Seediscussions,stats,andauthorprofilesforthispublicationat:https://www.researchgate.net/publication/276171282

RoutingandSpectrumAllocationinElasticOpticalNetworks:ATutorial

ArticleinIEEECommunicationsSurveys&Tutorials·July2015

ImpactFactor:6.81·DOI:10.1109/COMST.2015.2431731

CITATIONS

11

READS

499

3authors:

BijoyChandChatterjee

TheUniversityofElectro-Communications

23PUBLICATIONS79CITATIONS

SEEPROFILE

NityanandaSarma

TezpurUniversity

58PUBLICATIONS312CITATIONS

SEEPROFILE

EijiOki

TheUniversityofElectro-Communications

321PUBLICATIONS2,123CITATIONS

SEEPROFILE

Allin-textreferencesunderlinedinbluearelinkedtopublicationsonResearchGate,

lettingyouaccessandreadthemimmediately.

Availablefrom:BijoyChandChatterjee

Retrievedon:12June2016

1

Routing and Spectrum Allocation in Elastic OpticalNetworks: A Tutorial

Bijoy Chand Chatterjee IEEE, Member, Nityananda Sarma IEEE, Member, and Eiji Oki, IEEE, Fellow

Abstract—Flexgrid technology is now considered to be apromising solution for future high-speed network design. Inthis context, we need a tutorial that covers the key aspectsof elastic optical networks. This tutorial paper starts with abrief introduction of the elastic optical network and its uniquecharacteristics. The paper then moves to the architecture of theelastic optical network and its operation principle. To completethe discussion of network architecture, this paper focuses on thedifferent node architectures, and compares their performance interms of scalability and flexibility. Thereafter, this paper reviewsand classifies routing and spectrum allocation (RSA) approachesincluding their pros and cons. Furthermore, various aspects,namely — fragmentation, modulation, quality-of-transmission,traffic grooming, survivability, energy saving, and networkingcost related to RSA, are presented. Finally, the paper explores theexperimental demonstrations that have tested the functionalityof the elastic optical network, and follows that with the researchchallenges and open issues posed by flexible networks.

Index Terms—Elastic optical networks, node architecture,spectrum management, routing and spectrum allocation, andsliceable bandwidth-variable transponder.

NOMENCLATURE

AG Auxiliary GraphAoD Architecture on DemandAR Adaptive RoutingAST Average Setup TimeBP Blocking ProbabilityBPSK Binary Phase-Shift KeyingBV-SSS Bandwidth-Variable Spectrum Selective

SwitchBVT Bandwidth-Variable TransponderBV-WXC Bandwidth-Variable Cross-ConnectCF Central FrequencyDWDM Dense Wavelength Division MultiplexingEDFA Erbium-Doped Fiber AmplifierEF Exact FitFAR Fixed Alternate RoutingFF First Fit

Manuscript received xxxxx; revised xxxxx; accepted xxxxx. Date of pub-lication xxxx; date of current version xxxx. This work was supported in partby Strategic Information and Communication R&D Promotion Program ofthe Ministry of Internal Affairs in Japan and National Institute of Informationand Communications Technology, Japan.

B. C. Chatterjee and E. Oki are with the Department of Informationand Communication Engineering, The University of Electro-Communications,Tokyo - 182-8585, Japan. e-mail: {bijoycc, eiji.oki}@uec.ac.jp.

N. Sarma is with the Department of Computer Science and Engineering,Tezpur University, Assam-784028, India. e-mail: nitya@tezu.ernet.in.

FLF Fast-Last-FitFR Fixed RoutingGb/s Gigabit Per SecondGMPLS Generalized Multi-Protocol Label SwitchingHOPS Hitless Optical Path ShiftIEEE Institute of Electrical and Electronics Engi-

neersILP Integer Linear ProgrammingIP Internet ProtocolITU International Telecommunication UnionLCoS Liquid Crystal on SiliconLCR Least Congested RoutingLSP Label-Switched PathLU Least UsedMC-RSA Multicast-Capable Routing and Spectrum

AssignmentMEMS Micro-Electro Mechanical SystemMILP Mixed-Integer Linear ProgrammingMUX MultiplexerNSFNET National Science Foundation NetworkNP Non-Deterministic Polynomial TimeOFDM Orthogonal Frequency-Division Multiplex-

ingOSPF Open Shortest Path FirstPCE Path Computation ElementPIC Photonic Integrated CircuitPLI Physical Layer ImpairmentPSK Phase-Shift KeyingQAM Quadrature Amplitude ModulationQoT Quality-of-TransmissionQPSK Quadrature Phase-Shift KeyingR RandomRML Routing and Modulation LevelRODAM Reconfigurable Optical Add-Drop Multi-

plexerRSA Routing and Spectrum AllocationRSVP-TE Resource Reservation Protocol - Traffic En-

gineeringRWA Routing and Wavelength AssignmentRX ReceiverSA Spectrum AllocationSBVT Sliceable Bandwidth-Variable TransponderSDM Space Division MultiplexingSDN Software-Defined NetworkingSF Smallest FitSSS Spectrum Selective SwitchTAP-BR Time-Aware Provisioning with Bandwidth

Reservation

2

Tb/s Terabit Per SecondTC Time ComplexityTDM Time Division MultiplexingTS-EDFA Transient-Suppressed Erbium-Doped Fiber

AmplifierTX TransmitterWDM Wavelength Division Multiplexing

I. INTRODUCTION

THE rapid growth in world-wide communications and therapid adoption of the Internet has significantly modified

our way of life. This revolution has led to a vast growth incommunication bandwidth in every year. An optical networkhas the potential to support the continued demands for com-munication bandwidth. The high capacity of dense wavelengthdivision multiplexing (DWDM)-based optical networks [1], [2]is assisted by the use of upper layers to aggregate low-rate traf-fic flows into lightpaths in mechanism of traffic grooming [3]–[6]. DWDM-based systems with up to 40 Gb/s capacity perchannel have been deployed in backbone networks, while 100Gb/s interfaces are now commercially available, and 100 Gb/sdeployment is expected soon. TeleGeography [7] expects thatinternational bandwidth demands will be approximately 606.6Tb/s in 2018 and 1,103.3 Tb/s in 2020. Therefore, opticalnetworks will be required to support Tb/s class transmissionin the near future [8], [9]. Unfortunately, conventional opticaltransmission technology has inadequate scaling performanceto meet the growing traffic demands as it suffers from theelectrical bandwidth bottleneck limitation, and the physicalimpairments become more serious as the transmission speedincreases [10]. Moreover, the traffic behavior is changingrapidly and the increasing mobility of traffic sources makesgrooming more complex. Therefore, researchers are now fo-cusing on new technologies for high-speed optical networks.

To meet the needs of the future Internet, optical transmissionand networking technologies are moving toward to the goals ofgreater efficiency, flexibility, and scalability. Recently, elasticoptical networks [11]–[15] have been shown to be a promisingcandidate for future high-speed optical communication. Anelastic optical network has the potential to allocate spectrum tolightpaths according to the bandwidth requirements of clients.The spectrum is divided into narrow slots and optical connec-tions are allocated a different number of slots. As the result,network utilization efficiency is greatly improved compared toDWDM-based optical networks. In elastic optical networks, acertain number of transmission parameters, such as — opticaldata rate, modulation format, and wavelength-spacing betweenchannels, which are fixed in currently deployed networks, willbe made tunable. Given that future demands indicate thathigh-speed optical connections are needed to optimize datatransport, elastic optical networks are a suitable replacementfor DWDM-based optical networks.

As the elastic optical network has been widely acceptedas the next generation high-speed network, researchers havefocused on its architecture, spectrum allocation mechanism,and future scope. Accordingly, a comprehensive survey on or-thogonal frequency-division multiplexing (OFDM)-based opti-cal high-speed transmission and networking technologies, with

a specific focus on OFDM technology, was provided by G.Zhang et al. [15]. S. Talebi et al. [16] presented a review ofspectrum management techniques for elastic optical networks.Recently, a survey of the current ongoing research efforts intodefining elastic optical network control plane architectures waspresented in [17].

A. Contribution

This paper provides an integrated tutorial that covers differ-ent aspects of elastic optical networks. In this tutorial paper,we explore routing and spectrum allocation in elastic opticalnetworks. We begin with the basic concept of the elasticoptical network and its unique characteristics, and then thearchitecture of the elastic optical network and its operationprinciple are presented. We elucidate the functionalities ofthe bandwidth-variable transponder (BVT) and the sliceablebandwidth-variable transponder (SBVT). SBVT architectureand its advantages in the future optical network are detailed.Immediately after discussing network architecture, our discus-sion focuses on the different node architectures, namely —broadcast and select, spectrum routing, switch and select withdynamic functionality, and architecture on demand, along withtheir functionalities. We compare different node architecturesin order to clarify their performances. Next we look intothe basic concept of the routing and spectrum allocation(RSA) approach in elastic optical networks. We discuss thedifferences between RSA and routing and wavelength as-signment (RWA) in optical networks. Then, we move todifferent routing approaches along with their pros and cons.We compare different routing approaches in order to analyzetheir performances. Thereafter, we discuss different spectrumallocation policies. In addition, we separate the spectrumallocation polices into two categories based on spectrum rangeallocation for connection groups and spectrum slot allocationfor the individual connection request.

Various aspects, namely — fragmentation, modulation,quality-of-transmission, traffic grooming, survivability, energysaving, and networking cost related to RSA, all of which havebeen reported in the literature, are presented in this tutorialpaper. We classify the fragmentation-aware approaches, or de-fragmentation approaches, and explain how these approachesdeal with fragmentation problems. Distance adaptive RSA thatconsiders the modulation technique is one of the features of theelastic optical network, and is also presented in this tutorial pa-per. We discuss and compare traffic grooming in wavelength-division multiplexing (WDM)-based optical networks, elasticoptical networks with BVTs, and elastic optical networkswith SBVTs. Different survivability techniques, namely —protection and restoration, are addressed in this paper. Wediscuss the networking cost reduction made possible by theuse of SBVT in elastic optical networks. In addition, ourdiscussion on various aspects related to RSA is summarized.

Finally, we explore the experimental demonstrations thathave been published to confirm the functionality of the elasticoptical network. We address the research challenges and openissues posed by flexible networks.

3

B. Organization

The rest of the paper is organized as follows. Section IIprovides the basic concept of the elastic optical network.Section III presents the architecture of the elastic opticalnetwork. Section IV explains the different node architec-tures. Section V discusses the basic RSA approach in theelastic optical network. Section VI discusses and comparesthe major routing problems of the elastic optical network.Section VII focuses on different spectrum allocation policies.Various issues related to RSA are presented in Section VIII.Section IX presents the experimental demonstrations that havetested the functionality of the elastic optical network. Researchissues and challenges faced by optical network researchers arehighlighted in Section X. We close this tutorial in Section XI,where we draw our conclusions.

II. CONCEPT OF ELASTIC OPTICAL NETWORKS

The traditional WDM-based optical network divides thespectrum into separate channels. The spacing between ad-joining channels is either 50 GHz or 100 GHz, which isspecified by international telecommunication union (ITU)-Tstandards as shown in Fig. 1. The frequency spacing betweentwo adjacent channels is relatively large. If the channels carryonly low bandwidth, and no traffic can be transmitted in thelarge unused frequency gap, a large portion of the spectrumwill be wasted, which is reflected in Fig. 2.

50 GHz

Fig. 1. ITU-T grid.

Frequency

Unutilized spectrum

Fig. 2. Spectrum allocation in WDM based optical networks.

Filter guard band

Fig. 3. Overlapping subcarriers caused by OFDM technology.

To overcome the limitations of traditional optical networks,M. Jinno et al. [11] presented a spectrum efficient elastic op-tical network based on OFDM technology [15], [18]. Optical

OFDM allocates the data to several low data rate subcarrierchannels. As the spectrum of adjacent subcarrier channels areorthogonally modulated, they can overlap each other as shownin Fig. 3, which increases transmission spectral efficiency. Fur-thermore, optical OFDM can provide fine-granularity capacityto connections by the elastic allocation of low rate subcar-riers. A bandwidth-variable OFDM transponder generates anoptical signal using just enough spectral resources, in termsof subcarriers with appropriate modulation level, to satisfy theclients requirements. OFDM signals are usually generated inthe radio-frequency domain, so many transmission propertiescan be freely set, i.e., different subcarriers can be assigneddifferent numbers of modulated bits per symbol. To establisha connection, each bandwidth variable cross-connect on theroute allocates a cross-connection with sufficient spectrum inorder to create an appropriately sized end-to-end optical path.This end-to-end path is expanded and contracted according tothe traffic volume and user requests, as necessary. The maincharacteristics of an elastic optical network are bandwidth seg-mentation, bandwidth aggregation, efficient accommodation ofmultiple data rates, elastic variation of allocated resources,reach-adaptable line rate, etc. These are discussed in moredetail below.

• Bandwidth segmentation: Traditional optical networksrequire full allocation of wavelength capacity to an opticalpath between an end-node pair. However, elastic opticalnetworks provide a spectrum efficient bandwidth seg-mentation (sometimes called sub wavelength) mechanismthat provides fractional bandwidth connectivity service.If only partial bandwidth is required, elastic opticalnetwork can allocate just enough optical bandwidth toaccommodate the client traffic, as shown in Fig. 4, wherea 40 Gb/s optical bandwidth is segmented into three subwavelengths, such as — 5 Gb/s, 15 Gb/s, and 20 Gb/s.At the same time, every node on the route of the opticalpath allocates a cross-connection with the appropriatespectrum bandwidth to create an appropriate-sized end-to-end optical path. The efficient use of network resourceswill allow the cost-effective provisioning of fractionalbandwidth service.

• Bandwidth aggregation: Link aggregation is a packetnetworking technology standardized in IEEE 802.3. Itcombines multiple physical ports/links in a switch/routerinto a single logical port/link to enable incrementalgrowth of link speed as the traffic demand increasesbeyond the limits of any one single port/link. Similarly,the elastic optical network enables the bandwidth ag-gregation feature and so can create a super-wavelengthoptical path contiguously combined in the optical domain,thus ensuring high utilization of spectral resources. Thisunique feature is depicted in Fig. 4, where three 40 Gb/soptical bandwidths are multiplexed with optical OFDM,to provide a super-channel of 120 Gb/s.

• Efficient accommodation of multiple data rates: As shownin Fig. 4, the elastic optical network has the ability toprovide the spectrally-efficient direct accommodation ofmixed data bit rates in the optical domain due to its flex-

4

Optical fiberOptical fiber

Client node40 Gb/s

40 Gb/s

40 Gb/s

10 Gb/s

40 Gb/s

120 Gb/s

10 Gb/s

15 Gb/s

5 Gb/s

Bandwidth segmentation

Sub wavelength

Super wavelength

Multiple data rate

20 Gb/s

40 Gb/s Elastic variation

Link aggregation

Bandwidth aggregation

Fig. 4. Unique characteristics, namely — bandwidth segmentation, bandwidth aggregation, accommodation of multiple data rates, and elastic variation ofallocated resources, of elastic optical networks.

ible spectrum assignment. Traditional optical networkswith fixed grid leads to wastage of the optical bandwidthdue to the excessive frequency spacing for low bit ratesignals.

• Reach-adaptable line rate: The elastic optical networkhas the ability to support reach-adaptable line rate, aswell as dynamic bandwidth expansion and contraction,by altering the number of subcarriers and modulationformats.

• Energy saving: It supports energy-efficient operations inorder to save power consumption by turning off some ofthe OFDM subcarriers while traffic is slack.

• Network virtualization: It allows optical network visual-ization with virtual links supported by OFDM subcarriers.

In elastic optical networks, spectrum resources are allocatedin a contiguous manner. Accordingly, allocation of resourcesto short duration connections can increase the number of non-aligned available slots, unless the allocation is well organized.These non-aligned available slots can be reduced by partition-ing the total set of subcarrier slots as shown in Fig. 5. Thiswill increase the number of acceptable connections and hencedecrease the blocking probability.

Sometime, partitioning also negatively impacts the blockingprobability due to the lack of statistical multiplexing-gain [19].We explain this with a simple example wherein partitioning thetotal set of subcarrier slots decreases the number of channelsin each partition, which increases the networks blockingprobability. First, we calculate the blocking probability usingthe Erlang B loss formula [20] under a simple traffic model(a Poisson arrival process and exponential distribution of theconnection holding times). If the number of channels is 100and the offered traffic is 100 Erlang, the blocking probabilityis 0.0757. Dividing the same channel resources among fourpartitions and splitting the traffic among the partitions (i.e.,25 channels with offered traffic volume of 25 Erlang), theblocking probability for each partition becomes 0.1438, whichis higher than that of the non-partitioned case.

Therefore, based on the above discussion, partitioning has

(a)

Link 1

Link 2

Link 3

Slot 1 2 3 4 5 6 7 8 9

Partition 1 Partition 2 Partition 3

Link 1

Link 2

Link 3

Slot 1 2 3 4 5 6 7 8 9

(b)

Aligned available slots

Fig. 5. Comparison of spectrum allocation (a) without spectrum partitioning,and (b) with spectrum partitioning in elastic optical networks.

an advantage in reducing the bandwidth blocking probabilityin elastic optical networks, provided the number of partitionsis minimized. Taking this direction R. Wang et al. [21] andW. Fadini et al. [22] presented spectrum partitioning schemesin order to utilize the spectrum resources efficiently andmanage the fragmentation issue; both are discussed later inthe fragmentation subsection (Section VIII-A).

III. ELASTIC NETWORK ARCHITECTURE

To fulfill our ever-increasing bandwidth demands, the elasticoptical network is indispensable. This is due to its many desir-able properties including flexible data rate and spectrum allo-cation, low signal attenuation, low signal distortion, low powerrequirement, low material usage, small space requirement, andlow cost. This section discusses the architecture of the elasticoptical network. Figure 6 shows the typical architecture ofthe elastic optical network, which mainly consists of BVTsand BV-WXCs. These basic components and their workingprinciple are explained in the following subsections.

5

Optical fiber

Client Client

ClientClient

Client Client

Bandwidth variablecross-connect(BV-WXC)

Bandwidth variabletransponder(BVT)

Fig. 6. Architecture of elastic optical network.

A. Bandwidth-variable Transponder

BVTs [15] are used to tune the bandwidth by adjustingthe transmission bit rate or modulation format. BVTs supporthigh-speed transmission using spectrally efficient modulationformats, e.g., 16-quadrature amplitude modulation (QAM),with 64-QAM used for shorter distance lightpaths. Longerdistance lightpaths are supported by using more robust butless efficient modulation formats, e.g., quadrature phase-shiftkeying (QPSK) or binary phase-shift keying (BPSK). There-fore, BVTs are able to trade spectral efficiency off againsttransmission reach.

BVT ROADMTraffic towardsingle destination

Single variable bandwidthoptical flow

(a)

SBVT ROADMTraffic toward singleor multiple destinations

Single or multiple variablebandwidth optical flow

(b)

BVT: Bandwidth variable transponderROADM: Reconfigurable optical add-drop multiplexer

SBVT: Sliceable bandwidth variable transponder

Fig. 7. (a) Functionalities of (a) BVT, and (b) SBVT.

However, when a high-speed BVT is operated at lower thanits maximum rate due to required reach or impairments in theoptical path, part of the BVT capacity is wasted. In order

to address this issue, an SBVT [23]–[27] has been presentedthat offers improved flexibility; it is seen as a promisingtransponder technology. An SBVT has the capability to al-locate its capacity into one or several optical flows that aretransmitted to one or several destinations. Therefore, whenan SBVT is used to generate a low bit rate channel, its idlecapacity can be exploited for transmitting other independentdata flows. An SBVT generates multiple optical flows that canbe flexibly associated with the traffic coming from the upperlayers according to traffic requirements. Therefore, opticalflows can be aggregated or can be sliced based on the trafficneeds. Figure 7 distinguishes BVT and SBVT functionalities.

SBVTClient

Ele

ctro

nic

proc

essi

ngm

odul

e

Ele

ctro

nic

switc

h

source andMulti-wavelength

subcarrier filtering

PIC

Opt

ical

MU

X

PIC

PIC

PIC

Electrical line

Optical line

SBVT: Sliceable bandwidth variable transponderPIC: Photonic integrated circuitMUX: Multiplexer

Fig. 8. Architecture of SBVT.

The SBVT architecture [23], [24] was introduced in order tosupport sliceability, multiple bit rates, multiple modulation for-mats, and adaptive code rates. Figure 8 shows the architectureof an SBVT; it mainly consists of a source of N equally spacedsubcarriers, a module for electronic processing, an electronicswitch, a set of N photonic integrated circuits (PICs), and anoptical multiplexer. In this architecture, the N subcarriers aregenerated by a single multi wavelength source. However, sucha source may be replaced by N lasers, one per subcarrier. Eachclient is processed in the electronic domain (e.g., for filtering)and then is routed by the switching matrix to a specificPIC. The generated carriers are equally spaced according tothe spectral requirements and transmission technique adopted.Generated subcarriers are selected at the multi wavelengthsource, and they are routed the appropriate PICs. Each PIC isutilized as a single-carrier transponder that generates differentmodulated signals, such as 16-QAM and QPSK, in order tosupport multiple modulation formats. Finally, subcarriers areaggregated by the optical multiplexer in order to form a superchannel. Sometime, subcarriers may be sliced and directed tospecific output ports according to the traffic needs. A detaileddescription of PIC generation of different modulated signalsis given in [24].

6

B. Bandwidth-variable Cross-connect

The BV-WXC [11] is used to allocate an appropriate-sizedcross-connection with the corresponding spectrum bandwidthto support an elastic optical lightpath. Therefore, a BV-WXCneeds to configure its switching window in a flexible manneraccording to the spectral width of the incoming optical signal.

Split

ter

Split

ter

Split

ter

BV-SSS

BV-SSS

BV-SSS

Input fiber 1

Input fiber 2

BV-SSSBV-SSS

Output fiber 1

Output fiber 2

BVTClient signal

BV-SSS: Bandwidth-variable spectrum selective switch

BVT: Bandwidth-variable transponder

Fig. 9. Architecture of BV-WXC.

Figure. 9 shows an implementation example of a BV-WXC, where bandwidth-variable spectrum selective switches(BV-SSSs) in the broadcast-and-select configuration are usedto provide add-drop functionality for local signals as wellas groomed signal, and routing functionality for transit sig-nals. Typically, a BV-SSS performs wavelength demultiplex-ing/multiplexing and optical switching functions using inte-grated spatial optics. The light from an input fiber is dividedinto its constituent spectral components using a dispersiveelement. The spatially-separated constituent spectra are fo-cused on a one-dimensional mirror array and redirected tothe desired output fiber. Liquid crystal on Silicon (LCoS) orMicro-Electro Mechanical System (MEMS)-based BV-SSSscan be employed as switching elements to realize an opticalcross-connect with flexible bandwidth and center frequency.As the LCoS is deployed according to phased array beamsteering, which utilizes a large number of pixels, LCoS-based BV-SSSs can easily provide variable optical bandwidthfunctionality. A detailed description of a BV-WSS employingLCoS technology can be found in [18], [28]. Similarly, detailsof an MEMS-based BV-SSS can be found in [18], [29].

IV. NODE ARCHITECTURES

This section discusses various node architectures [30], [31],which are the building blocks of spectrum efficient elasticoptical networks.

A. Broadcast-and-Select

The broadcast-and-select architecture has been used to de-termine the elastic optical node architecture that uses spectrumselective switches [31]. Figure 10 shows the node architectureof broadcast-and-select, which is implemented using splittersat the input ports. Splitters are used to generate copies of theincoming signals that are subsequently filtered by spectrumselective switches in order to select the required signals at thereceiver side. The add/drop network may implement colorless,direction-less, and contention-less elastic add/drop function-ality, thus allowing the addition of one or more wavelengthchannels to an existing multi-wavelength signal automatically.It can also drop (remove) one or more channels from thepassing signals to another network path dynamically. The maindrawbacks of the broadcast-and-select node architecture are asfollows - (i) it requires synchronization and rapid tuning, (ii)it cannot support wavelength reuse and hence a large numberof wavelength channels is required, (iii) the signal power issplit among various nodes, so this type of node cannot beused for long distance communication. The broadcast andselect architecture is mostly being used in high-speed localarea networks and metropolitan area networks. It must notedthat the broadcast-and-select architecture struggles to supportadditional functionality to cope with dynamic requirements,e.g., spectrum defragmentation [32]–[34].

1

N -1

1

N -1

Split

ter

1

SSS

1

1

N -1

N -1

1

1

N -1

1

N -1

N N

NN

Add/Drop Network

SSSN

-1

Split

terN

-1

SSS: Spectrum selective switch

Add ports Drop ports

Fig. 10. Node architecture of broadcast-and-select.

B. Spectrum Routing

The spectrum routing node architecture is being designed toovercome the problems with the broadcast-and-select node ar-chitecture. It is basically implemented with arrayed wavebandgratings [18] and optical switches as shown in Fig. 11. Inspectrum routing, both switching and filtering functionalitiesare controlled by the spectrum selective switches. The basic

7

advantage of this architecture, compared to the broadcast-and-select architecture, is that the through loss is not dependent onthe number of degrees. However, it requires additional spec-trum selective switches at the input fibers, which makes it moreexpensive to realize. Furthermore, the additional functionalityneeded to cope with dynamic requirements, e.g., spectrumdefragmentation [32]–[34], is still difficult to implement inthis architecture.

N -1

1

N -1

SSS

1

1

N -1

N -1

1

1

N -1

1

N -1

N N

N

Add/Drop Network

SSSN

-1

SSSN

-1

1

SSS

1

SSS: Spectrum selective switch

N

Add ports Drop ports

Fig. 11. Node architecture of spectrum routing.

C. Switch and Select with Dynamic FunctionalityWe have already observed that the broadcast-and-select

architecture and spectrum routing architecture are unable tosupport dynamic requirements, such as, spectrum defragmen-tation, time multiplexing, regeneration, etc. To overcome theselimitations, the switch and select architecture with dynamicfunctionality has been introduced. In this architecture, anoptical switch is used to direct copies of the input to a specificspectrum selective switch or to a module (f ) that providesadditional functionalities, such as — defragmentation, timemultiplexing, and regeneration. The outputs of the modulesconnect to spectrum selective switches, where the requiredsignals are filtered for delivery to the corresponding outputfiber. Figure 12 shows the node architecture of the elasticoptical network with the dynamic functionalities that supportdynamic requirements, namely — spectrum defragmentation,time multiplexing, and regeneration. These dynamic function-alities come at the price of additional large port count opticalswitches and larger spectrum selective switch port counts. Thenumber of ports is dedicated to provide a specific functionality,and hence the number of modules may be calculated from theexpected demand.

D. Architecture on DemandThe architecture on demand (AoD) [35] consists of an

optical backplane that is implemented with a large port-count optical switch connected to several processing modules,

1

N -1

1

N -1

Split

ter

1

SSS

1

1

N+P

N+P

1

1

N

1

N+P

Add/Drop Network

SSSN

-1

Split

terN

-1 Opt

ical

Switc

h

N

N+1

N+P

N+1

SSS: Spectrum selective switch

Add ports Drop ports

f

f

Fig. 12. Node architecture of switch and select with dynamic functionality.

namely — spectrum selective switch, fast switch, erbium-doped fiber amplifier (EDFA), spectrum defragmenter, splitter,etc. The inputs and outputs of the node are connected viathe optical backplane as shown in Fig. 13. The differentarrangements of inputs, modules, and outputs are realized bysetting appropriate cross connections in the optical backplane.Therefore, it provides greater flexibility than the architecturesexplained above. This is mainly due to the non-mandatorynature of the components (such as — spectrum selectiveswitch, power splitters and other functional modules) unlikestatic architectures, but they can be interconnected togetherin an arbitrary manner. The number of spectrum selectiveswitches and other processing devices is not fixed but can bedetermined based on the specific demand for that functionality.Thus, savings in the number of devices can balance theadditional cost of the optical backplane, and hence this typeof architecture provides a cost-efficient solution. Furthermore,AOD provides considerable gains in terms of scalability andresiliency compared to conventional static architectures.

1

N

1

N

Optical backplaneinputsFiber

outputsFiber

Processing moduleOther signal

Fast switchEDFASSSEDFA Splitter

EDFA: Erbium-doped fiber amplifierSSS: Spectrum selective switch

1 P1 P

Fig. 13. Node architecture on demand with N input/outputs, and signalprocessing modules.

8

TABLE ICOMPARISON OF DIFFERENT NODE ARCHITECTURES FOR THE ELASTIC OPTICAL NETWORK

Broadcast-and-select Spectrum routing Switch and select with dynamicfunctionality

Architecture-on-demand

Power loss [in dB] [31] 3dlog2(N)e + LSSS 2LSSS 3dlog2(N + P )e + LSSS +LSwitch + LM

Switch and select architecture +(m+1)LSwitch for path with mmodules

Switch/backplane port count [31] Not required Not required N (N+P ) 2(N -1) + N (N+P )Routing flexibility No No Medium HighSSS port count N N N+P PDefragmentation No No Yes YesTime multiplexing No No Yes YesRegeneration No No Yes Yes

SSS: Spectrum selective switch, LSSS: SSS loss, LSwitch: Switch/backplane loss, LM : Module loss

E. Comparing Node ArchitecturesTable I summarizes the above discussed node architectures

in terms of total power loss, port count of switch/backplane,routing flexibility, port count of spectrum selective switches,defragmentation capability, time multiplexing, and regenera-tion capability. The calculation of total power loss [31] isdetermined by the type of node architecture implemented. Incase of AOD, total power loss depends on the architectureimplemented and the number of cross connections used inthe optical backplane. The total power loss in the switch andselect with dynamic functionality architecture depends on thespectrum selective switches, backplane, and modules used.However, the total power losses of the broadcast-and-selectarchitecture and spectrum routing architecture mainly dependon the spectrum selective switches.

Port count of switch/backplane [31] varies the networkingcost. The switch and select node and AOD node architecturesneed optical switches. However, the number of SSSs and otherprocessing devices may be tailored to suit the specific demand.Therefore, as savings in the number of devices can offset theadditional cost of the optical backplane, it provides an overallcost-effective solution. The number of SSS ports required byan AoD node is not strictly related to the node degree. Thisis because several small-port-count SSSs may be connectedtogether in order to increase the number of available ports.

Routing flexibility is the capability of the system to carrysignals from source to destination along different routes. Thistype of flexibility is required when strengthening systemresilience to failures along working paths; signals may bedirected to their backup paths. Time multiplexing is usedto transmit and receive independent signals over a commonsignal path by synchronized switches at each end of thetransmission line. As a result, each signal appears on theline only a fraction of the time in an alternating pattern. Onthe other hand, all-optical 3R (Re-amplification, Re-shaping,and Re-timing) signal regeneration is needed to avoid theaccumulation of noise, crosstalk and non-linear distortion, andto ensure good signal quality for transmission over any pathin an optical network. Spectral defragmentation is a techniqueto reconfigure the network so that the spectral fragments canbe consolidated into contiguous blocks.

V. BASIC CONCEPT OF RSAThis section presents the basic concept of RSA in the elastic

optical network. RSA is considered one of the key function-

alities due to its information transparency and spectrum reusecharacteristics. RSA is used to (i) find the appropriate routefor a source and destination pair, and (ii) allocate suitablespectrum slots to the requested lightpath.

The RSA problem in elastic optical networks is equiva-lent to the RWA problem in WDM-based optical networks.The problem of establishing lightpaths for each connectionrequest by selecting an appropriate route and assigning therequired wavelength is known as the routing and wavelengthassignment (RWA) problem [1], [2]. In WDM-based opticalnetworks without wavelength converters, the same wavelengthmust be used on all hops in the end-to-end path of a con-nection. This property is known as the wavelength continuityconstraint.

The difference of RSA and RWA is due to the capabilityof the elastic optical network architecture to offer flexiblespectrum allocation to meet the requested data rates. In RSA,a set of contiguous spectrum slots is allocated to a connectioninstead of the wavelength set by RWA in fixed-grid WDM-based networks. These allocated spectrum slots must be placednear to each other to satisfy the spectrum contiguity constraint.If enough contiguous slots are not available along the desiredpath, the connection can be broken up into small multiple de-mands. Each one of these smaller demands would then requirea lower number of contiguous subcarrier slots. Furthermore,the continuity of these spectrum slots should be guaranteed ina similar manner as demanded by the wavelength continuityconstraint. If a demand requires t units of spectrum, then tcontiguous subcarrier slots must be allocated to it (due to thespectrum contiguity constraint), and the same t contiguousslots must be allocated on each link along the route of thedemand (due to the spectrum continuity constraint).

The concept of the contiguity and continuity constraintsof the spectrum allocation is explained with an example.For this purpose, we consider the network segment shownin Fig. 14. We assume a connection request that requires abit-rate equivalent to two slots for RSA from source node1 to destination node 4. The connection request cannot beestablished through the shortest route 1-2-4 because the linksfrom 1-2 and 2-4 have two contiguous slots that are notcontinuous, so the continuity and contiguity constraints are notsatisfied. However, the continuity and contiguity constraintsare satisfied if the connection uses the route 1-2-3-4, andspectrum slots 5 and 6.

RWA in WDM-based optical networks is an NP-hard prob-

9

1

2 3

4

1

2 3

4

Connection established

Connection rejected

Link

1

Link 2

Link4

Link3

1 2 3 4 5 6 7 8Slots

Link 4Link 3Link 2Link 1

Only aligned available slots

Link

1

Link 2

Link4

Link3

1 2 3 4 5 6 7 8Slots

Link 4Link 3Link 2Link 1

Aligned continuous available slots

Used slot

Available slot

Fig. 14. Example of continuity and contiguity constraints.

lem, and has been well studied over the last twenty years.The RWA problem is reducible to the RSA problem as thenumber of wavelengths equals the number of spectrum slotsin each fiber link. For any lightpath request, if RWA requires 1wavelength along the lightpath, it is equivalent to a 1 spectrumslot request along the lightpath in the RSA problem. Thisreduction is within poly-nominal time. The RWA problem hasa solution if and only if the constructed RSA problem has asolution. Therefore, from the above discussion, we can saythat the RSA problem is an NP-hard problem [14], [36].

Although RSA is a hard problem, it can simplified bysplitting it into two separate subproblems, namely — (i) therouting subproblem, and (ii) the spectrum allocation subprob-lem. These subproblems are discussed in Section VI andSection VII, respectively.

VI. ROUTING

Approaches for solving the routing subproblem in the elasticoptical network fall into two main groups, namely — (i)routing without elastic characteristics, and (ii) routing withelastic characteristics. In the following, we explain these tworouting approaches.

A. Routing without Elastic Characteristics

This subsection focuses on different routing algorithms[37]–[42], namely — (i) fixed routing, (ii) fixed alternaterouting, (iii) least congested routing, and (iv) adaptive routing,with no consideration given to the elastic characteristics of op-tical networks. These routing approaches are mainly intendedto discover suitable routes between source-destination pairs.These algorithms are discussed below.

1) Fixed Routing: In fixed routing (FR) [1], [38], a sin-gle fixed route is precomputed for each source-destinationpair using some shortest path algorithm, such as Dijkstra’salgorithm [43]. When a connection request arrives in thenetwork, this algorithm attempts to establish a lightpath alongthe predetermined fixed route. It checks whether the requiredslot is available on each link of the predetermined route ornot. If even one link does not have the slot desired, theconnection request is blocked. In the situation when more thanone required slot is available, a spectrum allocation policy isused to select the best slot.

2) Fixed Alternate Routing: Fixed alternate routing (FAR)[1], [38] is an updated version of the FR algorithm. In FAR,each node in the network maintains a routing table (thatcontains an ordered list of a number of fixed routes) forall other nodes. These routes are computed off-line. When aconnection request with a given source-destination pair arrives,the source node attempts to establish a lightpath through eachof the routes from the routing table taken in sequence, until aroute with the required slot is found. If no available route withrequired slot is found among the list of alternate routes, theconnection request is blocked. In the situation when more thanone required slot is available on the selected route, a spectrumallocation policy is used to choose the best slot. Althoughthe computation complexity of this algorithm is higher thanthat of FR, it provides comparatively lower call blockingprobability than the FR algorithm. However, this algorithmmay not be able to find all possible routes between a givensource-destination pair. Therefore, the performance of FARalgorithm in terms of call blocking probability is not optimum.

3) Least Congested Routing: Least congested routing(LCR) [1], [38] predetermines a sequence of routes for eachsource-destination pair similar to FAR. Depending on thearrival time of connection requests, the least-congested routesare selected from among the predetermined routes. The con-gestion on a link is measured by the number of slots availableon the link. If a link has fewer available slots, it is consideredto be more congested. The disadvantage of LCR is its highercomputation complexity; its call blocking probability is almostthe same as that of FAR.

4) Adaptive Routing: In adaptive routing (AR) [38], [39],routes between source-destination pairs are chosen dynami-cally, depending on link-state information of the network. Thenetwork link-state information is determined by the set ofall connections that are currently active. The most acceptableform of AR is adaptive shortest path routing, which is wellsuited for use in optical networks. Under this approach, eachunused spectrum in the network has a cost of 1 unit, whereasthe cost of each used spectrum in the network is taken tobe α. When a connection arrives, the shortest path betweena source-destination pair is determined. If there are multiplepaths with the same distance, one of them is chosen at random.In AR, a connection is considered blocked mainly when thereis no route with a required slot between the source-destinationpair. Since AR considers all possible routes between source-destination pair, it provides lower call blocking probability,but its setup time is comparatively higher than other routingalgorithms. AR requires extensive support from control and

10

management protocols to continuously update the routingtables at the nodes. AR suits centralized implementation rathermore than the distributed alternative.

WA

CA

CA1

UTCO

TX

NE

L

MINY

NJ

MD

PA

GA

α

α

α

α

α

α1 1

1

1

1

11

1

Fig. 15. Fixed (solid line), alternate (dotted line) and adaptive (dashed line)routes are shown from source city CA to destination city L

The functionality of the above mentioned routing algorithmsis explained with the help of the sample example network, seeFig. 15. It consists of 14 nodes (representing cities) and 21bi-directional optical links. The fixed shortest route, alternateroute, and adaptive route from source city CA to destinationcity L are shown by the solid, dotted, and dashed lines,respectively. Furthermore, the congested links are denoted asα. If a connection request for a connection from source cityCA to destination city L arrives, only AR is able to find aroute between CA and L.

5) Comparisons of Routing Algorithms: A significantamount of work on the different issues of routing has beenreported. Table II summarizes the major routing algorithms,and compares their performance in terms of blocking prob-ability, average setup time, and time complexity [127]. Theblocking probability [44], [45] in the network is defined asthe ratio of the number of blocked connection requests to thenumber of connection requests in the network. The averagesetup time [6] in the network is defined as total executiontime required to establish all the connections in the networkto the number of successful connections. From Table II, weobserve that FR has the lowest average setup time and timecomplexity of all routing algorithms. However, its blockingprobability is the highest. AR provides the best performancein terms of blocking probability, but its time complexity is thehighest. FAR offers a trade-off between time complexity andblocking probability.

B. Routing with Elastic Characteristics

An elastic optical network has the capability to slice thespectrum into slots with finer granularity than WDM-basednetworks. M. Jinno et al. [12] presented, for first time, themethod referred to as single slot on the grid approach, seeFig. 16. In this approach, the frequency slots are based on theITU-T fixed grids, where the central frequency is set at 193.1THz. The width of a frequency slot depends on the transmis-sion system. In this example, one frequency slot is 12.5 GHz.According to the bandwidth demand of a connection request, a

group of frequency slots, usually consecutive in the frequencydomain, are allocated.

0 1 2 3 4 5 6 7-1-2-3-4-5-6-7

f=193.1 THz12.5 GHz/slot

Spectrum

Fig. 16. Frequency slot approach for elastic optical networks.

1 3

2

4

Link 1-2

Link 2-3

Link 1-4

Link 4-3

R1(1, 2, 1)

R2(1,

2, 2)

R3 (2, 3, 1)

R4(1, 3, 1)

R6(4,

3, 2)R

5(1, 4, 2)

P1

P2

Fig. 17. Example of multi-path routing (spectrum-split routing) to handlespectrum fragmentation in elastic optical networks.

In elastic optical networks, single path routing via theRSA approach can create the spectrum fragmentation problemin and thus inefficiency. The spectrum fragmentation issueis explained in detail in Section VIII-A. To overcome thisproblem, multi-path routing [79], [80], [126] has been con-sidered for the elastic optical network. An example of thisrouting is shown in Fig. 17. Let us consider connection requestC(S,D, F ), where S, D and F are source, destination, and thenumber of required contiguous sub-carrier slots, respectively.We assume connection requests arrive in time order, and eachguard-band is assumed to be two sub-carrier slots wide. Thegroup of consecutive frequency slots that are available afterspectrum allocation for R1, R2, · · · , R6 is referred to asa spectrum fragment. In this context, if connection requestR7 arrives at node 1 for destination node 3 with demandof four consecutive sub-carrier slots, it is rejected as theyare unavailable. However, two spectrum paths, i.e., P1 andP2, can be established in parallel for R7 by using multipathrouting. Multi-path routing can be used to handle the spectrumfragments that are very common in dynamic traffic scenarios.

VII. SPECTRUM ALLOCATION

With the aim of better fitting the bandwidth requirementsat each moment, the lightpaths established in a network maydynamically change their allocated spectrum. This capability

11

TABLE IISUMMARIES OF DIFFERENT ROUTING ALGORITHMS

Problem Approach ReferencePerformance Analysis

On/Off lineBP AST TCFR [1], [2] Higher BP than

othersShorter AST thanothers

O(L1Z) Off line

Static FAR [1], [2] Lower BP than FR Longer AST thanFR

O(L2KZ) Off line

Rou

ting

LCR [1], [2] Almost same asFAR

Almost same asFAR

O(L3KZ) Off line

Dynamic AR [1], [2] Lower BP thanothers

Longer AST thanothers

O(L2N2Z) On line

L1: Length of the longest fixed route for any node pair, L2: Length of the longest candidate route for any node pair, L3: Number of Hops in the longest candidate route,K: Number of candidate routes for any node pair, N : Number of nodes in the network, and Z: Number of connection requests in the network, BP: Blocking probability,AST: Average setup time, TC: Time complexity

is defined as elastic spectrum allocation [46], [47] and itsimplementation in future flexgrid networks is expected toprovide better network performance. Spectrum allocation maybe performed either after finding a route for a lightpath orin parallel during the route selection process. This sectiondiscusses the different spectrum allocation policies. We cat-egorize the spectrum allocation based on spectrum range forconnection groups and spectrum slot for individual connectionrequest, as presented below.

A. Spectrum Range Allocation for Connection Groups

Policies used to allocate spectrum range for connectiongroups can be categorized into three types, namely — (i) fixedspectrum allocation, (ii) semi-elastic spectrum allocation and(iii) elastic spectrum allocation, based on the changes allowedto the resources allocated to lightpaths in terms of centralfrequency (CF) and spectrum width.

1) Fixed Spectrum Allocation: In the fixed spectrum al-location (fixed SA) policy [46], [47], both CF and assignedspectrum width remain static for ever. At each time period,demands may utilize either whole or only a fraction of theallocated spectrum to convey the bit rate requested for thatperiod. Therefore, this policy does not provide any elasticity.Under this policy, the spectrum allocation of lightpaths isindependent of variations of bandwidth requirements. Whenthe bandwidth demand of a connection is lower than thecapacity of the assigned spectrum, the connection request isestablished. In this case, the utilized spectrum for carryingtraffic is lower than that of allocated spectrum. This can leadto a sub-optimal use of network capacity. When the bandwidthdemand is higher than the capacity of assigned spectrum, somebandwidth is not served.

Figure 18 shows the concept of the fixed-SA policy. Thebandwidth demand for the lightpath at time T is equal to thecapacity of the assigned spectrum. In an underused spectrumcondition, the bandwidth demand is lower than the capacityof the assigned spectrum in time T ′ as shown in Fig. 18(a).Similarly, the bandwidth demand is higher than the capacity ofthe assigned spectrum in time T ′ under insufficient spectrumcondition as shown in Fig. 18(b).

2) Semi-Elastic Spectrum Allocation: In the semi-elasticspectrum allocation (semi-elastic SA) policy [46], [47], theCF remains fixed, but the allocated spectrum width can varyin each time interval. The frequency slices are allocated to

T

Fixed allocated spectrum

CF

T ′

Unused spectrum slots

Utilized spectrum slotsCF

(a)

CF

T

Fixed allocated spectrum

CF

T ′

Insufficient spectrum slots

(b)

Fig. 18. Different conditions (a) underused spectrum condition, and (b)insufficient spectrum condition of fixed spectrum allocation policy.

a lightpath so as to suit the required bandwidth at any time.As a result, the unutilized slots can be used for subsequentconnection requests. Therefore, this spectrum allocation policyprovides higher flexibility than the fixed SA policy. To explainsemi-elastic SA, two scenarios are considered below.

(i) If the required bandwidth demand is reduced, the ca-pacity of the allocated spectrum can also be reduced.The unnecessary spectrum slices at each end of theallocated spectrum can be released and may be used forsubsequent connection requests. Figure 19(a) representsa spectrum slot reduction condition, where both utilizedand allocated spectra of the channel occupy the samenumber of slices at time T ′.

(ii) If the required bandwidth demand is increased, newcontiguous spectrum slices can be allocated at bothends of the CF. The capacity of the allocated spectrumcan be increased in order to serve the maximum re-quired bandwidth. Figure 19(b) represents a spectrumslot expansion condition, where a lightpath increases itsrequired bandwidth from six to eight slots.

3) Elastic Spectrum Allocation: In the elastic spectrumallocation (elastic SA) policy [46], [47], both assigned CF andthe spectrum width can be changed in each time interval. Thisspectrum allocation policy adds a new degree of freedom tothe previous policy. It not only allows the number of slots perlightpath to be varied at any time, but also CF can be changed.Furthermore, we distinguish two conditions that differ in thegrade of flexibility of CF movement as follows.

(i) CF movement within a range: As CF movement islimited to a certain range, the spectrum reallocation is

12

T

Allocated spectrum

CF

(b)

T ′

Spectrum slots expansion

CF

Utilized spectrum slots

TAllocated spectrum

CF

(a)

T ′

Spectrum slots reduction

Utilized spectrum slots

CF

Fig. 19. Different conditions of the semi-elastic spectrum allocation policywith (a) spectrum slot reduction, and (b) spectrum slot expansion.

restricted to neighboring CFs. Figure 20(a) represents alightpath that varies its requirement at time T and T ′.In this figure, both the assigned spectrum width and theCF are varied, but the CF is varied within the rangespecified.

(ii) Elastic spectrum reallocation: This condition reallocatesthe spectrum completely, and there is no CF movementlimitation. The elastic spectrum allocation policy offersthe best spectrum utilization performance among allspectrum allocation policies. Figure 20(b) represents atypical elastic spectrum allocation policy, where twolighpaths are reallocated and their allocated spectrumwidths are varied.

T

T ′

CF

CF

(b)

CF

CF

Re-allocation of the whole channel spectrum

T

T ′

CF

CF

(a)CF movement range

Fig. 20. Different conditions of the elastic spectrum allocation policy with(a) CF movement within a range, and (b) elastic spectrum reallocation.

A comparison of the CF movement approach within a rangeand the reallocation approach is given below. The first onelimits CF movement and so has lower hardware requirements.However, the limitation yields only limited flexibility, whilethe elastic SA policy with reallocation approach is completelyflexible.

B. Comparison of Spectrum Range Allocation policies forConnection Groups

Table III summarizes the different policies available toallocate spectrum range for connection groups in terms ofhardware requirement, control plane, complexity of signalprocedures, computation complexity, spectrum utilization, andnature of spectrum allocation. We observe that both CF andassigned spectrum width remain static in fixed SA policy.However, the control plane can be configured in order to

allocate a fixed channel consisting of a fixed number of slices.The main drawback of the fixed spectrum allocation policy isthe sub-optimal use of capacity, which makes this policy lessconvenient.

Compared to the fixed SA policy, the semi-elastic SA policyrequires additional hardware. Moreover, the control planecan be configured in such way that it can allow establishedlightpaths to be modified. Since the amount of frequency slicesare assigned to each lightpath dynamically, extension of theRSVP-TE protocol [48] should be designed so as to notifyall BV-WXCs along the path to adjust their filter bandwidth,and modify the number of slices allocated to a path. Further-more, some hardware is required to increase or decrease theutilized spectrum as needed. Therefore, bandwidth variabletransponders and bandwidth variable switches should workwith frequency steps in accordance with the frequency slicewidth. The semi-elastic SA policy has better performance thanof the fixed SA policy in terms of spectrum efficiency at thecost of some extra hardware resources.

In the elastic SA policy, extra hardware and a control planeare required to vary both CF and spectrum width dynamically.As both CF and spectrum width vary dynamically, the elasticSA policy provides best performance in terms of spectrumutilization. However, the computation complexity and extrahardware requirements are high compared to other spectrumallocation policies.

C. Spectrum Slot Allocation for Individual Connection Re-quest

The spectrum slot allocation of an individual connectionrequest can be performed using one of the following allocationpolicies.

1) First fit: In the first fit spectrum allocation policy [49],[50], the spectrum slots are indexed and a list of indexes ofavailable and used slots is maintained. This policy alwaysattempts to choose the lowest indexed slot from the list ofavailable slots and allocates it to the lightpath to serve theconnection request. When the call is completed, the slot isreturned to the list of available slots. By selecting spectrumin this manner, existing connections will be packed into asmaller number of spectrum slots, leaving a larger numberof spectrum slots available for future use. Implementing thispolicy, does not require global information of the network.The first fit spectrum allocation policy is considered to be oneof the best spectrum allocation policies due to its lower callblocking probability and computation complexity.

2) Random fit: In the random fit policy [1], [49], a listof free or available spectrum slots is maintained. When aconnection request arrives in the network, this policy randomlyselects a slot from the list of available slots and allocates itto the lightpath used to serve the connection request. Afterassigning a slot to a lightpath, the list of available slots isupdated by deleting the used slot from the free list. Whena call is completed, its slot is added to the list of free oravailable slots. By selecting spectrum in a random manner, itcan reduce the possibility of multiple connections choosingthe same spectrum which is possible if spectrum allocation isperformed in a distributed manner.

13

TABLE IIISUMMARIES OF DIFFERENT SPECTRUM RANGE ALLOCATION POLICIES FOR CONNECTION GROUPS

Fixed SA Semi-Elastic SA Elastic SAHardware Requirement No extra requirement BVT and BV-WXC should allow to increase

or decrease the number of allocated slices.Higher requirements for the range of lasertunability

Control Plane Extension: GMPLS,RSVP-TE, OSPF

Extension in RSVP-TE protocol to modifythe number of assigned slices.

Complex algorithms in PCE to prevent con-flict (simultaneous access to spectrum re-sources).

Complexity of signal procedures Control plane does notneed to notify BV-WXCs along the pathto adjust filter band-width.

RSVP-TE protocol is used to notify all BV-WXCs along the path to adjust their filterbandwidth.

Similar procedure as semi-elastic SA andextra waiting time to adjust filters and lasers

Computation Complexity Lower time complex-ity than others

Higher time complexity than Fixed SA, butlower than Elastic SA

Highest time complexity among all

Spectrum Utilization Lower spectrum uti-lization than others

Higher spectrum utilization than Fixed SA,but lower than Elastic SA

Highest spectrum utilization among all

On/Off line Off line On line On line

SA: Spectrum allocation, BVT: Bandwidth-variable transponder, BV-WXC: Bandwidth-variable wavelength cross-connect, GMPLS: Generalized multi-protocol label switching,RSVP-TE: Resource reservation protocol - traffic engineering, OSPF: Open shortest path first, PCE: Path computation element

TABLE IVSUMMARIES OF DIFFERENT POLICIES TO ALLOCATE SPECTRUM SLOTS TO A SINGLE CONNECTION REQUEST

Spectrum Allocation Policy Reference Time Complexity Applicable networkLeast used [1], [2] O(L1EWZ) Single/multi-fiber networksMost used [1], [2] O(L1EWZ) Single/multi-fiber networksRandom fit [1], [49] O(L1WZ) single/multi-fiber networks

First fit [49], [50] O(L1WZ) single/multi-fiber networksLast fit [1], [22] O(L1WZ) single/multi-fiber networks

Exact fit [49] O(L1WZ) single/multi-fiber networksFirst-last fit [50] O(L1WZ) single/multi-fiber networks

L1: Length of the longest fixed route for any node pair, E: Number of links in the network, M : Number of fibers in the network, N : Numbers of nodesin the network, W : Number of spectrum slots per fiber link, and Z: Number of connection requests in the network

TABLE VNUMERICAL RESULTS OF DIFFERENT SPECTRUM ALLOCATION POLICIES IN RSA IN TERMS OF BLOCKING PROBABILITY [129]

Spectrum Allocation PolicyBlocking probability

20 [Erlang] 40 [Erlang] 60 [Erlang] 80 [Erlang] 100 [Erlang]First-last fit 0.005 0.007 0.010 0.017 0.025

First-exact fit 0.006 0.008 0.012 0.020 0.027First fit 0.006 0.009 0.014 0.021 0.029Last fit 0.007 0.008 0.015 0.020 0.030

Random fit 0.017 0.022 0.031 0.043 0.055

3) Last fit: This policy [1], [22] always attempts to choosethe highest indexed slot from the list of available slots andallocates it to the lightpath to serve the connection request.When the call is completed, the slot is returned back to thelist of available slots.

4) First-last fit: In the first-last fit spectrum allocationpolicy [50], all spectrum slots of each link can be divided intoa number of partitions. The first-last fit spectrum allocationpolicy always attempts to choose the lowest indexed slotsin the odd number partition from the list of available slots.For the even number partitions, it attempts to choose thehighest indexed slots from the list of available spectrum slots.Use of first fit and random fit spectral allocation approachesalways attempt to choose the lowest indexed slots for eachpartition and randomly selects slots, respectively, from thelist of available spectrum slots. This may lead to a situationwhere spectrum slots may be available, but connection requestscannot be established due to unavailability of contiguousaligned slots. The first-last fit allocation policy is expected togive more contiguous aligned available slots than the random

fit and first fit allocation policies.5) Least used: The least used spectrum allocation policy

[1], [2] allocates a spectrum to a lightpath from a list ofavailable spectrum slots that have been used by the fewestfiber links in the network. If several available spectrum slotsshare the same minimum usage, the first fit spectrum allocationpolicy is used to select the best spectrum slot. Selectingspectrum in this manner is an attempt to spread the load evenlyacross all spectrum slots.

6) Most used: The most used spectrum allocation policy[1], [2] assigns spectrum to a lightpath from a list of availablespectrum slots, which have been used by the most fiber linksin the network. Similar to the least used spectrum allocationpolicy, if several available spectrum slots share the samemaximum usage, first fit spectrum allocation policy is usedto break the tie. Selecting spectrum slots in this way is anattempt to realize maximum spectrum reuse in the network.

7) Exact fit: Starting from the beginning of the frequencychannel, the exact fit allocation policy [49] searches for theexact available block in terms of the number of slots requested

14

for the connection. If there is a block that matches the exactsize of requested resources, this policy allocates that spectrum.Otherwise, the spectrum is allocated according to the firstfit spectrum allocation policy. By selecting spectrum slots inthis way, we can reduce the fragmentation problem in opticalnetworks.

Link 1

Aavailable slotOccupied slot

Link 2

Link 3

Link 4

Slots 1 2 3 4 5 6 7 8 9 10 11 12

Link 5

Fig. 21. Spectrum slot usage pattern for a network segment.

To illustrate the functionality of the above mentioned spec-trum allocation policies, we use the example shown in Fig. 21.If two connection requests arrive that use link 2 and link 3with one slot demand for establishing lightpaths, the strategiesproceed as follows. First fit spectrum allocation policy selectsspectrum slot 2 for the first connection request, and slot 3 forthe second connection request. First-last fit spectrum allocationpolicy selects spectrum slot 2 for the first connection request,and slot 12 for the second connection request. Slot 6 and slot 4have been used three times and two times, respectively, in thisexample. Therefore, slot 6 and slot 4 are used by most usedspectrum allocation policies. As slot 2 and slot 9 have not beenused so far, the least used spectrum allocation policy selectsthese two slots for the two connections requests. Random fitallocation policy selects any two of slot 2, slot 3, slot 4 andslot 12 with equal probability. Exact fit spectrum allocationpolicy selects spectrum slot 6 for the first connection request,and slot 2 for the second connection request.

D. Comparisons of Spectrum Allocation policies for Individ-ual Connection Request

A significant amount of published research has addresseddifferent policies to allocate spectrum slots to individualconnection requests. Table IV summarizes some major spec-trum allocation policies. It is observed from Table IV thatthe least-used and most-used allocation policies have highertime complexity than the other allocation policies. These twospectrum allocation policies also require global information ofthe network. As random fit, first fit, last fit, exact fit, first-lastfit spectrum allocation policies have lower time complexity, wepresent the performance of these spectrum allocation policiesin terms of blocking probability under differ traffic loads (inErlang), see Table V. These numerical results [129] wereobtained from a simulation study performed on NSFNET.The simulation parameters followed those of [22]. We observefrom the numerical results that first-last fit is lower blockingprobability than the other spectrum allocation schemes, as itprovides less fragmentation than the other spectrum allocationpolicies. The blocking probability of first-exact fit is higher

than that of first-last fit, but its blocking probability is lowerthan those of other spectrum allocation policies. First fit andlast fit spectrum allocation policies provide almost similarperformance. Finally, the blocking probability of random fitis highest among all spectrum allocation policies.

E. Joint RSA

We have already discussed routing and spectrum allocationin elastic optical networks separately in Section VI andSection VII, respectively. However, many researchers havepresented joint RSA [51]–[53] by considering routing andspectrum allocation at the same time. They usually employ amatrix to describe link or path spectral status by consideringspectrum continuity and contiguity constraints, and choose thebest performance from among all available matrix candidates.

In this direction, X Liu. et al. [51] presented a layer-based approach to design integrated multicast-capable routingand spectrum assignment (MC-RSA) algorithms for servingmulticast requests efficiently and minimizing the bandwidthblocking probability in the elastic optical network. For eachmulticast request, the presented algorithms decompose thephysical topology into several layered auxiliary graphs ac-cording to the network spectrum utilization. Then, basedon the bandwidth requirement, a proper layer is selected,and a multicast light-tree is calculated for the layer. Theseprocedures realize routing and spectrum assignment for eachmulticast request in an integrated manner. Similarly, two jointrouting and spectrum allocation algorithms [52], namely —(i) fragmentation-aware RSA and (ii) fragmentation-awareRSA with congestion avoidance, have been presented by Y.Yin et al. to alleviate spectral fragmentation in the lightpathprovisioning process.

There are some works [14], [54], [55] that focus on solvingjoint RSA with modulation selection. This type of problem isreferred to as the routing, modulation, and spectrum allocation(RMSA) problem. Taking this direction, X. Zhou et al. [54]introduced a RMLSA problem for elastic optical networks. Intheir work, the authors introduced an integer linear program-ming (ILP) for RMLSA algorithm that minimizes the spectrumused to serve the traffic matrix, and presented a decompo-sition method that breaks RMLSA into its two substituentsubproblems, namely (i) routing and modulation level and(ii) spectrum allocation, and solved them sequentially. Finally,they presented a heuristic algorithm, which serves connectionsone-by-one in order to solve the planning problem. In [55],the authors investigated the principle of how dynamic serviceprovisioning fragments the spectral resources on links along apath, and presented corresponding RMSA algorithms to allevi-ate spectrum fragmentation in dynamic network environments.

VIII. VARIOUS ASPECTS RELATED TO RSA AND SBVT

The performance of the elastic optical network depends onnot only its physical resources, like — transponders, physicallinks, usable spectral width, optical switches, etc., but also howthe network is controlled. The objective of an RSA algorithmis to achieve the best performance within the limits set by thephysical constraints. Recently, an increasing number of studies

15

have investigated solutions to the RSA problem in the elasticoptical network. The RSA problem can be cast in numerousforms. This section discusses various issues related to RSAand SBVT as follows.

A. Fragmentation

Elastic optical networks allocate spectrum on contiguoussubcarrier slots. As the size of contiguous subcarrier slots iselastic, it can be a few GHz or even narrower. Therefore,dynamically setting up and tearing down connections cangenerate the bandwidth fragmentation [14], [56] problem. Itis the condition where available slots become isolated fromeach other by being misaligned along the routing path ordiscontiguous in the spectrum domain. Thus, it is difficultto utilize them for upcoming connection requests. If noavailable slot can fulfill the required bandwidth demand ofa connection request, the connection request is consideredto be rejected/blocked. This is called bandwidth blockingand drives network operators to periodically reconfigure theoptical paths and spectrum slots. This is referred to as networkdefragmentation. Figure 22 shows the fragmentation problem.

Slots

Link 1

Link 2

Link 3

1 2 3 4 5 6 7 8 9

Fragmentation problem Fragmentation problem can be minimizedusing defragmentation technique

Slots

Link 1

Link 2

Link 3

1 2 3 4 5 6 7 8 9

Fig. 22. Fragmentation problem, and reducing its effect by incorporatingdefragmentation technique

To overcome the bandwidth fragmentation problem, manyRSA approaches [12], [14], [34], [36], [57] have been pub-lished. In this direction, Kadohata et al. [32] and Zhanget al. [33] developed bandwidth defragmentation schemesby considering the green field scenario, where connectionsare totally rerouted. A. N. Patel, et al. [58] formulated thedefragmentation problem in an elastic optical network asan ILP formulation to provide optimal defragmentation withconsideration of the spectrum continuity and continuity con-straints. They presented two heuristic algorithms, namely —(i) greedy-defragmentation algorithm and (ii) shortest-path-defragmentation for large-scale networks in order to maximizethe spectrum utilization. Fragmentation-aware RSA algorithmsor defragmentation approaches can be classified into two cate-gories, namely — (i) proactive fragmentation-aware RSA and(ii) reactive fragmentation-aware RSA, which are discussed inSection VIII-A1 and Section VIII-A2, respectively.

1) Proactive Fragmentation-aware RSA: When a new re-quest is admitted to the network, the proactive fragmentation-aware RSA technique attempts to prevent or minimize spec-trum fragmentation in the network. In this direction, R. Wanget al. [50] have presented four spectrum management tech-niques for allocating spectrum resources to connections of dif-ferent data rates. In their approaches, all connections share the

whole spectrum using the first-fit spectrum allocation policy.A similar concept of spectrum reservation has been presentedby K. Christodoulopoulos et al. [14]. In their approach, ablock of contiguous subcarriers is reserved for each source-destination pair. In addition, subcarriers that are not reservedmay be shared among all connections on demand.

2) Reactive Fragmentation-aware RSA: In a dynamic en-vironment, the fragmentation problem cannot be completelyeliminated. Therefore, reactive fragmentation-aware RSA al-gorithms attempt to restore the network’s ability to accommo-date high-rate and long-path connections. The main objectiveof defragmentation is to reconfigure the spectrum allocationof existing connections in order to consolidate available slotsinto large contiguous and continuous blocks that may be usedfor upcoming connection requests. In this direction, R.Wanget al. [50] have presented a set of reactive defragmentationstrategies that exploit hitless optical path shift (HOPS). HOPStechnology shifts a connection to a new block of spectrumas long as the route of the connection does not change andthe movement of the new spectrum does not affect other estab-lished connections. They presented a scheme that consolidatesthe spectrum slots freed by a terminated connection with otherblocks of spectrum available along the links of its path.

Most of the approaches [12], [14], [34], [36], [50], [56],[57] in the literature perform bandwidth defragmentation af-ter bandwidth fragmentation occurs in subcarrier-slots. Thismeans that the traffic is disrupted by connection rerouting,as the bandwidth defragmentation is performed. Bandwidthdefragmentation that uses connection rerouting increases thetraffic delay and system complexity.

To overcome this serious issue, R. Wang and B. Mukher-jee [21] have presented a scheme that prevents bandwidthfragmentation without performing any connection rerouting.Typically, when connection requests with lower-bandwidth andhigher-bandwidth are not separated during spectrum alloca-tion, it is more likely that the higher-bandwidth connectionrequests are blocked. In order to circumvent this drawback,they explore an admission control mechanism that capturesthe unique challenges posed by heterogeneous bandwidths.They adopt a preventive admission control scheme based onspectrum partitioning to achieve higher provisioning efficiency.As a result, it prevents the blocking of connections due to theunfairness of bandwidth issue.

Similarly, W. Fadini et al. [22] have presented a subcarrier-slot partition scheme for spectrum allocation in elastic opticalnetworks; it reduces the number of non-aligned available slotswithout connection rerouting. Thus the bandwidth blockingprobability in the network is reduced. In their approach, theydefine a connection group as a set of connections whoseroutes are exactly the same. When the spectrum resources oftwo connections from different connection groups sharing acommon link are allocated to adjacent slots, some availableslots might be non-aligned with each other. When anotherconnection request arrives in the network and its route needsthese available slots, the connection request is rejected ifthese available slots are non-aligned. The partitioning schemedivides the subcarrier slots into several partitions. Disjointconnections whose routes do not share any link are allocated to

16

the same partition, while non-disjoint connections are allocatedto different partitions. In this way, their approach increases thenumber of aligned available slots in the network and hence thebandwidth blocking probability is reduced.

B. Modulation

The traditional WDM-based optical network assigns spec-trum resources to optical paths without considering the ap-propriate modulation technique, which leads to an inefficientutilization of the spectrum. However, the OFDM-based elasticoptical network allocates optical paths with consideration ofadaptive modulation and bit rate to further improve the spec-trum efficiency. In the modulation-based spectrum allocationscheme of [12], [59], the necessary minimum spectral resourceis adaptively allocated to an optical path. The adaptationconsiders the physical conditions while ensuring a constantdata rate. The modulation-based spectrum allocation schemeimproves the spectrum efficiency, as the allocated spectralbandwidth can be reduced for shorter paths by increasing thenumber of modulated bits per symbol.

In this direction, M. Jinno et al. [12] have presented adistance-adaptive spectrum allocation scheme that adopts ahigh-level modulation format for long distance paths, and alow-level modulation format to shorter paths. As the opticalsignal-to-noise ratio (OSNR) tolerance of 64-QAM is lowerthan that of QPSK, it suits shorter distance lightpaths as shownin Fig. 23.

BPSK

QPSK

8QAM

16QAM

Mod

ulat

ion

leve

l

L1L2L3L4

Transmission distance

L1 > L2 > L3 > L4 > · · ·

Fig. 23. Modulation level versus transmission distance.

The modulation based spectrum allocation schemes can beclassified into two categories, namely — (i) offline modulationbased spectrum allocation, and (ii) online modulation basedspectrum allocation, and are discussed below.

1) Offline Modulation based Spectrum Allocation: K.Christodoulopoulos et al. [14] have presented an offline mod-ulation based spectrum allocation scheme, where a mappingfunction is provided as input to the problem. In their scheme,each demand is mapped to a modulation level according tothe requested data rate and the distance of the end-to-endpath. Initially, they presented a path-based ILP formulationfor their scheme, and then decomposed the problem into twosub-problems, namely (i) routing and modulation level (RML),and (ii) spectrum allocation. They solved the sub-problemssequentially using ILPs. Finally, a sequential algorithm waspresented to serve connections one-by-one, and to solve theplanning problem sequentially.

2) Online Modulation based Spectrum Allocation: Most ofthe studies on online modulation based spectrum allocation[60]–[63] have introduced heuristic algorithms, which dealwith randomly arriving connection requests. Initially, thesealgorithms compute a number of fixed-alternate paths foreach source-destination pair, and arrange them in decreasingorder of their end-to-end path length. In the second step, aspectrum allocation policy is used to allocate a lightpath toeach connection request by considering alternate path routingand modulation.

Recent studies [15], [60], [61] on modulation based spec-trum allocation claim that this type of spectrum allocationscheme increases the spectral utilization by approximately 9%-60% compared to fixed-modulation based spectrum alloca-tion in the elastic optical network. Fixed-modulation basedspectrum allocation schemes do not consider the most appro-priate modulation technique for different connection requestsaccording to their lightpath distance. Typically, they select,conservatively, one modulation technique for all connectionrequests regardless their lightpath distance. As an example,a fixed-modulation based spectrum allocation scheme adoptsBPSK modulation format for all connection requests regardlesstheir lightpath distance. As a result, this type of spectrumallocation schemes does not utilize the spectrum efficiently. Onthe other hand, modulation-based spectrum allocation schemesdetermine the modulation technique that best suits each light-path distance. As an example, a modulation-based spectrumallocation scheme adopts BPSK for long distance lightpaths,and 16-QAM for shorter distance lightpaths. This minimizesthe numbers of spectrum slots that must be assigned, whichyields better utilization of spectrum resources compared tofixed-modulation based spectrum allocation schemes.

C. Quality of Transmission

The elastic optical network architecture offers the abilityto choose the modulation format and channel bandwidth tosuit the transmission distance and quality of transmissiondesired. One version of the online modulation based spectrumallocation scheme is referred to as quality-of-transmissionaware RSA. In this direction, H. Beyranvand et al. [63]have presented a quality of transmission (QoT) aware onlineRSA scheme for the elastic optical network. Their approachemploys three steps, namely — (i) path calculation, (ii)paths election, and (iii) spectrum assignment to constructthe complete framework. The Dijkstra and k-shortest pathalgorithms have been adapted for computing paths, whilefiber impairments and non-linear effects on the physical layerare modeled to estimate the QoT along the given route. S.Yang et al. have presented [62] a QoT-aware RSA scheme inorder to select a feasible path for each requested connectionand allocate subcarrier slots by using modulation formatsappropriate for the transmission reach and requested data rate.

D. Traffic Grooming

In WDM-based wavelength-routed optical networks, trafficgrooming [1], [5], [64]–[70] is used to multiplex a number oflow-speed connection requests into a high-capacity wavelength

17

channel for enhancing channel utilization. Traffic groomingimproves the resource utilization by aggregating multipleelectrical channels (packet or circuit flows) onto one opticalchannel.

The elastic optical network allocates spectral resources withjust enough bandwidth to satisfy the traffic demands. However,traffic grooming [26], [67], [71]–[73] is still essential for thefollowing reasons, (i) BVT is normally designed so as tomaximize the traffic rate in the network, and it does not supportslicing at a very early stage [74]. Electrical traffic groomingis applied in order to use transponder capacity efficiently. (ii)Generally speaking, a filter guard band between two adjacentchannels should be assigned to resolve optical filter issues.Traffic grooming can minimize filter guard band usage byaggregating traffic electrically. The electrical switching fabricis still needed for traffic grooming in the elastic opticalnetwork, similar to WDM networks. The main difference isthat the transponder used in elastic optical networks does notstrictly follow the ITU-T central frequency. As a result, it canprovide flex-optical channels.

To further improve the flexibility and eliminate the electricalprocessing, researchers have designed SBVT for the elasticoptical network. In SBVT-based elastic optical networks, thetraffic grooming [26], [75] function can be partly offloadedfrom the electrical layer to the optical layer. Multiple electricalchannels are groomed onto one sub-transponder channel, inwhich each sub-transponder channel is associated with aflex-optical channel. Multiple sub-transponder channels (flex-optical channels) are groomed optically onto one transponderby using an optical switching fabric (e.g., BV-OXC), whichis called optical traffic grooming. Figure 24 distinguishes be-tween traffic grooming in WDM-based optical networks, trafficgrooming with BVT in elastic optical networks, and trafficgrooming with SBVT in elastic optical networks. Spectrumefficiency and transponder usage are improved in WDM-basedoptical networks, whereas traffic grooming with BVT in theelastic optical network improves transponder usage and re-duces guard band usage. Finally, traffic grooming with SBVTin the elastic optical network eliminates electrical processingby offloading parts of the grooming function to the opticallayer.

Y. Zhang et al. [71] have incorporated, for the first time,a grooming approach for the RSA problem in the elasticoptical network with BVT. In their approach, multiple low-speed connection requests are groomed into elastic opticalpaths by using electrical layer multiplexing. They presenteda mixed integer linear program (MILP) formulation to re-duce the average spectrum utilization in the traffic-groomingscenario. S. Zhang et al. [67] have presented a multi-layerauxiliary graph to implement various traffic-grooming policiesby properly adjusting the edge weights in the auxiliary graph.With their approach, they have shown that there is a trade-off among different traffic-grooming policies, and that thespectrum reservation scheme can be incorporated into varioustraffic-grooming scenarios. Recently, J. Zhang et al. [26] havepresented dynamic traffic grooming in SBVT-enabled elasticoptical networks. In their approach, a three-layered auxiliarygraph (AG) model has been presented to address mixed-

electrical-optical grooming under the dynamic traffic scenario.By adjusting the edge weights of AG, various traffic-groomingpolicies can be achieved for different purposes. Furthermore,two spectrum reservation schemes have been introduced inorder to efficiently utilize transponder capacity. Finally, theycompared different traffic-grooming policies under two spec-trum reservation schemes, and the tradeoff among the policieswas shown.

E. Survivability

The elastic optical network has the capability to support in-dividual data rates from 400-1000 Gb/s [76]. It also aggregatesthe throughput per fiber link to approximately 10-100 Tb/s.Therefore, failure of a network component, such as opticalfiber or network node, can disrupt communications for millionsof users, which can lead to a great loss of data and revenue.As an example, in 2004, the Gartner Research Group had lostapproximately 500 million dollars due to failure of a opticalnetwork [77]. Thus, survivability against failure has becomean essential requirement of the elastic optical network. Failurerecovery [78] is defined here as “the process of re-establishingtraffic continuity in the event of a failure condition affectingthat traffic, by re-routing the signals on diverse facilitiesafter the failure”. A network is defined as survivable [78],if its recovery can be secured rapidly. Similar to WDM-basedoptical networks, the survivability mechanisms [79]–[81] forthe elastic optical network can be classified into two broadcategories, namely — (i) protection and (ii) or restoration,which are discussed briefly in the following subsections.

1) Protection: The protection techniques of [81]–[85] usebackup paths to carry optical signals after fault occurrence.The backup paths are computed prior to fault occurrence, butthey are reconfigured after fault occurrence. In this direction,M. Klinkowski et al. [85] have presented an RSA approachwith dedicated protection for static traffic demands. Althoughdedicated protection can provide more reliability, it is unableto utilize the spectrum slots properly as some of the spectrumslots are reassigned prior to fault occurrence. To overcomethis problem, M. Liu et al. [81] presented a shared protectionscheme to enhance the spectrum utilization by sharing backupspectrum slots between two adjacent paths on a link, if thecorresponding working paths are link-disjoint. They exploredthe opportunity of sharing enabled by tunable transponders.The elasticity of the transponder enables the expansion andcontraction of paths. As a result, the backup spectrum is usedby only one of the adjacent paths at a time. Similarly, G. Shenet al. [82] addressed a shared protection technique for elasticoptical networks to minimize both required spare capacity andmaximum number of used link spectrum slots.

2) Restoration: In restoration [80], [86]–[91], backup pathsare computed dynamically on the basis of link-state informa-tion after fault occurrence, and hence can provide more effi-ciency in terms of resource utilization compared to protection.In this direction, F. Ji et al. [92] presented three algorithms fordynamic preconfigured-cycle (p-cycle) configuration in orderto provide the elastic optical network with 100% restorationagainst single-link failure. The first algorithm configures the

18

ESF

ESF: Electrical switch fabric

Electrical channel

Transponder (fixed)

(a)

R1

R2 R3R1 R2 R3

Electrical grooming

ESF

Electrical channel

Transponder (flexible)

(b)

R1

R2 R3R1 R2 R3

Electrical grooming

ESF

Electrical channel

Sub-transponder (flexible)

(c)

R1

R2 R3R1 R2 R3

Electrical grooming SBVT

Optical groomingOSF

Transponder (flexible)

OSF: Optical switch fabric

Transponder(fixed)

Transponder(flexible)

Fig. 24. Comparison of traffic grooming in (a) WDM-based optical networks, (b) elastic optical networks with BVTs, and (c) elastic optical networks withSBVTs.

working path and p-cycles of a request together according tothe protection efficiencies of the p-cycles. In order to reducethe blocking probability, they presented a spectrum planningtechnique that regulates the working spectrum and protectionresources, and finally, two algorithms based on the protectedworking capacity envelope cycles and Hamiltonian cycles.

F. Paolucci et al. [91] have presented a restoration techniqueenabling multipath recovery and bit rate squeezing in theelastic optical network. They exploited the advanced flexibilityprovided by sliceable bandwidth-variable transponders thatsupport the adaptation of connection parameters in terms of thenumber of sub-carriers, bit rate, transmission parameters, andreserved spectrum resources. They formulated their problemsas an ILP model and finally, presented an heuristic algorithm

that efficiently recovers network failures by exploiting limitedportions of the spectrum resources along multiple routes. Asrestoration finds the backup paths after fault occurrence, itoffers slower recovery than protection. Depending on the typeof rerouting used, restoration can be considered as consistingof three categories, namely — (i) link restoration, (ii) pathrestoration and (iii) segment-based restoration. Link restora-tion discovers a backup path for the failed connection onlyaround the failed link. In path restoration, the failed connectionindependently discovers a backup path on an end-to-end basis.Segment-based restoration discovers a segment backup path ofthe failed connection.

19

F. Energy Saving

The energy consumption of telecom networks is drasticallyincreasing with the increase in traffic. The IP router con-sumes the maximum amount of energy in IP-over WDM-based optical networks [94]. When transmission rate increases,the optical transponder associated with the IP router is ahuge energy consumer in optical networks [95]. Therefore tominimize energy consumption, it is essential to reduce thenumber of IP router ports and optical transponders. By usingthe advantages of SBVT, the elastic optical network can offersome new features for traffic grooming (shown in Fig. 24)and optical layer bypass, which can help to reduce the energyconsumption. In this direction, J. Zhang et al. [96] studiedthe power consumption of IP-over-elastic optical networkswith different elastic optical transponders. The results of theirstudies show that significant energy savings are possible ifSBVT is used rather than the fixed BVT.

Recently, researchers [93], [97], [98] have focused on en-ergy efficient RSA schemes for the elastic optical network.In this direction, A. Fallahpour el al. [93] have presenteda dynamic energy efficient RSA algorithm that considersregenerator placement to suppress the total network energyconsumption. In their work, a virtual graph is designed basedon the given network topology, where the cost functions ofthe virtual graph are computed according to the energy con-sumption of the corresponding links and intermediate routers.Furthermore, a newly arrived connection request is servedby finding the most energy-efficient path among the possiblecandidate paths. Similarly, S. Zhang et al. [98] have presentedenergy-efficient dynamic provisioning in order to significantreduce energy consumption and make efficient use of spectrumresources. In their research work, they adopt an auxiliarygraph, and from it created a dynamic provisioning policycalled time-aware provisioning with bandwidth reservation(TAP-BR). The TAP-BR policy incorporates the two importantfactors of time awareness and bandwidth reservation, in orderto facilitate energy-efficient provisioning.

Several studies on energy saving in the elastic opticalnetwork are anticipated, and more research work is neededto develop truly effective energy-saving RSA schemes.

G. Networking cost of SBVT

This subsection discusses the networking cost reductionmade possible by the use of SBVTs in the elastic optical net-work. We have observed from the literature that SBVTs allowthe reuse of hardware and optical spectrum by transmittingdata to multiple destinations. SBVTs enable point to multiplepoint transmission where the traffic rate to each destinationand the number of destinations can be freely set to satisfythe request. On the other hand, the non-sliceable transponderrequires at least one interface for each destination, whichincreases networking cost.

In this direction, V. Lopez et al. [25], [27] have presentedtwo node models, namely — (i) non-sliceable transpondermodel, and (ii) SBVT model in order to compare the net-working cost, please see Fig. 25. The main difference betweenthese two models is that the non-sliceable transponder model

requires at least one interface for each destination, while theSBVT transponder reuses hardware and optical spectrum totransmit to multiple destinations. The model without sliceabletransponders considers coherent modulation formats, such as— 40 Gb/s, 100 Gb/s, 400 Gb/s and 1 Tb/s, whereas only 400Gb/s or 1 Tb/s SBVTs are considered by the SBVT model.

RX TX

Traffic demand in Gb/s

Transponder40 Gb/s, 100 Gb/s,400 Gb/s or 1 Tb/s

RXTXRX TX

RXTX

Traffic demand in Gb/s

SBVT400 Gb/s or 1 Tb/s

RX TXRXTX

(a)

(b)

Fig. 25. Different models (a) non-sliceable transponder model, and (b) SBVTmodel for the analysis of networking cost.

Finally, the result of the research work [25] has claimed thatusing 400 Gb/s and 1 Tb/s SBVTs reduces transponder costsin the network by at least 50% in a core network scenario.This reduction was calculated relative to BVTs of 400 Gb/sand 1 Tb/s in the non-sliceable scenario.

A significant number of works have addressed variousaspects of the RSA problem in the elastic optical network. Ta-ble VI summarizes these different aspects, namely — fragmen-tation, modulation, quality-of-transmission, traffic grooming,survivability, energy saving, and networking cost reductionsby SBVT.

TABLE VISUMMARIES OF DIFFERENT ISSUES RELATED TO RSA

Various issues Reference

Fragmentation Proactive [14],[21], [22]Reactive [12], [14], [34], [36], [50], [56],

[57]

Modulation Online [60]–[63]Offline [14]

Quality of transmission [62], [63]Traffic grooming [67], [71], [72], [91],[26], [73]

SurvivabilityProtection [81]–[84]Restoration [80], [86]–[90],[92]

Energy saving [93],[97], [98]Networking cost caused by SBVT [25], [27]

20

IX. EXPERIMENTAL DEMONSTRATIONS

This section presents the experimental demonstrations thathave tested the functionality of the elastic optical network. Thefirst experimental demonstration was conducted by M. Jinnoet al. [99]. They demonstrated a spectrum-efficient elasticoptical path network for 100 Gb/s services and beyond; it usedflexible rate transceivers and BV-WXCs. Subsequently, manyresearchers moved from the conventional WDM-based opticalnetwork to the elastic optical network architecture.

In this direction, B. Kozicki et al. [100] presented an exper-imental demonstration on the architecture of an elastic opticalnetwork. They used the optical OFDM modulation format andBV-WXCs to generate, transmit and receive optical paths withbandwidths of up to 1 Tb/s. They experimentally demonstratedelastic optical path setup and spectrally-efficient transmissionof multiple channels with bit rates ranging from 40 to 140Gb/s between six nodes in a mesh network. Furthermore, theydemonstrated multi-hop transmission in a single-mode fiberover 400 km with 1 Tb/s data rate. Finally, they investigatedthe filtering properties and the required guard bandwidth forthe spectrally-efficient allocation of optical paths in the elasticoptical network. Similarly, their other research work [101]experimentally demonstrated optical path aggregation in aspectrum-sliced elastic optical network. Multiple OFDM 100-Gb/s optical paths are aggregated in the optical domain to forma spectrally continuous 1-Tb/s super-wavelength optical pathand transmitted over a network of BV-WXCs. Finally, theyevaluated the potential implementation issues and concludedthat OFDM paths can be optically aggregated with an opticalsignal-to-noise ratio penalty of less than 1 dB.

A flexible-bandwidth network testbed with a real-time adap-tive control plane has been demonstrated by D. J. Geisleret al [102]. In their experiment, the modulation format andspectrum-positioning were adjusted to maintain QoT and highspectral efficiency. They have presented a performance mon-itoring method that dynamically reconfigures the modulationformat in such a way that the optical signals maintain therequired QoT and bit error rate even for signals that experiencetime-varying impairments. J. Zhang et al. [103] demonstratedenhanced software defined networking (SDN) over elastic gridoptical networks for data center service migration.

J. Zhang et al. [104] presented an experimental demonstra-tion of the openflow-based control plane for elastic lightpathprovisioning in flex-grid optical networks. Dynamic light-path establishment and adjustment were implemented in theircontrol plane testbed. Additionally, they reported the overalllatency including signaling and hardware for lightpath setupand adjustment.

A path computation element (PCE) architecture for flex-ible optical networks has been demonstrated in [105] thatmaximizes the spectral efficiency. Confirmation was providedthrough two different experiments; they successfully showedthe PCE ability to trigger dynamic rerouting with bit-rate ormodulation format adaptation. In particular, the experimentsdemonstrated, in a testbed, the dynamic allocation of spectrumslots and adoption of modulation formats from 16-QAM toQPSK at 100 Gb/s, and bit-rate adaptation with 16-QAM from

200 Gb/s to 100 Gb/s.Finally, S. Ma et al. [106] have presented and experimen-

tally demonstrated a control-plane framework to realize onlinespectrum defragmentation in software-defined elastic opticalnetworks in order to reduce the call blocking probability inthe network.

In conclusion, a significant number of experimental demon-strations have been reported in the literature. Table VI sum-marizes different experimental demonstrations related to theelastic optical network. However, the practical implementationof the elastic optical network remains in development, and thefull commercial deployment of the elastic optical network willbe seen in the next few years.

TABLE VIISUMMARIES OF DIFFERENT EXPERIMENTAL DEMONSTRATIONS

Experimental demonstrations Reference

Architecture Data plane [99], [100], [107]Control/management plane [101], [103], [105],

[108]–[110]

RSA

QoT [102]Survivability [111]Modulation [102], [110], [112]

Defragmentation [106], [113], [114]

X. RESEARCH ISSUES AND CHALLENGES

Flex-grid technology or elastic optical network is a promis-ing concept but its implementation remains some way off.There are several issues and challenges, which need fur-ther research to resolve [115], [116]. Taking this direction,I. Tomkos et al. [115] reviewed the recent developments on theresearch topic of flexible/elastic networking and highlightedthe future research challenges. In [116], the authors provideda comprehensive view of the different pieces composing the“flexible networking puzzle” with special attention given tocapturing the occurring interactions between different researchfields. In their work, physical layer technological aspects,network optimization for flexible networks, and control planeaspects were examined. Furthermore, future research direc-tions and open issues were presented. Figure 26 summarizesthe different areas demanding further work. In the following,we identify some interesting research opportunities for theelastic optical network.

Elastic optical network

Energy consumption Control plane requirmentCost Performance

Connection level Network level

Capacity Transparent reach Spectrum efficiency

Utilized interface AvailabilityCall blockingUtilize spectrum

Fig. 26. Different research areas in elastic optical networks

21

A. Hardware Development

Innovative and sophisticated devices and components mustbe developed in order to achieve high capacity spectrumefficient elastic optical networks. Novel optical switching andfiltering elements need to be developed in order to provide effi-cient client protocol data unit mapping procedures that extractthe incoming client signal via client-specific physical codingsublayers and media access controller layers, high resolutionand steep filtering performance, optimum modulation formatfor bandwidth variability and higher nonlinear impairmenttolerance, etc.

One of the important challenges faced by the researchcommunity is to develop a new sliceable bandwidth-variabletransponder, which supports sliceability, multiple bit rates,multiple modulation formats, and code rate adaptability. How-ever, 100 Gb/s OFDM transponders can adapt to lower bitrates. Therefore, fractional-bandwidth services may be pro-vided with the use of the same device. This kind of technologyand the use of optical integrated circuits may offer compactand cost-effective implementation. Similarly, the need formultiple temperature-stabilized, frequency controlled laserscan be improved by phase-locked carrier generation from asingle laser source.

Recently, space division multiplexing (SDM) technology[117], [118], [128] has been incorporated into elastic opticalnetworks in order to develop high-capacity, next-generation,and few-mode/multicore fiber infrastructures. The realizationof this type of infrastructure should be enabled by thedevelopment of novel multi-dimensional spectral switchingnodes, which can be fabricated by extending the designs ofexisting flexible SSS nodes, through the addition of advancedmode/core adapting techniques.

To achieve long transmission reach, optical signals must beamplified at periodic regeneration points along the fiber span tocompensate the power loss experienced in multi-core fiber. Forregeneration, one technique is to demultiplex the SDM signalsinto multiple single-core fibers and then amplify the signalsin each fiber using conventional single-core EDFAs. Theamplified signals are then recombined and injected back intothe multi-core fiber span, which increases the system delay.Therefore, to develop a single-core power transient-suppressedEDFA (TS-EDFAs) is one of the challenging research goalsthat must be accomplished; a key issue is to reduce the timetaken to adjust the operation point of the amplifier since anewly added signal may suddenly change the total power atthe EDFA input.

B. Network Control and Management

Traditional optical networks use a set of protocols (suchas — generalized multi-protocol label switching (GMPLS),open shortest path first (OSPF), and resource reservationprotocol - traffic engineering (RSVP)) for network controland management. Although these control protocols are welldesigned and standardized for traditional optical networks,evaluations that encompass the elastic technology are stillat an early stage. The control plane of the elastic opticalnetwork must support many unique properties, such as — (i)

optical channels are allowed to be flexible in size or width,(ii) optical channels can support various modulation formats,(iii) sub and super wavelength concept, (iv) support of multi-path routing of the composing waveband members of a splitspectrum super-channel, (v) fast restoration upon failure, withor without resource reservation, etc. Therefore, new controlprotocols need to be developed or the existing protocols shouldbe extended in order to support these unique properties of theelastic optical network.

During the last decade, the GMPLS protocol has beenbroadly standardized in different aspects including packetswitching, label-switched paths (LSPs), and time-division mul-tiplexing. However, its applicability to the elastic optical net-work has not been completely described yet. For this, GMPLSneeds to address frequency slots instead of wavelengths. Thecontrol plane protocols are required to maintain coherentglobal information representing up to 320 or 640 slots, for12.5 GHz or 6.25 GHz slot granularity, respectively [119]. G.Shen et al. [120] have indicated in their research work thatspectrum granularity may become even finer reaching 3 GHz,resulting in 1280 possible slots. Some research works [119]–[121] have been addressed management of the control plane.Extensive research is need to adequately manage the controlplane of the elastic optical network.

From the operation and control viewpoint, extendingsoftware-defined networking (SDN) toward transport net-works, while retaining flexibility, is a challenging issue thatneeds to be addressed properly. In this direction, the OpenNetworking Foundation group [122] are addressing SDN andOpenFlow-based control capabilities for optical transport net-works. In their research work, many activities such as — (i)identifying use cases, and (ii) defining reference architecturein order to control optical transport networks by incorporatingthe OpenFlow standard, have been performed to developOpenFlow protocol extensions. Based on their model, theextended OpenFlow protocol is responsible for interfacingwith network elements. The control virtual network interface isdeveloped in order to provide the required bridge between thedata center controllers. Future extensions and additional stan-dardization activities are needed in order to realize the SDNcontrollers that can manage TDM circuit and wavelength-based architectures (such as — generic packet/TDM/fiberswitching, bandwidth aggregation and segmentation). Finally,an efficient technology needs to be developed in order tosupport a combination of centralized and distributed controlof a multi-layer network.

C. Energy Consumption

The increase in the traffic in carried by optical networks willincrease the energy consumption. The elastic optical networkhas the ability to significantly reduce energy consumption. Incombination with SBVTs, the elastic optical network presentssome new features as regards optical traffic grooming andoptical layer bypass [26], which can help to reduce the energyconsumption. However, we still are unable to groom trafficoptically in a very early stage, and this omission must betackled.

22

The elastic optical network offers lower blocking proba-bility compared to traditional optical networks and so canaccept higher volumes of traffic. This clearly is a significantadvantage in terms of energy efficiency, as the deployment ofadditional network elements would not only increase cost, butalso increase the overall energy consumption. Several energysaving schemes are anticipated, setting some the networkelements into sleep mode when the traffic is below a certainthreshold. Another interesting topic for future researchers is toanalyze the energy efficiency of new protection and restorationschemes for the elastic optical network.

D. Physical Layer Impairments

As optical connections may span over many long links,physical layer impairments (PLIs), such as — dispersion, inter-ference, noise, and nonlinear effects accumulate and degradethe signal quality, which affects the quality-of-transmission(QoT). Accounting for PLIs is a challenging issue for networkdesigners, especially if we consider exact models and theinterdependencies. Many studies [45], [123]–[125] on PLIshave been carried out for WDM-based optical networks. PLIshave distinct impact on both WDM-based optical networksand elastic optical networks. With the introduction of co-herent detection and digital signal processing, impairmentsthat are related to dispersion can be substantially reduced orfully compensated. However, high levels of flexibility makethe minimization of these effects more complicated from analgorithmic perspective, which needs further research.

E. Spectrum Management

One of the most important problems in elastic opticalnetworks, both for planning and operation, is allocating net-work resources in a dynamic environment. The problem ofestablishing connections in fixed-grid WDM-based networksand the allocation of network resources is well addressed inthe literature [1], [2]. However, connection establishment inthe elastic optical network is more complicated for severalreasons. First, in contrast to WDM networks where eachconnection is assigned a single wavelength, in elastic opticalnetworks, spectrum slots can be allocated in a flexible manner.Apart from the difference in spectrum resource allocation, thechoice of the transmission parameters of the tunable transpon-ders present in flexible networks directly or indirectly impactsthe resource allocation decision and makes the problem evenmore complicated.

Some proposals [22] found in the literature partition theentire spectrum in an advance in order to handle spectrumresources efficiently. Most of these schemes assume the trafficdemand in an advance. However, considering the fact that thetraffic profile will change over time, connections with largersize slots demand will have to be accommodated over thesame partitions. In this case, it is obvious that there is noother way than dropping these connections. As a result, theblocking performance will decrease significantly. Even worse,larger slot connections are the ones that will be droppedmost often. Therefore, it is very important to take account ofpossible changes in the traffic profile. Therefore, partitioning

the spectrum in a dynamic traffic environment is a challengingissue, which needs further research.

F. Disaster Management

Recovery of the network resources after large scale disasterssuch as — tsunami, hurricanes, floods, etc, is becomingincreasingly important. The approach to disaster recovery isdifferent from the approach to network failures such as fibercuts or node failures. Network failures can be planned for byproviding sufficient resources to deal with all network failures.However, there is no way to anticipate the impact of a disaster,and therefore no cost effective way to plan for full recoveryfrom it. The recent development of SBVTs represents a newtool for handling disaster recovery. As SBVTs can play animportant role in providing sufficient flexibility, elastic opticalnetworks with SBVTs are an interesting research topic fordisaster management.

XI. CONCLUDING REMARKS

The elastic network paradigm is an emerging research area,and a promising technology for future high-speed transmissiondue to its flexible properties. This tutorial has introduced anddiscussed different aspects of the elastic optical network.

We started with the basic concept of the elastic opticalnetwork and its unique properties, and then turned to itsarchitecture and operation principle. The architecture of SBVTand its advantages in the future optical networking weredetailed. Immediately after discussing network architecture,our discussion focused on the different node architectures,namely — broadcast and select, spectrum routing, switch andselect with dynamic functionality, and architecture on demand,along with their functionalities. Next we looked into the basicconcept of the RSA approach in elastic optical networks. Then,we moved to different routing approaches along with their prosand cons. Thereafter, the different spectrum allocation policieswere addressed. In addition, we distinguished the spectrumallocation polices into two categories based on spectrum rangeallocation for connection groups and spectrum slot allocationfor the individual connection request.

Various aspects, namely — fragmentation, modulation,quality-of-transmission, traffic grooming, survivability, energysaving, and networking cost related to RSA, which havebeen reported in the literature, were addressed in this tuto-rial paper. We classified them as either fragmentation-awareapproaches or defragmentation approaches and explained howthese approaches deal with fragmentation. Distance adaptiveRSA that considers the modulation scheme is one of the newpossibilities of the elastic optical network, and was also beenpresented in this paper. We have discussed and comparedtraffic grooming in WDM-based optical networks, elasticoptical networks with BVTs, and elastic optical networkswith SBVTs. Different survivability techniques, namely —protection and restoration, were also covered. We discussedthe networking cost reduction made possible by the use ofSBVTs in elastic optical networks.

Finally, we explored the experimental demonstrations thathave tested the functionality of the elastic optical network.

23

Research challenges and open issues that flexible networkingposes were presented to guide future research.

REFERENCES

[1] B. Mukherjee, Optical WDM Networks. Springer, 2006.[2] R. M. C. Siva and G. Mohan, WDM Optical Networks: Concepts,

Design and Algorithms. PHI, 2003.[3] B. C. Chatterjee, N. Sarma, and P. P. Sahu, “Priority Based Routing and

Wavelength Assignment With Traffic Grooming for Optical Networks,”J. of Opt. Commun. and Netw., vol. 4, no. 6, pp. 480–489, 2012.

[4] ——, “Priority based Dispersion-Reduced Wavelength Assignment forOptical Networks,” IEEE/OSA J. of Lightwave Technol., vol. 31, no. 2,pp. 257–263, 2013.

[5] K. Zhu, H. Zang, and B. Mukherjee, “A Comprehensive Study on Next-Generation Optical Grooming Switches,” IEEE J. Sel. Areas Commun.,vol. 21, no. 7, pp. 1173–1186, 2003.

[6] B. C. Chatterjee, N. Sarma, and P. P. Sahu, “A Heuristic Priority basedWavelength Assignment Scheme for Optical Networks,” Optik - Int. J.Light Electron. Opt., vol. 123, no. 17, pp. 1505–1510, 2012.

[7] “How Much Bandwidth Do We Need?”http://arstechnica.com/business/2012/05/bandwidth-explosion-as-internet-use-soars-can-bottlenecks-be-averted/, Last Accessed: 2015-01-30.

[8] M. Jinno, H. Takara, and B. Kozicki, “Dynamic Optical Mesh Net-works: Drivers, Challenges and Solutions for the Future,” in 35thEuropean Conference on Optical Communication (ECOC). IEEE,2009, pp. 1–4.

[9] S. Roy, A. Malik, A. Deore, S. Ahuja, O. Turkcu, S. Hand, andS. Melle, “Evaluating Efficiency of Multi-Layer Switching in FutureOptical Transport Networks,” in National Fiber Optic Engineers Con-ference. OSA, 2013, pp. NTh4J–2.

[10] C. Saradhi and S. Subramaniam, “Physical Layer Impairment AwareRouting (PLIAR) in WDM Optical Networks: Issues and Challenges,”IEEE Commun. Surveys & Tuts., vol. 11, no. 4, pp. 109–130, 2009.

[11] M. Jinno, H. Takara, B. Kozicki, Y. Tsukishima, Y. Sone, and S. Mat-suoka, “Spectrum-Efficient and Scalable Elastic Optical Path Network:Architecture, Benefits, and Enabling Technologies,” IEEE Commun.Mag., vol. 47, no. 11, pp. 66–73, 2009.

[12] M. Jinno, B. Kozicki, H. Takara, A. Watanabe, Y. Sone, T. Tanaka,and A. Hirano, “Distance-Adaptive Spectrum Resource Allocationin Spectrum-sliced Elastic Optical Path Network [Topics in OpticalCommunications],” IEEE Commun. Mag., vol. 48, no. 8, pp. 138–145,2010.

[13] O. Gerstel, M. Jinno, A. Lord, and S. B. Yoo, “Elastic OpticalNetworking: A New Dawn for the Optical Layer?” IEEE Commun.Mag., vol. 50, no. 2, pp. s12–s20, 2012.

[14] K. Christodoulopoulos, I. Tomkos, and E. Varvarigos, “ElasticBandwidth Allocation in Flexible OFDM-Based Optical Networks,”IEEE/OSA J. of Lightwave Technol., vol. 29, no. 9, pp. 1354–1366,2011.

[15] G. Zhang, M. De Leenheer, A. Morea, and B. Mukherjee, “A Surveyon OFDM-Based Elastic Core Optical Networking,” IEEE Commun.Surveys & Tuts., vol. 15, no. 1, pp. 65–87, 2013.

[16] S. Talebi, F. Alam, I. Katib, M. Khamis, R. Salama, and G. N. Rouskas,“Spectrum Management Techniques for Elastic Optical Networks: ASurvey,” Opt. Switch. and Network., vol. 13, pp. 34–48, 2014.

[17] J. Socrates-Dantas, D. Careglio, J. Perello, R. M. Silveira, W. V.Ruggiero, and J. Sole-Pareta, “Challenges and Requirements of AControl Plane for Elastic Optical Networks,” Computer Networks,vol. 72, pp. 156–171, 2014.

[18] G. Keiser, Optical Fiber Communications. McGraw-Hill, 1991.[19] X. Lagrange and B. Jabbari, Multiaccess, Mobility and Teletraffic for

Wireless Communications. Springer, 1999.[20] D. Medhi, Network Routing: Algorithms, Protocols, and Architectures.

Morgan Kaufmann, 2010.[21] R. Wang and B. Mukherjee, “Spectrum Management in Heterogeneous

Bandwidth Networks,” Global Communications Conference (GLOBE-COM), IEEE, 2012, pp. 2907–2911.

[22] W. Fadini and E. Oki, “A Subcarrier-slot Partition Scheme for Wave-length Assignment in Elastic Optical Networks,” in International Con-ference on High Performance Switching and Routing (HPSR). IEEE,2014, pp. 7–12.

[23] M. Jinno, H. Takara, Y. Sone, K. Yonenaga, and A. Hirano, “MultiflowOptical Transponder for Efficient Multilayer Optical Networking,”IEEE Commun. Mag., vol. 50, no. 5, pp. 56–65, 2012.

[24] N. Sambo, A. DErrico, C. Porzi, V. Vercesi, M. Imran, F. Cugini,A. Bogoni, L. Potı, and P. Castoldi, “Sliceable Transponder Archi-tecture Including Multiwavelength Source,” J. of Opt. Commun. andNetw., vol. 6, no. 7, pp. 590–600, 2014.

[25] V. Lopez, B. d. l. Cruz, O. Gonzalez de Dios, O. Gerstel, N. Amaya,G. Zervas, D. Simeonidou, and J. P. Fernandez-Palacios, “Finding theTarget Cost for Sliceable Bandwidth Variable Transponders,” J. of Opt.Commun. and Netw., vol. 6, no. 5, pp. 476–485, 2014.

[26] J. Zhang, Y. Ji, M. Song, Y. Zhao, X. Yu, and B. Mukherjee, “Dy-namic Traffic Grooming in Sliceable Bandwidth-Variable TransponderEnabled Elastic Optical Networks,” IEEE/OSA J. of Lightwave Technol.(Accepted), 2015.

[27] V. Lopez, O. Gonzalez De Dios, O. Gerstel, N. Amaya, G. Zervas,D. Simeonidou, and J. P. Fernandez-Palacios, “Target Cost for SliceableBandwidth Variable Transponders in a Real Core Network,” in FutureNetwork and Mobile Summit (FutureNetworkSummit). IEEE, 2013,pp. 1–9.

[28] S. Frisken, G. Baxter, D. Abakoumov, H. Zhou, I. Clarke, and S. Poole,“Flexible and Grid-Less Wavelength Selective Switch using LCOSTechnology,” in OFC. OSA, 2011, p. OTuM3.

[29] R. Ryf, Y. Su, L. Moller, S. Chandrasekhar, X. Liu, D. T. Neilson, andC. R. Giles, “Wavelength Blocking Filter With Flexible Data Rates andChannel Spacing,” IEEE/OSA J. of Lightwave Technol., vol. 23, no. 1,pp. 54–61, 2005.

[30] O. Rival and A. Morea, “Elastic Optical Networks with 25–100G Format-versatile WDM Transmission Systems,” in 15th Optoe-Electronics and Communications Conference (OECC). IEEE, 2010,pp. 100–101.

[31] N. Amaya, G. Zervas, and D. Simeonidou, “Introducing Node Archi-tecture Flexibility for Elastic Optical Networks,” J. of Opt. Commun.and Netw., vol. 5, no. 6, pp. 593–608, 2013.

[32] A. Kadohata, A. Hirano, M. Fukutoku, T. Ohara, Y. Sone, andO. Ishida, “Multi-Layer Greenfield Re-Grooming With WavelengthDefragmentation,” IEEE Commun. Lett., vol. 16, no. 4, pp. 530–532,2012.

[33] M. Zhang, W. Shi, L. Gong, W. Lu, and Z. Zhu, “Bandwidth Defrag-mentation in Dynamic Elastic Optical Networks With Minimum TrafficDisruptions,” in International Conference on Communications (ICC).IEEE, 2013, pp. 3894–3898.

[34] M. Zhang, C. You, H. Jiang, and Z. Zhu, “Dynamic and AdaptiveBandwidth Defragmentation in Spectrum-Sliced Elastic Optical Net-works With Time-Varying Traffic,” IEEE/OSA J. of Lightwave Technol.,vol. 32, no. 5, pp. 1014–1023, 2014.

[35] N. Amaya, G. Zervas, and D. Simeonidou, “Architecture on Demandfor Transparent Optical Networks,” in 13th International Conferenceon Transparent Optical Networks (ICTON). IEEE, 2011, pp. 1–4.

[36] Y. Wang, X. Cao, and Y. Pan, “A Study of the Routing and SpectrumAllocation in Spectrum-Sliced Elastic Optical Path Networks,” inInternational Conference on Computer Communications (INFOCOM).IEEE, 2011, pp. 1503–1511.

[37] S. Subramaniam and R. Barry, “Wavelength Assignment in Fixed Rout-ing WDM Networks,” in International Conference on Communications(ICC). Montreal, IEEE, 1997, pp. 406–410.

[38] R. Ramamurthy and B. Mukherjee, “Fixed-Alternate Routing andWavelength Conversion in Wavelength-Routed Optical Networks,”IEEE/ACM Trans. on netw., vol. 10, no. 3, pp. 351–367, 2002.

[39] J. Jue and G. Xiao, “An Adaptive Routing Algorithm for Wavelength-Routed Optical Networks With a Distributed Control Scheme,” in 9thInternational Conference on Computer Communications and Networks(ICCCN). IEEE, 2002, pp. 192–197.

[40] A. Castro, L. Velasco, M. Ruiz, M. Klinkowski, J. FernaNdez-Palacios,and D. Careglio, “Dynamic Routing and Spectrum (Re)allocationin Future Flexgrid Optical Networks,” Computer Networks, vol. 56,no. 12, pp. 2869–2883, 2012.

[41] X. Wan, N. Hua, and X. Zheng, “Dynamic Routing and SpectrumAssignment in Spectrum-Flexible Transparent Optical Networks,” J. ofOpt. Commun. and Netw., vol. 4, no. 8, pp. 603–613, 2012.

[42] K. Chan and T. Yum, “Analysis of Least Congested Path Routingin WDM Lightwave Networks,” in 13th International Conference onComputer Communications (INFOCOM). IEEE, 1994, pp. 962–969.

[43] T. H. Cormen, Introductions to Algorithms. McGraw-Hill Companies,2003.

[44] B. C. Chatterjee, N. Sarma, and P. P. Sahu, “A Priority based Wave-length Assignment Scheme for Optical Network,” in Internationalworkshop on Network Modeling and Analysis (IWNMA), 2011, pp. 1–6.

24

[45] ——, “Dispersion-Reduction Routing and Wavelength Assignment forOptical Networks,” in 2nd International Conference on Trends inOptics and Photonics (IConTOP), 2011, pp. 456–463.

[46] M. Klinkowski, M. Ruiz, L. Velasco, D. Careglio, V. Lopez, andJ. Comellas, “Elastic Spectrum Allocation for Time-varying Traffic inFlexgrid Optical Networks,” IEEE J. Sel. Areas Commun., vol. 31,no. 1, pp. 26–38, 2013.

[47] A. A. Garcia, “Elastic Spectrum Allocation in Flexgrid Optical Net-works,” Universitat Politcnica de Catalunya, Tech. Rep., 2012.

[48] L. Berger, “Generalized Multi-Protocol Label Switching (GMPLS)Signaling Resource Reservation Protocol-Traffic Engineering (RSVP-TE) Extensions,” IETF RFC 3473, Jan. 2003.

[49] A. Rosa, C. Cavdar, S. Carvalho, J. Costa, and L. Wosinska, “SpectrumAllocation Policy Modeling for Elastic Optical Networks,” in 9thInternational Conference on High Capacity Optical Networks andEnabling Technologies (HONET). IEEE, 2012, pp. 242–246.

[50] R. Wang and B. Mukherjee, “Spectrum Management in HeterogeneousBandwidth Optical Networks,” Opt. Switch. and Network., vol. 11, pp.83–91, 2014.

[51] X. Liu, L. Gong, and Z. Zhu, “Design Integrated RSA for Multicastin Elastic Optical Networks with a Layered Approach,” in GlobalCommunications Conference (GLOBECOM). IEEE, 2013, pp. 2346–2351.

[52] Y. Yin, H. Zhang, M. Zhang, M. Xia, Z. Zhu, S. Dahlfort, and S. Yoo,“Spectral and Spatial 2D Fragmentation-Aware Routing and SpectrumAssignment Algorithms in Elastic Optical Networks [Invited],” J. ofOpt. Commun. and Netw., vol. 5, no. 10, pp. A100–A106, 2013.

[53] M. Zhang, W. Shi, L. Gong, W. Lu, and Z. Zhu, “Bandwidth Defrag-mentation in Dynamic Elastic Optical Networks with Minimum TrafficDisruptions,” in International Conference on Communications (ICC).IEEE, 2013, pp. 3894–3898.

[54] X. Zhou, W. Lu, L. Gong, and Z. Zhu, “Dynamic RMSA in ElasticOptical Networks with an Aadaptive Genetic Algorithm,” in GlobalCommunications Conference (GLOBECOM). IEEE, 2012, pp. 2912–2917.

[55] Y. Yin, Z. Zhu, S. Yoo et al., “Fragmentation-Aware Routing, Mod-ulation and Spectrum Assignment Algorithms in Elastic Optical Net-works,” in Optical Fiber Communication Conference (OFC). OSA,2013, pp. OW3A–5.

[56] P. S. Khodashenas, J. Comellas, S. Spadaro, J. Perello, and G. Junyent,“Using Spectrum Fragmentation to Better Allocate Time-Varying Con-nections in Elastic Optical Networks,” J. of Opt. Commun. and Netw.,vol. 6, no. 5, pp. 433–440, 2014.

[57] Y. Sone, A. Hirano, A. Kadohata, M. Jinno, and O. Ishida, “Routing andSpectrum Assignment Algorithm Maximizes Spectrum Utilization inOptical Networks,” in European Conference and Exposition on OpticalCommunications (ECOC). OSA, 2011, pp. 1–3.

[58] A. N. Patel, P. N. Ji, J. P. Jue, and T. Wang, “Routing, Wavelength As-signment, and Spectrum Allocation Algorithms in Transparent FlexibleOptical WDM Networks,” Opt. Switch. and Network, vol. 9, no. 3, pp.191–204, 2012.

[59] Z. Ding, Z. Xu, X. Zeng, T. Ma, and F. Yang, “Hybrid Routingand Spectrum Assignment Algorithms based on Distance-AdaptationCombined Co-Evolution and Heuristics in Elastic Optical Networks,”Optical Engg., vol. 53, no. 4, pp. 046 105–046 105, 2014.

[60] T. Takagi, H. Hasegawa, K. Sato, T. Tanaka, B. Kozicki, Y. Sone, andM. Jinno, “Algorithms for Maximizing Spectrum Efficiency in ElasticOptical Path Networks that Adopt Distance Adaptive Modulation,”in European Conference and Exposition on Optical Communications(ECOC), IEEE, 2010, pp. 1–3.

[61] T. Takagi, H. Hasegawa, K.-i. Sato, Y. Sone, B. Kozicki, A. Hirano,and M. Jinno, “Dynamic Routing and Frequency Slot Assignmentfor Elastic Optical Path Networks That Adopt Distance AdaptiveModulation,” in Optical Fiber Communication Conference (OFC).OSA, 2011, p. OTuI7.

[62] S. Yang and F. Kuipers, “Impairment-Aware Routing in TranslucentSpectrum-Sliced Elastic Optical Path Networks,” in 17th EuropeanConference on Networks and Optical Communications (NOC). IEEE,2012, pp. 1–6.

[63] H. Beyranvand and J. A. Salehi, “A Quality-of-Transmission AwareDynamic Routing and Spectrum Assignment Scheme for Future ElasticOptical Networks,” IEEE/OSA J. of Lightwave Technol., vol. 31, no. 18,pp. 3043–3054, 2013.

[64] A. Chiu and E. Modiano, “Traffic Grooming Algorithms for ReducingElectronic Multiplexing Costs in WDM Ring Networks,” IEEE/OSA J.of Lightwave Technol., vol. 18, no. 1, pp. 2–12, 2000.

[65] J. Wang, W. Cho, V. Vemuri, and B. Mukherjee, “Improved Approachesfor Cost-Effective Traffic Grooming in WDM Ring Networks: ILPFormulations and Single-Hop and Multi-Hop Connections,” IEEE/OSAJ. of Lightwave Technol., vol. 19, no. 11, pp. 1645–1653, 2001.

[66] M. Liu, M. Tornatore, and B. Mukherjee, “Survivable Traffic Groomingin Elastic Optical Networks — Shared Path Protection,” in Interna-tional Conference on Communications (ICC). IEEE, 2012, pp. 6230–6234.

[67] S. Zhang, C. Martel, and B. Mukherjee, “Dynamic Traffic Groomingin Elastic Optical Networks,” IEEE J. Sel. Areas Commun., vol. 31,no. 1, pp. 4–12, 2013.

[68] K. Zhu and B. Mukherjee, “A Review of Traffic Grooming in WDMOptical Networks: Architectures and Challenges,” Opt. Networks Mag.,vol. 4, no. 2, pp. 55–64, 2003.

[69] ——, “Traffic Grooming in an Optical WDM Mesh Network,” IEEEJ. Sel. Areas Commun., vol. 20, no. 1, pp. 122–133, 2002.

[70] E. Modiano, “Traffic Grooming in WDM Networks,” IEEE Commun.Mag., vol. 39, no. 7, pp. 124–129, 2001.

[71] Y. Zhang, X. Zheng, Q. Li, N. Hua, Y. Li, and H. Zhang, “TrafficGrooming in Spectrum-Elastic Optical Path Networks,” in OpticalFiber Communication Conference (OFC), OSA, 2011, p. OTuI1.

[72] G. Zhang, M. De Leenheer, and B. Mukherjee, “Optical TrafficGrooming in OFDM-Based Elastic Optical Networks [Invited],” J. ofOpt. Commun. and Netw, vol. 4, no. 11, pp. B17–B25, 2012.

[73] S. Zhang, M. Tornatore, G. Shen, J. Zhang, and B. Mukherjee, “Evolv-ing Traffic Grooming in Multi-Layer Flexible-Grid Optical NetworksWith Software-Defined Elasticity,” IEEE/OSA J. of Lightwave Technol.,vol. 32, no. 16, pp. 2905–2914, 2014.

[74] B. Kozicki, H. Takara, and M. Jinno, “Enabling Technologies for Adap-tive Resource Allocation in Elastic Optical Path Network (SLICE),”in Communications and Photonics Conference and Exhibition (ACP),2010, pp. 23–24.

[75] B. Jiawei Zhang; Yongli Zhao; Xiaosong Yu; Jie Zhang; Mukherjee,“Auxiliary Graph Model for Dynamic Traffic Grooming in ElasticOptical Networks With Sliceable Optical Transponder,” in EuropeanConference on Optical Communication (ECOC), 2014, pp. 1–3.

[76] P. Winzer, “Beyond 100G Ethernet,” IEEE Commun. Mag., vol. 48,no. 7, pp. 26–30, 2010.

[77] W. Fawaz and K. Chen, “Survivability-Oriented Quality of Servicein Optical Networks,” in Quality of Service Engineering in NextGeneration Heterogenous Networks , A. Mellouk, Ed. London, UK:Wiley Online Library, 2010, pp. 197–211.

[78] E. Bouillet, G. Ellinas, J.-F. Labourdette, and R. Ramamurthy, PathRouting in Mesh Optical Networks. Wiley Online Library, 2007.

[79] L. Ruan and N. Xiao, “Survivable Multipath Routing and SpectrumAllocation in OFDM-based Flexible Optical Networks,” J. of Opt.Commun. and Netw., vol. 5, no. 3, pp. 172–182, 2013.

[80] L. Ruan and Y. Zheng, “Dynamic Survivable Multipath Routing andSpectrum Allocation in OFDM-Based Flexible Optical Networks,” J.of Opt. Commun. and Netw., vol. 6, no. 1, pp. 77–85, 2014.

[81] M. Liu, M. Tornatore, and B. Mukherjee, “Survivable Traffic Groomingin Elastic Optical Networks — Shared Protection,” IEEE/OSA J. ofLightwave Technol, vol. 31, no. 6, pp. 903–909, 2013.

[82] G. Shen, Y. Wei, and S. K. Bose, “Optimal Design for Shared BackupPath Protected Elastic Optical Networks Under Single-Link Failure,”J. of Opt. Commun. and Netw., vol. 6, no. 7, pp. 649–659, 2014.

[83] K. Walkowiak, M. Klinkowski, B. Rabiega, and R. Goscien, “Routingand Spectrum Allocation Algorithms for Elastic Optical Networks WithDedicated Path Protection,” Opt. Switch. and Network., vol. 13, pp. 63–75, 2014.

[84] J. Wu, Y. Liu, C. Yu, and Y. Wu, “Survivable Routing and SpectrumAllocation Algorithm based on P-Cycle Protection in Elastic OpticalNetworks,” Optik - Int. J. Light Electron. Opt., vol. 125, no. 16, pp.4446–4451, 2014.

[85] M. Klinkowski and K. Walkowiak, “Offline RSA Algorithms for ElasticOptical Networks with Dedicated Path Protection Consideration,” in4th International Congress on Ultra Modern Telecommunications andControl Systems and Workshops (ICUMT), 2012, pp. 670–676.

[86] A. Giorgetti, F. Paolucci, F. Cugini, and P. Castoldi, “Fast Restorationin SDN-Based Flexible Optical Networks,” in Optical Fiber Commu-nication Conference (OFC). OSA, 2014, pp. Th3B–2.

[87] Y. Wei, G. Shen, and S. K. Bose, “Span-Restorable Elastic OpticalNetworks Under Different Spectrum Conversion Capabilities,” IEEETrans. on Reliability, vol. 63, no. 2, pp. 401–411, 2014.

[88] B. Chen, J. Zhang, Y. Zhao, C. Lv, W. Zhang, S. Huang, X. Zhang,and W. Gu, “Multi-Link Failure Restoration With Dynamic Load

25

Balancing in Spectrum-Elastic Optical Path Networks,” Opt. FiberTechnol., vol. 18, no. 1, pp. 21–28, 2012.

[89] Y. Sone, A. Watanabe, W. Imajuku, Y. Tsukishima, B. Kozicki,H. Takara, and M. Jinno, “Highly Survivable Restoration SchemeEmploying Optical Bandwidth Squeezing in Spectrum-sliced ElasticOptical Path (SLICE) Network,” in Optical Fiber CommunicationConference (OFC). OSA, 2009, p. OThO2.

[90] ——, “Bandwidth Squeezed Restoration in Spectrum-sliced ElasticOptical Path Networks (SLICE),” J. of Opt. Commun. and Netw., vol. 3,no. 3, pp. 223–233, 2011.

[91] F. Paolucci, A. Castro, F. Cugini, L. Velasco, and P. Castoldi, “Multi-path Restoration and Bitrate Squeezing in SDN-Based Elastic OpticalNetworks [Invited],” Photon. Net. Commun., Vol. 28, no. 1, pp. 45–57,2014.

[92] F. Ji, X. Chen, W. Lu, J. J. Rodrigues, and Z. Zhu, “Dynamic P -Cycle Configuration in Spectrum-Sliced Elastic Optical Networks,” inGlobal Communications Conference (GLOBECOM). IEEE, 2013, pp.2170–2175.

[93] A. Fallahpour, H. Beyranvand, S. A. Nezamalhosseini, and J. A. Salehi,“Energy Efficient Routing and Spectrum Assignment With RegeneratorPlacement in Elastic Optical Networks,” IEEE/OSA J. of LightwaveTechnol, vol. 32, no. 10, pp. 2019–2027, 2014.

[94] G. Shen and R. Tucker, “Energy-Minimized Design for IP Over WDMNetworks,” J. of Opt. Commun. and Netw., vol. 1, no. 1, pp. 176–186,2009.

[95] W. Van Heddeghem, F. Idzikowski, W. Vereecken, D. Colle, M. Pick-avet, and P. Demeester, “Power Consumption Modeling in OpticalMultilayer Networks,” Photon. Net. Commun., vol. 24, no. 2, pp. 86–102, 2012.

[96] J. Zhang, Y. Zhao, J. Zhang, and B. Mukherjee, “Energy Efficiency ofIP-over-Elastic Optical Networks with Sliceable Optical Transponder,”in Optical Fiber Communication Conference (OFC). OSA, 2014, pp.W3A–4.

[97] J. Zhang, Y. Zhao, X. Yu, J. Zhang, M. Song, Y. Ji, and B. Mukherjee,“Energy-Efficient Traffic Grooming in Sliceable-Transponder-EquippedIP-Over-Elastic Optical Networks [Invited],” J. of Opt. Commun. andNetw., vol. 7, no. 1, pp. A142–A152, 2015.

[98] S. Zhang and B. Mukherjee, “Energy-Efficient Dynamic Provisioningfor Spectrum Elastic Optical Networks,” in International Conferenceon Communications (ICC). IEEE, 2012, pp. 3031–3035.

[99] M. Jinno, H. Takara, B. Kozicki, Y. Tsukishima, T. Yoshimatsu,T. Kobayashi, Y. Miyamoto, K. Yonenaga, A. Takada, O. Ishidaet al., “Demonstration of Novel Spectrum-Efficient Elastic OpticalPath Network with Per-channel Variable Capacity of 40 Gb/s to Over400 Gb/s,” in 34th European Conference on Optical Communication(ECOC). IEEE, 2008, pp. 1–2.

[100] B. Kozicki, H. Takara, Y. Tsukishima, T. Yoshimatsu, K. Yonenaga,and M. Jinno, “Experimental Demonstration of Spectrum-Sliced ElasticOptical Path Network (SLICE),” Opt. Express, vol. 18, no. 21, pp.22 105–22 118, 2010.

[101] B. Kozicki, H. Takara, Y. Tsukishima, T. Yoshimatsu, T. Kobayashi,K. Yonenaga, and M. Jinno, “Optical Path Aggregation for 1-Tb/sTransmission in Spectrum-Sliced Elastic Optical Path Network,” IEEEPhoton. Technol. Lett., vol. 22, no. 17, pp. 1315–1317, 2010.

[102] D. J. Geisler, R. Proietti, Y. Yin, R. P. Scott, X. Cai, N. Fontaine,L. Paraschis, O. Gerstel, and S. J. B. Yoo, “The First TestbedDemonstration of a Flexible Bandwidth Network With A Real-TimeAdaptive Control Plane,” in 37th European Conference and Expositionon Optical Communications (ECOC). OSA, 2011, p. Th.13.K.2.

[103] J. Zhang, Y. Zhao, H. Yang, Y. Ji, H. Li, Y. Lin, G. Li, J. Han,Y. Lee, and T. Ma, “First Demonstration of Enhanced Software DefinedNetworking (eSDN) Over Elastic Grid (eGrid) Optical Networks forData Center Service Migration,” in Optical Fiber CommunicationConference (OFC). OSA, 2013, p. PDP5B.1.

[104] J. Zhang, J. Zhang, Y. Zhao, H. Yang, X. Yu, L. Wang, and X. Fu,“Experimental Demonstration of OpenFlow-Based Control Plane forElastic Lightpath Provisioning in Flexi-Grid Optical Networks,” Opt.Express, vol. 21, no. 2, pp. 1364–1373, 2013.

[105] F. Cugini, G. Meloni, F. Paolucci, N. Sambo, M. Secondini, L. Gerardi,L. Potı, and P. Castoldi, “Demonstration of Flexible Optical NetworkBased on Path Computation Element,” IEEE/OSA J. Lightwave Tech-nol., vol. 30, no. 5, pp. 727–733, 2012.

[106] S. Ma, C. Chen, S. Li, M. Zhang, S. Li, Y. Shao, Z. Zhu, L. Liu, andS. J. B. Yoo, “Demonstration of Online Spectrum DefragmentationEnabled by OpenFlow in Software-Defined Elastic Optical Networks,”in Optical Fiber Communication Conference (OFC). OSA, 2014, p.W4A.2.

[107] H. Takara, T. Goh, K. Shibahara, K. Yonenaga, S. Kawai, and M. Jinno,“Experimental Demonstration of 400 Gb/s Multi-flow, Multi-Rate,Multi-Reach Optical Transmitter for Efficient Elastic Spectral Routing,”in European Conference and Exposition on Optical Communications(ECOC). OSA, 2011, pp. Tu–5.

[108] L. Liu, R. Munoz, R. Casellas, T. Tsuritani, R. Martınez, and I. Morita,“OpenSlice: An OpenFlow-Based Control Plane for Spectrum SlicedElastic Optical Path Networks,” Opt. Express, vol. 21, no. 4, pp. 4194–4204, 2013.

[109] M. Channegowda, R. Nejabati, M. Rashidi Fard, S. Peng, N. Amaya,G. Zervas, D. Simeonidou, R. Vilalta, R. Casellas, R. Martınezet al., “Experimental Demonstration of An OpenFlow Based Software-Defined Optical Network Employing Packet, Fixed and Flexi-ble DWDM Grid Technologies on An International Multi-DomainTestbed,” Opt. Express, vol. 21, no. 5, pp. 5487–5498, 2013.

[110] R. Casellas, R. Munoz, J. M. Fabrega, M. S. Moreolo, R. Martinez,L. Liu, T. Tsuritani, and I. Morita, “Design and Experimental Valida-tion of a GMPLS/PCE Control Plane for Elastic CO-OFDM OpticalNetworks,” IEEE J. Sel. Areas Commun., vol. 31, no. 1, pp. 49–61,2013.

[111] X. Cai, K. Wen, R. Proietti, Y. Yin, R. P. Scott, C. Qin, and S. Yoo,“Experimental Demonstration of Adaptive Combinational QoT FailureRestoration in Flexible Bandwidth Networks,” in Optical Fiber Com-munication Conference (OFC). OSA, 2012, p. PDP5D–1.

[112] X. Jin, R. Giddings, and J. Tang, “Experimental Demonstration ofAdaptive Bit and/or Power Loading for Maximizing Real-Time End-to-End Optical OFDM Transmission Performance,” in Optical FiberCommunication Conference (OFC). OSA, 2011, p. JWA029.

[113] R. Proietti, R. Yu, K. Wen, Y. Yin, and S. Yoo, “Quasi-HitlessDefragmentation Technique in Elastic Optical Networks by a CoherentRX LO With Fast TX Wavelength Tracking,” in Photonics in SwitchingConference, 2012, pp. 1–3.

[114] R. Munoz, R. Casellas, R. Martınez, L. Liu, T. Tsuritani, and I. Morita,“Experimental Evaluation of Efficient Routing and Distributed Spec-trum Allocation Algorithms for GMPLS Elastic Networks,” Opt. Ex-press, vol. 20, no. 27, pp. 28532–28537, 2012.

[115] I. Tomkos, E. Palkopoulou, and M. Angelou, “A Survey of RecentDevelopments on Flexible/Elastic Optical Networking,” in 14th In-ternational Conference on Transparent Optical Networks (ICTON).IEEE, 2012, pp. 1–6.

[116] I. Tomkos, S. Azodolmolky, J. Sole-Pareta, D. Careglio, andE. Palkopoulou, “A Tutorial on the Flexible Optical NetworkingParadigm: State of the Art, Trends, and Research Challenges,” Vol. 102,no. 9, pp. 1317–1337, 2014.

[117] K. Abedin, T. Taunay, M. Fishteyn, D. DiGiovanni, V. Supradeepa,J. Fini, M. Yan, B. Zhu, E. Monberg, and F. Dimarcello, “Cladding-Pumped Erbium-Doped Multicore Fiber Amplifier,” Opt. Express,vol. 20, no. 18, pp. 20191–20200, 2012.

[118] B. C. Chatterjee, Y. Jayabal, N. Yamanaka, S. Okamoto, A. Fumagalli,and E. Oki, “Span Power Management Scheme for Rapid LightpathProvisioning in Multi-Core Fibre Networks,” IET Elect. Lett., vol. 51,no. 1, pp. 76–78, 2014.

[119] N. Sambo, F. Cugini, G. Bottari, P. Iovanna, and P. Castoldi, “Dis-tributed Setup in Optical Networks With Flexible Grid,” in Euro-pean Conference and Exposition on Optical Communications (ECOC).OSA, 2011, p. We–10.

[120] G. Shen and Q. Yang, “From Coarse Grid to Mini-Grid to Gridless:How Much Can Gridless Help Contentionless?” in Optical FiberCommunication Conference (OFC). OSA, 2011, p. OTuI3.

[121] R. Casellas, R. Munoz, J. M. Fabrega, M. S. Moreolo, R. Martınez,L. Liu, T. Tsuritani, and I. Morita, “GMPLS/PCE Control of Flexi-GridDWDM Optical Networks Using CO-OFDM Transmission [Invited],”J. of Opt. Commun. and Netw., vol. 4, no. 11, pp. B1–B10, 2012.

[122] “Open Networking Foundation (ONF) : Technical Library,”https://www.opennetworking.org/sdn-resources/technical-library,Last Accessed: 2015-01-30.

[123] A. Kretsis, P. Kokkinos, K. Christodoulopoulos, T. Varvarigou, andE. M. Varvarigos, “Mantis: Cloud-Based Optical Network Planningand Operation Tool,” Computer Networks, vol. 77, pp. 153–168, 2015.

[124] H. Pereira, D. Chaves, C. Bastos-Filho, and J. Martins-Filho, “OSNRModel to Consider Physical Layer Impairments in Transparent OpticalNetworks,” Photon. Net. Commun., vol. 18, no. 2, pp. 137–149, 2009.

[125] B. C. Chatterjee, N. Sarma, and P. P. Sahu, “A QoS-Aware WavelengthAssignment Scheme for Optical Networks,” Optik - Int. J. LightElectron. Opt., vol. 124, no. 20, pp. 4498–4501, 2013.

[126] M. Zhang, W. Shi, L. Gong, W. Lu, and Z. Zhu, “Multipath Routingin Elastic Optical Networks With Distance-Adaptive Modulation For-

26

mats,” in International Conference on Communications (ICC). IEEE,2013, pp. 3915–3920.

[127] B. C. Chatterjee, N. Sarma, and P. P. Sahu, “Review and PerformanceAnalysis on Routing and Wavelength Assignment Approaches forOptical Networks,” IETE Tech. Rev., vol. 30, no. 01, pp. 12–23, 2013.

[128] E. Oki, B. C. Chatterjee, Y. Jayabal, N. Yamanaka, S. Okamoto, andA. Fumagalli, “Performance Evaluation of Span Power Control Schemefor Fast Optical Lightpath Provisioning in Multi-Core Fiber Networks,”in 17th International Conference on Transparent Optical Networks(ICTON). IEEE, 2015, (to appear).

[129] B. C. Chatterjee, and E. Oki, “Performance Evaluation of SpectrumAllocation Policies for Elastic Optical Networks,” in 17th InternationalConference on Transparent Optical Networks (ICTON). IEEE, 2015,(to appear).

Bijoy Chand Chatterjee received the PhD de-gree from the Department of Computer Science& Engineering, Tezpur University, Assam in 2014.Presently, he is working as a Postdoctoral Re-searcher in the Department of Communication En-gineering and Informatics, the University of Electro-Communications, Tokyo, Japan. During his PhD, hewas awarded an IETE Research Fellowship by TheInstitute of Electronics and Telecommunication En-gineers (IETE), India. His research interests includeQoS-aware protocols, cross-layer design, optical net-

works and elastic optical networks. He is a member of IEEE, and a lifemember of IETE.

Dr. Nityananda Sarma received B E (ComputerScience & Engineering), M Tech (Computer Science& Engineering) and Ph D (Computer Science &Engineering) degrees from Dibrugarh University, IITKharagpur and IIT Guwahati respectively. In 1992,he served as a faculty member in the Departmentof Computer Science & Engineering, North EasternRegional Institute of Science & Technology, Itana-gar, India. In 1999 he has joined Tezpur University,India, where he is currently working as Professor inthe Department of Computer Science & Engineering.

Before joining Tezpur University, he worked as Assistant Professor in theDepartment of Computer Sc. & Engineering at Jorhat Engineering College,Jorhat, India. His current research interests include - Quality of Service andCross-layer design in Wireless Ad Hoc Networks, Cognitive Radio Networksdesign, Optical networks and Network Security. He has published more than70 journal/conference papers and edited 2 books. He is a professional memberof ACM, IEEE and a Fellow of IETE. He was awarded with Best paper awardin ADCOM 2007, IETE K S Krishnan Memorial Award in 2009 and AmiyaK Pujari Best Paper Award in ICIT 2014

Eiji Oki is a Professor at the University of Electro-Communications, Tokyo, Japan. He received theB.E. and M.E. degrees in instrumentation engineer-ing and a Ph.D. degree in electrical engineering fromKeio University, Yokohama, Japan, in 1991, 1993,and 1999, respectively. In 1993, he joined NipponTelegraph and Telephone Corporation (NTT) Com-munication Switching Laboratories, Tokyo, Japan.He has been researching network design and control,traffic-control methods, and high-speed switchingsystems. From 2000 to 2001, he was a Visiting

Scholar at the Polytechnic Institute of New York University, Brooklyn,New York, where he was involved in designing terabit switch/router sys-tems. He was engaged in researching and developing high-speed opticalIP backbone networks with NTT Laboratories. He joined the University ofElectro-Communications, Tokyo, Japan, in July 2008. He has been activein the standardization of the path computation element (PCE) and GMPLSin the IETF. He wrote more than ten IETF RFCs. Prof. Oki was therecipient of several prestigious awards, including the 1998 Switching SystemResearch Award and the 1999 Excellent Paper Award presented by IEICE,the 2001 Asia-Pacific Outstanding Young Researcher Award presented byIEEE Communications Society for his contributions to broadband network,ATM, and optical IP technologies, the 2010 Telecom System TechnologyPrize by the Telecommunications Advanced Foundation, IEEE HPSR 2012Outstanding Paper Award, and IEEE HPSR 2014 Best Paper Award Finalist,First Runner Up. He has authored/co-authored four books, Broadband PacketSwitching Technologies, published by John Wiley, New York, in 2001, GMPLSTechnologies, published by CRC Press, Boca Raton, FL, in 2005, AdvancedInternet Protocols, Services, and Applications, published by Wiley, NewYork, in 2012, and Linear Programming and Algorithms for CommunicationNetworks, CRC Press, Boca Raton, FL, in 2012. He is a Fellow of IEEE andIEICE.