs t~~llvwiiig ii single dcscriptioo ol traffic thal allows certain iittTibiitcs iif tlic traf-
iiitcrswilch link iiiip;iirirmnl iii the A'SM corc. lic to Iic 1i;ir'airictcrizcd. It is indcpcndcnt i l l tlic iiiiderlyiiig
Tlic trciid, IIICIC~O~C, is lo cvolvc core 11' networks ;iw;iy tccliiiologics. U i c ~~riihlciii of iiiappjng traffic trunks vntn ii
frvni llic ~ivcrlayi i i ~ ~ c liiiid
c l towarcl more iiilcgralcrl solutions. givcii network topilogy is ii central issiic 01' triillic ciigiiiccr-
This c v n l u t i o n is now possible hccausc iif dcvclrrpmcnls i o ing. I n MPLS nclwol-lis, tral'lic trunks arc iiiappcd onlo the
MPLS ancl rccciit atlviiiicc~in IiigIi~pcrC~irmiiiicc gigihitilcrabil nctwvl-k topology tlirough lhc sclccliiiii of routes lor explicit
riiutcrs i i i i d nptical iiitcrrictworlking ayatciiis. Pigurc 3 presents I S P s . Tlic tcrins I.SP limnel 131 and troJic riigineering t~nirrel
the cxpcclcd Lcclinvl~igyIiiycr cvnlulioii VI cnrc 11' nclwvrks (le-tunnel) 191 arc commonly used to rcfcr t u ilic ciiml~ination
from [I' over ATM over ayirclirouiius optical network (SONI'I) nf tralfic trunk ;ind explicit ISPs in MI'I .S.
over fiber 111 1P with MPLS nver SONHT over IIWIIM; iiiid, I Sl' tiiiiricls iilkiw tlic pcrlorin;iiicc of iiii iipcralivniil nctwvrk
I' with M P I 3 iivcr iiii adaplation I;iycr intcrfiiciiig 10 be opLiinizcd in varioiis wigs. For cx;unplc, if congestion p n b
itilc npticiil tnuisporl network (01"). lciiis ci~iisctlby rulioptinlnl routing arc dctcctcd, LSI' tunncls can
hc rcroiacd i o alleviate lhc problem. LSI' tiiniicls can he pwamc-
tcrircd, and network resources call bc iillocatcd 111tlicni h i s c d mi
MPLS AND TRAFFICENGINEERINGIN those pilamctcrs. Multiple LSP tunnels can be crciitcd bctwccn
twv i i ~ l c sand
, the traffic bctwccn thc ~ivclcsdividctl iimvng lhc
IP NETWORKS tunnels accrirdiiig til sonic l v c d policy. LSI' lumncls pcrinit l l i c
'l'hc Iiiiiclioiiiil capiihilitics inakiiig MI'LS iillriictivc f i r lraffic iiitrvductivn o l flcxiblc iiiid crrsl-effective surviviihility optiiim.
engineering iii I P iiclworks iirc dcscril)cd i n this section. Ciciicr- Stzitistics dcrivcd lrom LSI' lunricls can be used lo c m s t r i i c t ii
211 discussinns [if MI'LS lcchnoliigy itself arc
dctiiilcd i n n number of documciits from tlic
IIITTL'MPLS wnrking group [Z,5 , XI.
MPLS allows sophislicatcd riiuliiig coiitrvl
cap;ihilitics lo bc introduced into I P networks.
'Shcsc capabililics iirc buttressed oii llic fact
that MI'LS clliciciilly. siii>rmrIs .. orilriiitilinn
I

ctiiincclinn cniitrol tlirough explicit l a l i c l -

switched pallis (ISPs). An explicit LSP is one
wlivsc rvulc is dclcrmiiiccl 211 lhc origination
iiiidc. Origiiiativn cvmiection cnntrol pcrinils
explicit r v u t c s ti1 h c established wliicli arc
iiidcpciidcnt v i tlic ~lcstiniition-h;iscdTI' stivrl-
CSI path routing inodcl. Oiicc an explicit riiiiic
is dclcrmiiied, ii signaling p r o t o c o l is lhcn
u s e d lv insl;ill tlic ISI'. Tlirrrugli explicit LSI's,
a qiiiisi circuit swilcliing cap;hility is siipcrini-
poscd o i i t h c 11' rouliiig modcl [2].W h e n
d c p l o y c d i n 11' (ivcr SONLlT n r I P vvcr
UWDM c m I i g i i d i i i i s , llic tniditional 1.3 a n d
L2 I'iiiictioiis iirc virlualizcd in vile network
clcniciit callccl l l i c I i i l i c l swilchiiig rciiitcr
rudiiiicntaiy trarlic matrix. I'igiirc 5 sIi(iws liow LSP tunocls can iiicnlctl in ii inulliclass c ~ i v i r n ~ ~ ~using
i i c n t Iircciiiption. Rcsiliencc
l i e used t i i rcdistrihutc traffic lo add ct~ngcstioiiprohlciiis iittrilwlcs specify tlic response of im LSP tomicl ti, impairmciits
caused by shortest pnlh IGPs. t h t impict i t s route. A hasic rcsilicncc atlributc specifics whcilicr
a n LSP tunnel i s to be uilnini~iiczillyrcroutcil Sollowing f;iults
COMPONENTS OF THE MPLS TRAFFIC
ENGINEERINGMODEL along its cs1;iblislicd pili.Exlcmlcd icsiliciicc atirihiitcs iist,rpo-
An M P L S lrallic engineering mndcl consists of the following ratc morc sophislicatcd rccovcry policies, iiicludiog p l i c i c s 111;it
basic l'unctinnel cti~nponcnts: inst;intialc moltiplc pir:ilIcl I.SP tiimicls together with riilcs tir

-
* I'iitli in;io;igcmcnl
'Traffic assignment
* Network stiitc inliiIm;ition disscniiiiiilion
dclcrniinc tlicir rcl;llivc prcfcrcncc ~indcrS,iolts. Ilcsoul-cc class
;iffiiiilg xttributcs iiiqmsc additionel policy icslriclioiis o n tlic
qudificiition of sets ol rcsourccs lor IS'I' tuiirml path sclcctirro.
* Network inaiuigcniciil An affinity rclatioiisliip l ~ ~ t w c ciiiii i I S ' lunncl a n d a rcsniircc
Thcsc arc elements nf the MPLS control plane and arc disliiict CIBSS indicates wlictlicr tlic icsourcc cliiss is tu bc includcd o r
from Ilic tiiwardiiig pkmc. cxclodcd from IIMpath ol ihc IS P tumicl.
Resource atlribiitcs dcSinc additional priipcriics o f net-
Path M a n a g e m e n t - P;lth nian;igcmciil concerns all work rcsnurccs tbal further constraiii tlic r w t i n g 0l LSI' ton-
aspects rclatcd to tlic sclcctioii ol' explicit roulcs, iiiiil Llic iicls through i h c i n I<csourcc xtlrihulcs iiicludc the iniixiiiiiim
iiistontiatiation ancl inaintcnaiicc of LSP iunocls. A path niaii- ;illr~c;ition inulliplicr (MAM), IIic ilclaull traffic engineering
agcmcnt policy defines l h c path scIcclioii criteria as well a s metric, iind resourcc cliiss iittril~ulcs.Thc MAM conccpl i s
rules f o r sustaining alrcady cst;ililishcd LSP t u n ~ i c l s121. I'alli ;io;ilogous t n siihscripiiuii iind rivcrhii~ikingfactors i n h m c
managcmcnt coiisists nl' 1111.c~ primary funclions: p i t h selec- relay and ATM iiclworks. The clefnull tr;ilfic cugiiiccring mcl-
tinn. path pkiccincnl, i i i i c l pith m a i i i l c o w x rics can Ihe used 10 cstal~lishriiutc o ~ ~ t i ~ n i ~ crilcria , ~ ~ t i i lor
m
'llic p;ith sclecfioii fuiictioii spcciEics the explicit rootc Cor I.SP twmcls iitdepcndcnl o f IGP mcltics.
an LSP lunncl at tlic i i r i g i n n t i i i n node cif t l i c l u n i i c l . An xttribuics arc iiscd t r i c;itcgorix scsoiirccs, pri-
explicit roulc c a n h e rclircscnlcd 21s 21 scqucricc o l bops or ii inarily links, into rlilfcrc[it ckisscs. UiiiC~iImpulicics, siicli iis iiiclii-
scqiiciicc nl' a b s l r a c t nodes. 11 may contain both strict a n d s i o n i i n d cxclusicm, can tlicii hc applied LO c;ich rcsourcc clitss
1oosc suliscts. An abstract ~ i o d ci s ti group of ~iodcsw l i ~ i s c with respect to LSI' tiiiiiicl piitli selection. A I
internal lopiiliigy is iip;iquc to tlic origiii;ition iiodc. Explicil iniurc th;m one rcsoiircc class. l l i c rcsourcc cla
routes can hc dc1Iiicd ;idministr;ilivcly or com~nitcdautom;iti- of tlic link al;itc pammctcrs. Rcsnurcc cliiss iitlr
cally by a constrainl-h;iscd routing ciitily. In coiitiiin 1r;iffic williin spccilic 1opokigic;il rcgioiis of ii network.
Constraiiit-li;isc~Irtiuting i s a gcncr~ilizalii~ii (IC00s routing. I t
i s tiscd to compute routcs that sxtisfi a scl uE rcqoircmcncs, EL^,- Traffic Assignment - Ti;iIlic must Ihc assigned t u iiii [.SI'
jcct to ciiiistr;iints imposcd by the iictwork a i d ndministrativc lunncl 011cc the t i i n n e l is cstahlishcd. l'niffic assignincot coii-
policics. Constraint-based routing reduces tlic lcvcl nf maiiud ccriis all aspects rclatcil to l l i c allocalion (11lraffic to cslab-
inlctvcntion iiivulvcd iii traffic cngiiiccriiig. lislied I . S P ii!iincIs. IL cniisists of ti pariitioning Sunction and
Ilic scairtd conipniient of pith management i s the path 11lacc- an apportiommcnt function. Thc pzirlilioning S i i n ~ t i i mparli-
mcnt funclion. 'L'liis is iiscd f n instxntiatc I S ' timiicls usiiig n sig- t i o i l s ingress trall'ic according t o snnic principle (iL division.
iialing protocol, wliicli d s n scivcs iis ii 1al)cl clislrihution prati~col. The apportiorumcnt function allots the 1xulitioncd lraffic to
Two MPLS sigtuiling prot(icnls arc currently ileEincd: Resource cst;ihlislicd 1,SP timncls according to siiiiic principle nl' allocii-
Rescrvatiim Protoctil (RSVP) cxlciisions [3, 41 aiid constraint- lioii. Tlic po1cnti;il flexibility in tniffic ;issignmciil r u u h n c n -
based routccl LDP (CR-I.DP) [6J. Tlic lliird cornpoiictit o f pal11 tally distinguislics MPLS Frirm ATM.
management i s tlic path miiiiitciiancc fimcliiin, which sustains Otic way ti1 ;iutomatc tlic traflic iissigiimcnt prolilcin i s to
and tcriniiuitcs alrcady cstabiishcil U P hmiicls. view I S P tuniicls iis shorlculs thrniigli thc IGl' domaiii [OI.
A set of allrihutcs can he associated with LSI' timiicls mid ncl- Addilion;il altrihiitcs inay be introd?iccd lo control Ilic assign-
work rcsoiirccs l o guide the p:ith inanzigcincnl functinns ;ind to ment lunctioii wlicn there arc niulliplc paths tii a given niidc.
priwklc controls over ciinstraint-hased routing. An i n i p o r t u i t Filtration riilcs may bc applied to restrict the class of lraffic
opcratioiial requirement i s ilic capability tn manipnlaic tlic mapped oiitc ii given I S P luiincl. F i l t r i i t i o i i rolcs may, for
attributes nf active LSP tuuncls to ciiiisc cerlain transitions (e.g., cxamplc, l i e i i s c d io clcfinc Lhc way d i f f c r c i i l i a t c d scrviccs
explicit route clitingcs) to occ~irgracefully without advcrscly heliavior aggrcg;itcs arc miippcd ontn LSF tunnels.
i m p s l i n g network opcnitions. Lotid distribulioii acrnss inulliplc LSP tunnels Iietwccn two
LSP tunnel atlrilxitcs include traflic parameters, adaptivity ~iodcsi s an importtint traffic assignment issue. The load distri-
attributes, prinrity altribules, prccniption atlributcs, resilience h u t i i i i i psiihlcin ciin he ndclrcsscd by implicitly or explicitly
attributes, rcsn~irccclass all'inity atlributcs, and otlicr policy assigning wciglils l o each LSI' tiiiiiicl iind appi~rtioninglraffic
options siicli as policing attributes 121. Traffic parameters specify in relative propiirtinii 111 tlic weights. I.o;id distributirin across
the bandwidtli characteristics nf [lie LSI' tonncl, aiid may include p r a l l c l LSP tunnels ciiii also h e implcmcntcd as a fccdhack
peak ratcs, mean riitcs and burst skcs - or llic parameters may funclion nf tlic st;itc o f tlic nclwwk.
simply spccify iin cficctivc bandwidth. l ' l i c adnptivity ;ittrihulcs
indicate ilic scnsilivity cif ao LSP tooncl to the dynamics of the Network State information Dissemination - Network
nctwork stale. Adaptive U P tunncls can bc rerouted automatical- statc inSonnation disscuiinaliirn cnnccr~isthe distribiiti~iiiof rcl-
ly when better routcs bccomc available. Non;vlnptivc ISP tumicls want topology stale information throiigli~iiiltlic MFLS di~iiiaiii.
arc pimicd to their cstablislied m i t e s cxccpt undcr faults. Priorily This i s ;~ccomplishcdhy cxlcruliiig comciilio1i;il 1(;Ps lo propn-
attributes impose a partial ordcr on multiple ISP timiiels, accord- gate ;~dditionalinCorin;ition about tlic stale (if tlic iiclwurk in
ing ti, wliich p i t h selection and palh placement arc scqucnccd. link state adverliscmciils 171. Tlic idditioiial iiifurmalioii dis-
Currently, ciglit setup priority levels arc specified [3]. Tlic prc- tributed incluclcs tnaximum link batidwidtli, miixiiiiiini alloca-
emption attributes clclerminc whether a iicw I S P tumicl can t i o n niulliplicr, d c h u l t t r a l l i c cngiiiecrii
acquire the rcsmrccs allocated to i i n cxisiing tunnel. Prccniption b;indwidtli per priority cl;iss, and rcsourcc
is implcmcntcd using a combinatirio of setup and rescivaiion pri- topology stntc iniorm;itiori i s used by constr;iini-hascd routing
orities 131. V;iriiius prinrilircd rcsto~iitiniiscliciiies can hc implc- entities to sclcct feasible riiulcs for LSI' tuiiiicls.
Network Management - N c l w o r k nr;in;igcrncnl is ;ill reviewed. 'I'lic iivcrlay mudcl is dcscrilicd. This ~ n i i d cis
l hascil
iinporlant ;ispeel crl trallic cnginccring over MI'LS. The sue- on 11' over ATM and is an altcrnativc to MI'IA'. l h a l l y , l l i c
cess 01 the MI'1.S a p ~ r n ~ i clol i triillic cnginccring cvcnlually functional capiihilitics making MI'I .S IISCIUIfnr trxllic ciigi-
dcpcnds on ihc case wiih which the nctwork c u i be discrvcd iiccriiig iii 11' inclworks iirc highliglitcd.
;mil conlrnllcd. Ciciicr;illy, a n MI'LS nctwork ~ n ; i ~ ~ a g c i n o i i t
REFERENCES

..
mctisuriog llic transit tiincs. liaciit i1otiCic;itions can be issiicd iirt draft. work in'prrigrers, Feb. 1919.
hen thc slate of n nianagcd MP1.S object cxcccds prcscribcd I71 T. ti, G. Swallow, and D. Awduihc, "IGP Rcqiiirementi fur l r d i f i c E n g i ~
neering with MPLS." IETF Internet draft, work in pmqrerr. Feb. 1999.
tlircsholds. Rulk rclricvd cif LSP tunnel 1r;iClic statislies can [RI E. Roren, A. Virwanathan, anti R. callon, "Multilmtocol Labcl Switchinq
I,c used lor time scrics nnelysis iind cepicily planning [JLIrplW Architccture," IETF Inteinct h f l , work in pragreir, July 1998.
cs. Ai1 olicrational rcqoircmcnt is lhc cap;ihilily tn lisl, at any I91 N. Shcn and H. Smit, "Calculating IGP Rautcr over Traffic Engiileering
given p i n t in time, iill the iiirdcs tl-;ivcrscd hy iin I.SP tuntiel, Tunnels," I E l F Intcmet draft, work in progrcir, June 1999.
;ind lor each nodc t o list ;ill 01 tlic r.sP tunncls iiriginaling
1'mm it, lcriiiintiling on it, iiiid traversing tlirougli it.
Uccausc o p h i z i i r g ilic pcrlormancc ot large-scale nclwnrks BIOGRAPHY
is an intractahlc priihlciii, nffline tr;dllc cogiivxring supporL lnols DANIEL0 . Awo~cilr[MI (awduche@!,uu.not) is the manager of advanced t o c b
iiiiiv hc rciiuirctl 11) iiiieiiicnt tlic online caixihilitics nf MPLS. noloqy a t U U N E l ~ ,811 MCI Worldcom Company, a glob.il provider of Illternet

IETF w h e r e h serves As ~ ' o ~ c h oof

i r the Internet lraffic Enqincerinq Working
CONCLUSION GTOUI). l l c i r a co-author uf s e ~ c i d lIETF documents un MPLS. He I S also
Tllis cliscusses lllc ~Il,,~lic.~lil,ns l l l l l ~ t ~ p ~ l i l o e l active
,~ in the Opticai Internctworking Forum. where lie s c r ~ editor of thc
"Optical lnternctwnrk Architecture: Fiinctionr and Reference Modclr" ~ O L L I -
hbcl switching to tK1ICic cnginccriiig i n 11' nclwnrks. 'I'hc Coil- mcnt.H~ is a memho of the ACM. bic did graduate rtudics it, calnplltcr r y r ~
ccpls ;Ind ch;illcngcs 01 lrlifilc enginccring ill tllc lntcrllct arc temi engineerinq at the University o i Mamchusetti, Amherrt.