IncrementalPlanningofMulti-layerElasticOpticalNetworks

P.Papanikolaou,K.Christodoulopoulos,E.VarvarigosSchoolofElectricalandComputerEngineering,NationalTechnicalUniversityofAthens,Greece

ONDM 201721th International Conference on Optical Network Design

and ModelingMay 15, 2017 Budapest, Hungary

RS1:ElasticOpticalNetworks

High Speed Communication Networks Laboratory

National Technical University of ATHENS

ONDM 2017

MotivationConventionalapproach:

ý optical transport networks have long upgrade cycles

ý optical network and its edges are upgraded independently

ý both result to overprovisioning, underutilized equipment and unnecessary investments

Longupgradecyclesandsinglelayerplanningwereadoptedinthepasttoaccountforthelackofopticalagilityanddynamiccontrol/management ofopticalnetworkresources

Emerging elastic optical technology and Software Defined Networking paradigm increaseflexibility, enabling a joint multi-layer operation and short network re-optimization/upgrade cycles:

þ operate the network close to its true capabilities and postpone or avoid investmentsþ multilayer coordination allows more efficient resources usageþ capture traffic dynamicity and technology maturation (depreciation & better technology)

NetworkModelIP-over-ElasticOpticalNetwork

PlanninganIPoverEONconsistsof3inter-relatedsub-problems:

§ Opticalnetwork:virtualtopologydesign

§ RoutingoflightpathsandModulationLevelselection(RML)

§ SpectrumAllocation(SA)

§ IProuting(IPR)ontopofthevirtualtopology

MultilayerCAPEXmodel

§ Optical: flex-grid enabled ROADMs and tunable - Bandwidth Variable

Transponders (BVTs)

§ IP: Modular IP/MPLS routers organized into 3 component classes: basic

node (3 types of chassis), line-cards, and short reach transceivers

ONDM 2017

IncrementalMulti-layerNetworkPlanning(1/2)

ONDM 2017

Periodt0

Initial planning period both

(IP and optical) layers are

simultaneously optimized

with the objective being the

minimization of the cost.

t0

topology

trafficdemands

availableequipment

spectrumdescription

constraints

Period tN:Input� new traffic demands� previous state of the network at tN-1� state of the resources (established

lightpaths and IP tunnels)� information about physical resources

(installed/available equipment and itslocation)

Objectiveminimize:� CapEx of added network equipment

� Control the re-configurationbetween the two successive networkstates (low disruption and manualinterventions)

tN

previous state of thenetworkattN-1

established lightpathsandIPtunnels

installed/availableequipment and itslocation

technologymaturationequipmentcostdecrease

new trafficdemands

….

IncrementalMulti-layerNetworkPlanning(2/2)

Incrementalmodel

þ flexibility ofBVTsdifferentconfigurationstocarryclienttrafficadapt transmissionparameters

þ FlexibilityofIPexploitIPgroomingcapabilitiestoenablesparecapacityutilization

þ Exploittechnologymaturation(depreciation)

þ Capture trafficdynamicity

ONDM 2017

tN-1 tNNetwork re-optimization

capex minimizationIP tunnelsEstablished lightpaths

- exploit spare capacity

- BVT reconfiguration

opex minimization- optimize changes made over the entire period of the network lifecycle

100 150 200 250 300 350 400 500 600 700 800 900 1T

Client port rate

1 2 3 4 5 6 7 8

Number of optical carriers Modulation scheme

BPSK 16QAM

BVT reconfiguration

Multi-periodplanningtechniquesTechniqueI(reference):Jointmultilayerplanningwithoutpreviousstate (J-ML)î Dimension each period from scratch (as if period is the initial) – lowest possible cost

TechniqueII:Incrementalmultilayerplanningontopofthepreviousstate(Inc)î Incremental dimensioning with no reconfiguration of transponders & IP equipment(maintain the previous network state)

TechniqueIII:Incrementalmultilayerplanningwithoptimizedadaptationsî Allow but penalize the adaptations from the previous network stateî Study two variations:

ê Allow (for free) IP layer adaptations, forbid optical layer adaptations (Inc-ML)ê Allow IP and allow but penalize optical layer adaptations (J-Inc-ML)

ONDM 2017

Transitionbetweennetworkstates(1/3)

ONDM 2017

tNtN+1demandDN=180GbpsDN+1=240Gbps

IProuter

200G

200G

IProuter

200G

100G

IProuter

IProuter

100G

200G

Incremental planning without being able to

perform any change from the previous

network state.

ý No transponders and IP equipment

reconfigurations

ý capacity overprovisioning

ý underutilized equipment

ý unnecessary investments

Transitionbetweennetworkstates(2/3)

ONDM 2017

IProuter

flex200G

flex200G

IProuter

IProuter

flex250G

flex250G

IProuter

tNdemandDN=180GbpsDN+1=240Gbps

Incremental multilayer planning optimizing

lightpath reconfiguration between periods

þ flexibility ofBVTs andIPlinecards

þ re-optimization of the previousnetwork state

tN+1

Transitionbetweennetworkstates(3/3)

ONDM 2017

tNIncremental multilayer planning

optimizing modifications between periods

þ flexibility ofBVTsþ exploit IP grooming capabilitiesto enable spare capacity utilization

þ re-optimization of the previousnetwork state

tN+1

A

B

C

flex250G

flex250G

flex680G

flex280G IP

router

flex680G

IProuter

IProuter

flex280G

A

B

C

flex380G

flex380G

flex780G

flex400G IP

router

flex780G

flex400G

IProuter

IProuter

80Gbps

𝐷"#$%&' =250Gbps𝐷"#$%&( =280Gbps𝐷"#$%'( =680Gbps

𝐷"#$%&' =300Gbps𝐷"#$%&( =480Gbps=AC:400+ABC:80𝐷"#$%'( =700Gbps

[AàB]=380Gbps[AàC]=400Gbps[BàC]=780Gbps

[AàB]=250Gbps[AàC]=280Gbps[BàC]=680Gbps

BVT400G

BVT1T maximum rate of 1 Tbps

maximum rate of 400 Gbps

ILPmodel

ONDM 2017

Input• ThenetworktopologyrepresentedbygraphG(V,L).• ThemaximumnumberZofavailablespectrumslots(of12.5GHz)• ThetrafficdescribedbythetrafficmatrixΛ.• AsetBoftheavailabletransponders(BVTs).• A set T of feasible transmission tuples, which represent the

transmission options of the available transponders, with tuplet=(Dt,Rt,St,Ct)indicatingfeasibilityoftransmisionatdistanceDt,withrateRt(Gpbs),usingStspectrumslots,forthetransponderofcostCt.Also,Tbrepresentsthetransmissiontuplesoftransponderb B.

• A set of line-cards represented by H, where a line-card fortransponderb Bisrepresentedbyatuplehb=(Nh,Ch),whereNhisthenumberoftranspondersoftypebthattheline-cardsupports.

• The IP/MPLS router cost, specified by amodular costmodel.WeassumethatanIP/MPLSrouterconsistsofline-cardchassisofcostCLCC,thatsuportNLCCline-cardseach,andfabriccardchassisofcostCFCC,thatsuportNFCCline-cardchassis.

• Theweightingcoefficient,WC,takingvaluesbetween0and1.Setting𝑊𝑐 = 1minimizessolelythecostwhereassetting𝑊𝑐 ≈ 0minimizesthemaximumspectrumused.

• The weighting coefficient,Wd, taking values between 0 and 1.SettingWd=1minimizessolelythecurrentstatecostignoringtheprevious network state, whereas setting𝑊𝑑 ≈ 0 maintains thepreviousstatelightpathsandminimizesanyadditionalcosttothat.

Variables• 𝑓𝑠𝑑

𝑝 :Floatvariables,equaltotherateoftheIPtunnelfromIPsourcestodestinationdthatpassesoveralightpaththatusespathp.• xpt:Integervariables,equaltothenumberoflightpathsofpath-transmissiontuplepairs(p,t)used.• ynh:Integervariables,equaltothenumberofline-cardsoftypehatnoden.• qn:Integervariables,equaltothenumberofline-cardchassisatnoden.• on:Integervariables,equaltothenumberoffabric-cardchassisatnoden.• z:Integervariable,equaltothemaximumindexedspectrumslot.• θnb:Integervariables,equaltothenumberofutilizedtranspondersoftypebatnoden.• vnb:Integervariables,equaltothenumberofdeployedtranspondersoftypebatnoden.• dpt:Integervariables,equaltothenumberofremoved(p,t)tuplesfromthepreviousstate.• c:Floatvariable,equaltothecostofnetworkequipment.

Constants• 𝐹𝑠𝑑

′ 𝑝 :Integerconstants,equaltotheIPtrafficofend-nodesstodthatistransferredoveropticalpathpinthepreviousnetworkstate.• 𝑋𝑝𝑡′ :Integerconstants,equaltothenumberoflightpathsofpath-transmissiontuplepairs(p,t)usedinthepreviousnetworkstate.• 𝛩𝑛𝑏′ :Integerconstants,equaltothenumberoftranspondersoftypebatnodenusedinthepreviousnetworkstate.

Objective

Costcalculationconstraints:

IPflowcontinuityconstraints:

Path-transmissiontupleassignmentconstraints:

Previousstateconstraints(opticallayer):

Maximumspectrumslotusedconstraints:

Previousstateconstraints(IPlayer):

Utilizedtranspondersconstraints:

Deployedtranspondersconstraints:

Numberofline-cardspernodeconstraints:

Numberofline-cardchassispernodeconstraints:

Numberoffabriccardchassispernodeconstraints:

Î

Î

( )min (1 )C CW c W z× + - ×| ( , )

(

) (1 )

d b nb h nhn V b B n V h H

ptLCC n CH n dp P t T p tn V n V

c W C v C y

dC q C o WÎ Î Î Î

Î Î $Î Î

+= × × ×

+ × + × + - ×

åå ååå åå å

( ) 2, , Î Î" s d V n V

,,

0, ,in nj

sdp psd sd sd

i V p P j V p P

n sf f n d

n s dÎ Î Î Î

-L =ìïæ ö- = L =íç ÷ ïè ø ¹î

å å åå

2( , )" Îi j V

2 | ( , )( )

ij

psd t pt

p P t T p tsd V

f R xÎ Î $Î

£ ×å å å

( )feasible p,t"

pt pt ptd X x¢³ -

( )

2

| | ( , )

, ( , ) ,( )

maxij

lt pt

p P l p t T p t

l

l L i j Vz S x

z zz Z

Î Î Î $

" Î Î

= ×

=£

å å

( ) , ( ) , | 0,2 2 pij sds,d V i, j V p P F¢" Î Î Î >

p psd sdf F ¢>

, n V b B" Î Î

b

nb ptp starts at n t T

xqÎ

= å å

, n V b B" Î Î

nb nbv q³

nb nbv ¢³ Q

," Î În V h H

is supported by

/nh nb hb h

y v N³ å

/ ,n nh LCChVq y N n" γå

/ ,n n CHo q N n V³ " Î

Constraintsê jointly consider multi-layer andincremental planning / detailedcost model for optical and IPequipment

ê Penalize the reconfiguration ofexisting lightpaths, to controlthe extent of modificationsperformed between periods

ObjectiveThree-objective minimization(CapEx, Spectrum, reconfigurations)

Inputq new traffic demandsq previous state of the networkq Weight to penalize the

deviation from previous state

Illustrativeresults- Scenarioê DT topology, 2012 traffic, uncompensated links, 12.5 GHz slots

ê Incremental planning for 2016-2026 - yearly

ê Traffic increase by 35% uniformly per year

ê 2 types of BVTs/regens (i) 400 Gbps, (ii) 1 Tbps, available at 2020

ê 10% price depreciation per year

ê Objective: weighted minimization of Capex, reconfigurations,spectrumê 80% CaPex, 10% reconfigurations, and the rest 10% spectrum

ONDM 2017

Capacity (Gb/s) Reach (km) Data slots cost (c.u.) Capacity (Gb/s) Reach (km) Data slots cost (c.u.)

100 2000 4 500 950 7150 1350 4 600 800 8200 1050 5 700 700 9250 950 5 800 650 11300 700 6 900 550 12350 600 6 1000 450 14

400 450 6

1.762*

*available from 2020

BVT 1 BVT 2

ILP termination

ê Integer gap tolerance: 0.05

IllustrativeResults(1/2)

1c.u.:costofa100Gb/scoherentopticaltransponder

Ý J-ML : the minimum CaPex, as if the network was planned from scratch on each yearÞ Inc: exhibits the worst performance, due to the inability to exploit IP & optical equipment reconfigurations

Ý Inc-ML : exploits reconfiguration capabilities of IP layer to achieve cost savingsÝ J-Inc-ML: exploits reconfiguration capabilities of both layers and achieves even higher cost savings

ONDM 2017

0

500

1000

1500

2000

2500

3000

3500

4000

4500

5000

2016 2017 2018 2019 2020 2021 2022 2023 2024 2025 2026

Cape

x (c

.u.)

Year

J-MLIncInc-MLJ-Inc-ML

BVT_2available

J-ML: Joint multilayer planning withoutprevious state(benchmark- minimum)

Inc: Incremental multilayer planning on topof existing state (no adaptation of optical andIP equipment)

Inc-ML: Incremental multilayer planning(allow for free IP layer adaptation, noadaptation of optical layer)

J-Inc-ML: Incremental multilayer planning(penalize IP and optical adaptations)

IllustrativeResults(2/2)

Spectrum UtilizationÝ J-Inc-ML also achieves spectrum savingsThe savings are slightly lower compared to CapExdue to deployment of more regenerators* for Incand Inc-ML*(regenerators provide wavelength conversionpossibilities)

Trade-off betweenthe added equipmentand reconfigurationsbetween consecutivenetwork states

ONDM 2017

0

500

1000

1500

2000

2500

3000

3500

4000

2016 2017 2018 2019 2020 2021 2022 2023 2024 2025 2026

Spec

trum

(G

Hz)

Year

J-MLIncInc-MLJ-Inc-ML

BVT_2available

q Challenge: optimize the reconfigurations made together with the minimization of the cost

0

20

40

60

80

100

120

140

160

2017 2018 2019 2020 2021 2022 2023 2024 2025 2026

reco

nfig

urat

ion

proc

esse

s

Year

J-ML (# of reconfigurations)

J-Inc-ML (# of reconfigurations)

0

20

40

60

80

100

120

140

2017 2018 2019 2020 2021 2022 2023 2024 2025 2026

# of

add

ed li

ghtp

aths

Year

J-MLIncInc-MLJ-Inc-ML

ExtendedmodelsWorkinprogress

Short network cycles:þ are able to capture the effects of traffic dynamicity and avoid overprovisioning (smallbutfrequentnetworkupdates)

þ postpone the investments and exploit technology maturation

ONDM 2017

cost vs network upgrade cycles

2-months vs multi-year

network upgrade cycles

0

1000

2000

3000

4000

5000

6000

7000

2017

_a20

17_b

2017

_c20

17_d

2017

_e20

17_f

2018

_a20

18_b

2018

_c20

18_d

2018

_e20

18_f

2019

_a20

19_b

2019

_c20

19_d

2019

_e20

19_f

2020

_a20

20_b

2020

_c20

20_d

2020

_e20

20_f

2021

_a20

21_b

2021

_c20

21_d

2021

_e20

21_f

2022

_a20

22_b

2022

_c20

22_d

2022

_e20

22_f

2023

_a20

23_b

2023

_c20

23_d

2023

_e20

23_f

2024

_a20

24_b

2024

_c20

24_d

2024

_e20

24_f

2025

_a20

25_b

2025

_c20

25_d

2025

_e20

25_f

2026

_a20

26_b

2026

_c20

26_d

2026

_e20

26_f

Cap

Ex (

cost

uni

ts)

Year

2-months increments12-months increments24-months increments60-months incrementswithout previous state

Conclusionsq Long and independent (between network layers) upgrade cycles lead tocapacity overprovisioning, underutilized equipment and unnecessaryinvestments

q Shorter upgrade cycles and joint multi-layer upgrades increase thenetwork efficiency, operate the network closer to its true capabilities,and postpone or avoid investments

q ILP model that combines multi-layer and incremental planning andtradeoffs:ê the capital expenditure (CapEx) of the added equipment at both layers

ê the reconfigurations for the transition between two consecutive periods

ONDM 2017

Questions?

ONDM 2017