2738 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 18, NO.

4, FOURTH QUARTER 2016

Software Defined Optical Networks (SDONs):

A Comprehensive Survey
Akhilesh S. Thyagaturu, Anu Mercian, Member, IEEE, Michael P. McGarry, Senior Member, IEEE,
Martin Reisslein, Fellow, IEEE, and Wolfgang Kellerer, Senior Member, IEEE

Abstract—The emerging software defined networking (SDN) logical centralization of control, and (iii) programmability of
paradigm separates the data plane from the control plane and network functions. The first two architectural principles are
centralizes network control in an SDN controller. Applications related in that they combine to allow for network control func-
interact with controllers to implement network services, such
as network transport with quality of service. SDN facilitates tions to have a wider perspective on the network. The idea
the virtualization of network functions so that multiple virtual is that networks can be made easier to manage (i.e., control
networks can operate over a given installed physical network and monitor) with a move away from significantly distributed
infrastructure. Due to the specific characteristics of optical control. A tradeoff is then considered that balances ease of
(photonic) communication components and the high optical management arising from control centralization and scalability
transmission capacities, SDN-based optical networking poses
particular challenges, but holds also great potential. In this issues that naturally arise from that centralization.
article, we comprehensively survey studies that examine the The SDN abstraction layering consists of three generally
SDN paradigm in optical networks; in brief, we survey the accepted layers [5] inspired by computing systems, from the
area of software defined optical networks (SDONs). We mainly bottom layer to the top layer: (i) the infrastructure layer,
organize the SDON studies into studies focused on the infras- (ii) the control layer, and (iii) the application layer, as
tructure layer, the control layer, and the application layer.
Moreover, we cover SDON studies focused on network virtual- illustrated in Fig. 1. The interface between the application
ization, as well as SDON studies focused on the orchestration of layer and the control layer is referred to as the NorthBound
multilayer and multidomain networking. Based on the survey, Interface (NBI), while the interface between the control layer
we identify open challenges for SDONs and outline future and the infrastructure layer is referred to as the SouthBound
directions. Interface (SBI). There are a variety of standards emerg-
Index Terms—Control layer, infrastructure layer, optical net- ing for these interfaces, e.g., the OpenFlow protocol [7]
work, orchestration, software defined networking (SDN), virtual for the SBI.
network. The application layer is modeled after software applications
that utilize computing resources to complete tasks. The con-
I. I NTRODUCTION trol layer is modeled after a computer’s Operating System (OS)
that manages computer resources (e.g., processors and mem-
T LEAST a decade ago [1] it was recognized that new
A network abstraction layers for network control functions
needed to be developed to both simplify and automate network
ory), provides an abstraction layer to simplify interfacing with
the computer’s devices, and provides a common set of ser-
vices that all applications can leverage. Device drivers in
management. Software Defined Networking (SDN) [2]–[4] is
a computer’s OS hide the details of interfacing with many
the design principle that emerged to structure the develop-
different devices from the applications by offering a sim-
ment of those new abstraction layers. Fundamentally, SDN is
ple and unified interface for various device types. In the
defined by three architectural principles [5], [6]: (i) the separa-
SDN model both the unified SBI as well as the control layer
tion of control plane functions and data plane functions, (ii) the
functionality provide the equivalent of a device driver for inter-
Manuscript received November 4, 2015; revised May 26, 2016; accepted facing with devices in the infrastructure layer, e.g., packet
June 29, 2016. Date of publication July 1, 2016; date of current version switches.
November 18, 2016. This work was supported in part by the Alexander von Optical networks play an important role in our modern
Humboldt Foundation through a Friedrich Wilhelm Bessel Research Award.
(Corresponding author: Martin Reisslein.) information technology due to their high transmission
A. S. Thyagaturu, A. Mercian, and M. Reisslein are with the capacities. At the same time, the specific optical (pho-
School of Electrical, Computer, and Energy Engineering, Arizona State tonic) transmission and switching characteristics, such as
University, Tempe, AZ 85287-5706 USA. Parts of this article were
written while M. Reisslein visited the Technische Universität München, circuit, burst, and packet switching on wavelength channels,
Munich 80290, Germany (e-mail: [email protected]; [email protected]; pose challenges for controlling optical networks. This article
[email protected]). presents a comprehensive survey of Software Defined Optical
M. P. McGarry is with the Department of Electrical and Computer
Engineering, University of Texas at El Paso, El Paso, TX 79968 USA (e-mail: Networks (SDONs). SDONs seek to leverage the flexibility of
[email protected]). SDN control for supporting networking applications with an
W. Kellerer is with the Lehrstuhl für Kommunikationsnetze, underlying optical network infrastructure. This survey com-
Technische Universität München, Munich 80290, Germany (e-mail:
[email protected]). prehensively covers SDN related mechanisms that have been
Digital Object Identifier 10.1109/COMST.2016.2586999 studied to date for optical networks.
1553-877X  c 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
THYAGATURU et al.: SDONs: A COMPREHENSIVE SURVEY 2739

Fig. 1. Illustration of Software Defined Networking (SDN) abstraction layers: The infrastructure layer implements the data plane, e.g., with OpenFlow (OF)
switches [7] or network elements (devices) controlled with the NETCONF protocol [8]. A controller at the control layer, e.g., the ONOS controller [9], controls
the infrastructure layer based on the application layer requirements. The interface between the application and control layers is commonly referred to as the
NorthBound Interface (NBI), while the interface between the control and infrastructure layers is commonly referred to as the SouthBound Interface (SBI).
The WestBound Interface (WBI) interconnects multiple SDN domains, while the EastBound Interface (EBI) interconnects with non-SDN domains.

A. Related Work the overview of SDON challenges by considering the incor-

The general principles of SDN have been extensively cov- poration of 5G wireless systems. Cvijetic [36] has noted that
ered in several surveys, see for instance, [2], [6], [7], [10]–[27]. SDN access networks are highly promising for low-latency and
SDN security has been surveyed in [28] and [29], while man- high-bandwidth back-hauling from 5G cell base stations and
agement of SDN networks has been surveyed in [26] and briefly surveyed the requirements and areas of future research
SDN-based satellite networking is considered in [30]. required for integrating 5G with SDON access networks. A
To date, there have been relatively few overview and related overview of general software defined access networks
survey articles on SDONs. Zhang et al. [31] have pre- based on a variety of physical transmission media, including
sented a thorough survey on flexible optical networking based copper Digital Subscriber Line (DSL) [37] and Passive Optical
on Orthogonal Frequency Division Multiplexing (OFDM) in Networks (PONs), has been presented by Kerpez et al. [38].
core (backbone) networks. The survey briefly notes how Bitar [39] has surveyed use cases for SDN controlled broad-
OFDM-based elastic networking can facilitate network virtu- band access, such as on-demand bandwidth boost, dynamic
alization and surveys a few studies on OFDM-based network service re-provisioning, as well as value-added services and
virtualization in core networks. service protection. Bitar [39] has discussed the commer-
Bhaumik et al. [32] have presented an overview of SDN cial perspective of the access networks that are enhanced
and network virtualization concepts and outlined principles with SDN to add cost-value to the network operation.
for extending SDN and network virtualization concepts to the de Almeida Amazonas et al. [40] have surveyed the key issues
field of optical networking. Their focus has been mainly on of incorporating SDN in optical and wireless access networks.
industry efforts, reviewing white papers on SDN strategies They briefly outlined the obstacles posed by the different spe-
from leading networking companies, such as Cisco, Juniper, cific physical characteristics of optical and wireless access
Hewlett-Packard, Alcatel-Lucent, and Huawei. A few selected networks.
academic research projects on general SDN optical networks, Although our focus is on optical networks, for complete-
namely projects reported in the journal articles [33], [34] and ness we note that for the field of wireless and mobile
a few related conference papers, have also been reviewed by networks, SDN based networking mechanisms have been sur-
Bhaumik et al. [32]. In contrast to Bhaumik et al. [32], we pro- veyed in [41]–[47] while network virtualization has been
vide a comprehensive up-to-date review of academic research surveyed in [48] for general wireless networks and in [49]
on SDONs. Whereas Bhaumik et al. [32] presented a small for wireless sensor networks. SDN and virtualization strate-
sampling of SDON research organized by research projects, gies for LTE wireless cellular networks have been surveyed
we present a comprehensive SDON survey that is organized in [50]. SDN-based 5G wireless network developments for
according to the SDN infrastructure, control, and application mobile networks have been outlined in [51]–[54].
layer architecture.
For the SDON sub-domain of access networks, Cvijetic [35] B. Survey Organization
has given an overview of access network challenges that can We have mainly organized our survey according to the
be addressed with SDN. These challenges include lack of three-layer SDN architecture illustrated in Fig. 1. In partic-
support for on-demand modifications of traffic transmission ular, we have organized the survey in a bottom-up manner,
policies and rules and limitations to vendor-proprietary poli- surveying first SDON studies focused on the infrastructure
cies, rules, and software. Cvijetic [35] also offers a very brief layer in Section III. Subsequently, we survey SDON studies
overview of research progress for SDN-based optical access focused on the control layer in Section IV. The virtualization
networks, mainly focusing on studies on the physical (pho- of optical networks is commonly closely related to the SDN
tonics) infrastructure layer. Cvijetic [36] has further expanded control layer. Therefore, we survey SDON studies focused
2740 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 18, NO. 4, FOURTH QUARTER 2016

on virtualization in Section V, right after the SDON con- EBIs establish communication links to legacy network archi-
trol layer section. Resuming the journey up the layers in tectures (i.e., non-SDN networks); whereas, links to con-
Fig. 1, we survey SDON studies focused on the application current (side-by-side) SDN architectures are facilitated by
layer in Section VI. We survey mechanisms for the overarch- the WBIs.
ing orchestration of the application layer and lower layers, 1) Infrastructure Layer: The infrastructure layer includes
possibly across multiple network domains (see Fig. 2), in an environment for (payload) data traffic forwarding (data
Section VII. Finally, we outline open challenges and future plane) either in virtual or actual hardware. The data plane
research directions in Section VIII and conclude the survey in comprises a network of network elements, which expose their
Section IX. capabilities through the SBI to the control plane. In tradi-
tional networking, control mechanisms are embedded within
II. BACKGROUND an infrastructure, i.e., decision making capabilities are embed-
ded within the infrastructure to perform network actions, such
This section first provides background on Software Defined
as switching or routing. Additionally, these forwarding actions
Networking (SDN), followed by background on virtualiza-
in the traditional network elements are autonomously estab-
tion and optical networking. SDN, as defined by the Internet
lished based on self-evaluated topology information that is
Engineering Task Force (IETF) [55], is a networking paradigm
often obtained through proprietary vendor-specific algorithms.
enabling the programmability of networks. SDN abstracts and
Therefore, the configuration setups of traditional network
separates the data forwarding plane from the control plane,
elements are generally not reconfigurable without a service
allowing faster technological development both in data and
disruption, limiting the network flexibility. In contrast, SDN
control planes. We provide background on the SDN archi-
decouples the autonomous control functions, such as for-
tecture, including its architectural layers in Section II-A. The
warding algorithms and neighbor discovery of the network
network programmability provides the flexibility to dynami-
nodes, and moves these control functions out of the infras-
cally initialize, control, manipulate, and manage the end-to-end
tructure to a centrally controlled logical node, the controller.
network behavior via open interfaces, which are reviewed
In doing so, the network elements act only as dumb switches
in Section II-B. Subsequently, we provide background on
which act upon the instructions of the controller. This decou-
network virtualization in Section II-C and on optical network-
pling reduces the network element complexity and improves
ing in Section II-D.
reconfigurability.
In addition to decoupling the control and data planes, packet
A. Software Defined Networking (SDN) Architectural Layers modification capabilities at the line-rates of network elements
SDN offers a simplified view of the underlying network have been significantly improved with SDN. P4 [59] is a pro-
infrastructure for the network control and monitoring appli- grammable protocol-independent packet processor, that can
cations through the abstraction of each independent network arbitrarily match the fields within any formatted packet and
layer. Fig. 1 illustrates the three-layer SDN architecture model is capable of applying any arbitrary actions (as programmed)
consisting of application, control, and infrastructure layers on the packet before forwarding. A similar forwarding mecha-
as defined by the Open Networking Foundation (ONF) [5]. nism, Protocol-oblivious Forwarding (PoF) has been proposed
The ONF is the organization that is responsible for the pub- by Huawei Technologies [60].
lication of specifications for the OpenFlow protocol. The 2) Control Layer: The control layer is responsible for pro-
OpenFlow protocol [2], [7], [56] has been the first proto- gramming (configuring) the network elements (switches) via
col for the SouthBound Interface (SBI, also referred to as the SBIs. The SDN controller is a logical entity that identifies
Data-Controller Plane Interface (D-CPI)) between the con- the south bound instructions to configure the network infras-
trol and infrastructure layers. Each layer operates indepen- tructure based on application layer requirements. To efficiently
dently, allowing multiple solutions to coexist within each manage the network, SDN controllers can request information
layer, e.g., the infrastructure layer can be built from any from the SDN infrastructures, such as flow statistics, topol-
programmable devices, which are commonly referred to as ogy information, neighbor relations, and link status from the
network elements [57] or network devices [55] (or sometimes network elements (nodes). The software entity that imple-
as forwarding elements [58]). We will use the terminology ments the SDN controller is often referred to as Network
network element throughout this survey. The SouthBound Operating System (NOS). Generally, a NOS can be imple-
Interface (SBI) and the NorthBound Interface (NBI, also mented independently of SDN, i.e., without supporting SDN.
referred to as Application-Controller Plane Interface (A-CPI)) On the other hand, in addition to supporting SDN operations,
are defined as the primary interfaces interconnecting the a NOS can provide advanced capabilities, such as virtualiza-
SDN layers through abstractions. An SDN network archi- tion, application scheduling, and database management. The
tecture can coexist with both concurrent SDN architectures Open Network Operating System (ONOS) [9] is an example
and non-SDN legacy network architectures. Additional inter- of an SDN based NOS with a distributed control architec-
faces are defined namely the EastBound Interface (EBI) ture designed to operate over Wide Area Networks (WANs).
and the WestBound Interface (WBI) [17] to interconnect Furthermore, Cisco has recently developed the one Platform
the SDN architecture with external network architectures Kit (onePK) [61], which consists of a set of Application
(the EBI and WBI are also collectively referred to as Program Interfaces (APIs) that allow the network applica-
Intermediate-Controller Plane Interfaces (I-CPIs)). Generally, tions to control Cisco network devices without a command
THYAGATURU et al.: SDONs: A COMPREHENSIVE SURVEY 2741

In contrast, the SDN orchestrator makes control decisions

that are generally not predefined. More specifically, the SDN
orchestrator could make an automated decision with the help
of SDN applications or seek a manual recommendation from
user inputs; therefore, results are generally not predefined.
These orchestrator decisions (actions/configurations) are then
delegated via the SDN controllers and the SBIs to the network
elements.
Intuitively speaking, SDN orchestration can be viewed as a
distinct abstracted (higher) layer for coordination and manage-
ment that is positioned above the SDN control and application
layers. Therefore, we generalize the term SDN orchestrator as
an entity that realizes a wider, more general (more encompass-
ing) network functionality as compared to the SDN controllers.
For instance, a cloud SDN orchestrator can instantiate and
tear down Virtual Machines (VMs) according to the cloud
Fig. 2. Overview of SDN orchestrator and SDN controllers: The SDN orches-
tration coordinates and manages at a higher abstracted layer, above the SDN workload, i.e., make decisions that span across multiple net-
applications and SDN controllers. SDN controllers, which may be in a hier- work domains and layers. In contrast, SDN controllers realize
archy (see left part), implement the orchestrator decisions. A virtualization more specific network functions, such as routing and path
hypervisor may intercept the SouthBound Interfaces (SBIs) to create multiple
virtual networks from a given physical network infrastructure. (The optical computation.
orchestrator on the right can be ignored for now and will be addressed in
Section VIII-F.) B. SDN Interfaces
1) Northbound Interfaces (NBIs): A logical interface that
line interface. The onePK libraries act as an SBI for Cisco interconnects the SDN controller and a software entity oper-
ONE controllers and are based on C and Java compilers. ating at the application layer is commonly referred to as
3) Application Layer: The application layer comprises a NorthBound Interface (NBI), or as Application-Controller
network applications and services that utilize the control Plane Interface (A-CPI).
plane to realize network functions over the physical or vir- a) REST: REpresentational State Transfer (REST) [64] is
tual infrastructure. Examples of network applications include generally defined as a software architectural style that supports
network topology discovery, provisioning, and fault restora- flexibility, interoperability, and scalability. In the context of
tion. The SDN controller presents an abstracted view of the SDN NBI, REST is commonly defined as an API that
the network to the SDN applications to facilitate the real- meets the REST architectural style [65], i.e., is a so-called
ization of application functionalities. The applications can RESTful API:
also include higher levels of network management, such as • Client-Sever: Two software entities should follow the
network data analytics, or specialized functions requiring pro- client-server model. In SDN, a controller can be a server
cessing in large data centers. For instance, the Central Office and the application can be the client. This allows multiple
Re-architected as a Data center (CORD) [62] is an SDN heterogeneous SDN applications to coexist and operate
application based on ONOS [9], that implements the typical over a common SDN controller.
central office network functions, such as optical line termina- • Stateless: The client is responsible for managing all the
tion, as well as BaseBand Unit (BBU) and Data Over Cable states and the server acts upon the client’s request. In
Interface (DOCSIS) [63] processing as virtualized software SDN, the applications collect and maintain the states of
entities, i.e., as SDN applications. the network, while the controller follows the instructions
4) Orchestration Layer: Although the orchestration layer is from the applications.
commonly not considered one of the main SDN architectural • Caching: The client has to support the temporary local
layers illustrated in Fig. 1, as SDN systems become more com- storage of information such that interactions between
plex, orchestration becomes increasingly important. We intro- the client and server are reduced so as to improve
duce therefore the orchestration layer as an important SDN performance and scalability.
architectural layer in this background section. Typically, an • Uniform/Interface Contract: An overarching technical
SDN orchestrator is the entity that coordinates software mod- interface must be followed across all services using the
ules within a single SDN controller, a hierarchical structure of REST API. For example, the same data format, such as
multiple SDN controllers, or a set of multiple SDN controllers Java Script Object Notation (JSON) or eXtended Markup
in a “flat” arrangement (i.e., without a hierarchy) as illustrated Language (XML), has to be followed for all interactions
in Fig. 2. An SDN controller in contrast can be viewed as a sharing the common interface.
logically centralized single control entity. This logically cen- • Layered System: In a multilayered architectural solu-
tralized single control entity appears as the directly controlling tion, the interface should only be concerned with the
entity to the network elements. The SDN controller is respon- next immediate node and not beyond. Thus, allowing
sible for signaling the control actions or rules that are typically more layers to be inserted, modified, or removed without
predefined (e.g., through OpenFlow) to the network elements. affecting the rest of the system.
2742 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 18, NO. 4, FOURTH QUARTER 2016

2) Southbound Interfaces (SBIs): A logical interface that single or multiple physical infrastructures (e.g., geographically
interconnects the SDN controller and the network element separated WAN segments). Network Virtualization (NV) can
operating on the infrastructure layer (data plane) is commonly flexibly create independent virtual networks (slices) for dis-
referred to as a SouthBound Interface (SBI), or as the Data- tinct users over a given physical infrastructure. Each network
Controller Plane Interface (D-CPI). Although a higher level slice can be created with prescribed resource allocations.
connection, such as a UDP or TCP connection, is sufficient When no longer required, a slice can be deleted, freeing up
for enabling the communication between two entities of the the reserved physical resources.
SDN architecture, e.g., the controller and the network ele- Network hypervisors [78], [79] are the network elements
ments, specific SBI protocols have been proposed. These SBI that abstract the physical network infrastructure (including net-
protocols are typically not interoperable and thus are lim- work elements, communication links, and control functions)
ited to work with SBI protocol-specific network elements into logically isolated virtual network slices. In particular,
(e.g., an OpenFlow switch does not work with the NETCONF in the case of an underlying physical SDN network, an
protocol). SDN hypervisor can create multiple isolated virtual SDN
a) OpenFlow protocol: The interaction between an networks [80], [81]. Through hypervisors, NV supports the
OpenFlow switching element (data plane) and an OpenFlow implementation of a wide range of network services belong-
controller (control plane) is carried out through the OpenFlow ing to the link and network protocol layers (L2 and L3),
protocol [7], [56]. This SBI (or D-CPI) is therefore also such as switching and routing. Additionally, virtualized infras-
sometimes referred to as the OpenFlow control channel. tructures can also support higher layer services, such as
SDN mainly operates through packet flows that are identified load-balancing of servers and firewalls. The implementation
through matches on prescribed packet fields that are specified of such higher layer services in a virtualized environment
in the OpenFlow protocol specification. For matched pack- is commonly referred to as Network Function Virtualization
ets, SDN switches then take prescribed actions, e.g., process (NFV) [82]–[86]. NFV can be viewed as a special case of
the flow’s packets in a particular way, such as dropping the NV in which network functions, such as address translation
packet, duplicating it on a different port or modifying the and intrusion detection functions, are implemented in a vir-
header information. tualized environment. That is, the virtualized functions are
b) Path computation element protocol (PCEP): The implemented in the form of software entities (modules) run-
PCEP enables communication between the Path Computation ning on a data center (DC) or the cloud [75]. In contrast,
Client (PCC) of the network elements and the Path the term NV emphasizes the virtualization of the network
Computation Element (PCE) residing within the controller. resources, such as communication links and network nodes.
The PCE centrally computes the paths based on constraints
received from the network elements. Computed paths are then
forwarded to the individual network elements through the D. Optical Networking Background
PCEP protocol [66], [67].
c) Network configuration (NETCONF) protocol: The 1) Optical Switching Paradigms: Optical networks are
NETCONF protocol [8] provides mechanisms to config- networks that either maintain signals in the optical domain
ure, modify, and delete configurations on a network device. or at least utilize transmission channels that carry signals in
Configuration of the data and protocol messages are encoded the optical domain. In optical networks that maintain signals in
in the NETCONF protocol using an eXtensible Markup the optical domain, switching can be performed at the circuit,
Language (XML). Remote procedure calls are used to realize packet, or burst granularities.
the NETCONF protocol operations. Therefore, only devices a) Circuit switching: Optical circuit switching can be
that are enabled with required remote procedure calls allow the performed in space, waveband, wavelength, or time. The opti-
NETCONF protocol to remotely modify device configurations. cal spectrum is divided into wavelengths either on a fixed
d) Border gateway protocol link state distribution wavelength grid or on a flexible wavelength grid. Spectrally
(BGP-LS) protocol: The central controller needs a topol- adjacent wavelengths can be coalesced into wavebands. The
ogy information database, also known as Traffic Engineering fixed wavelength grid standard (ITU-T G.694.1) specifies
Database (TED), for optimized end-to-end path computation. specific center frequencies that are either 12.5 GHz, 25 GHz,
The controller has to request the information for building the 50 GHz, or 100 GHz apart. The flexible DWDM grid (flexi-
TED, such as topology and bandwidth utilization, via the SBIs grid) standard (ITU-T G.694.1) [31], [87]–[89] allows the
from the network elements. This information can be gathered center frequency to be any multiple of 6.25 GHz away from
by a BGP extension, which is referred to as BGP-LS. 193.1 THz and the spectral width to be any multiple of
12.5 GHz. Elastic Optical Networks (EONs) [90]–[92] that
take advantage of the flexible grid can make more effi-
C. Network Virtualization cient use of the optical spectrum but can cause spectral
Analogously to the virtualization of computing fragmentation, as lightpaths are set up and torn down, the
resources [68], [69], network virtualization abstracts the spectral fragmentation counteracts the more efficient spectrum
underlying physical network infrastructure so that one or utilization [93].
multiple virtual networks can operate on a given physical b) Packet switching: Optical packet switching performs
network [15], [70]–[77]. Virtual networks can span over a packet-by-packet switching using header fields in the optical
THYAGATURU et al.: SDONs: A COMPREHENSIVE SURVEY 2743

domain as much as possible. An all-optical packet switch report their bandwidth demands to the OLT and the OLT
requires [94]: then assigns upstream transmission windows according to a
• Optical synchronization, demultiplexing, and Dynamic Bandwidth Allocation (DBA) algorithm [99]–[102].
multiplexing Conventional PONs cover distances up to 20 km, while so-
• Optical packet forwarding table computation called Long-Reach (LR) PONs cover distances up to around
• Optical packet forwarding table lookup 100 km [103]–[105].
• Optical switch fabric Recently, hybrid access networks that combine multi-
• Optical buffering ple transmission media, such as Fiber-Wireless (FiWi) net-
Optical packet switches typically relegate some of these design works [106]–[110] and PON-DSL networks [111], have been
elements to the electrical domain. Most commonly the packet explored to take advantage of the respective strengths of the
forwarding table computation and lookup is performed electri- different transmission media.
cally. When there is contention for a destination port, a packet b) Networks connected to access networks: Optical
needs to be buffered optically, this buffering can be accom- access networks provide Internet connectivity for a wide range
plished with rather impractical fiber delay lines. Fiber delay of peripheral networks. Residential (home) wired or wireless
lines are fiber optic cables whose lengths are configured to pro- local area networks [112] typically interconnect individual
vide a certain time delay of the optical signal; e.g., 100 meters end devices (hosts) in a home or small business and may
of fiber provides 500 ns of delay. An alternative to buffering is connect directly with an optical access network. Cellular
to either drop the packet or to use deflection routing, whereby wireless networks provide Internet access to a wide range
a packet is routed to a different output that may or may not of mobile devices [113]–[115]. Specialized cellular backhaul
lead to the desired destination. networks [116]–[122] relay the traffic to/from base stations
c) Burst switching: Optical burst switching alleviates the of wireless cellular networks to either wireless access net-
requirements of optical packet forwarding table computation, works [123]–[128] or optical access networks. Moreover,
forwarding table lookup, as well as buffering while accom- optical access networks are often employed to connect Data
modating bursty traffic that would lead to poor utilization of Center (DC) networks to the Internet. DC networks intercon-
optical circuits. In essence, it permits the rapid establishment nect highly specialized server units that process and store
of short-lived optical circuits to support the transfer of one large data amounts with specialized networking technolo-
or more packets coalesced into a burst. A control packet is gies [129]–[133]. Data centers are typically employed to
sent through the network that establishes the lightpath for the provide the so-called “cloud” services for commercial and
burst and then the burst is transmitted on the short-lived cir- social media applications.
cuit with no packet lookup or buffering required along the c) Metropolitan area networks: Optical Metropolitan
path [94]. Since the circuit is only established for the length (metro) Area Networks (MANs) interconnect the optical
of the burst, network resources are not wasted during idle access networks in a metropolitan area with each other and
periods. To avoid any buffering of the burst in the optical with wide-area (backbone, core) networks. MANs have typ-
network, the burst transmission can begin once the lightpath ically a ring or star topology [134]–[139] and commonly
establishment has been confirmed (tell-and-wait) or a short employ optical networking technologies.
time period after the control packet is sent (just-enough-time). d) Backbone networks: Optical backbone (wide area)
Note: Sending the burst immediately after the control packet networks interconnect the individual MANs on a national or
(tell-and-go) would require some buffering of the optical burst international scale. Backbone networks have typically a mesh
at the switching nodes. structure and employ very high speed optical transmission
2) Optical Network Structure: Optical networks are typi- links.
cally structured into three main tiers, namely access networks,
metropolitan (metro) area networks, and backbone (core)
networks [95]. III. SDN C ONTROLLED P HOTONIC C OMMUNICATION
a) Access networks: In the area of optical access I NFRASTRUCTURE L AYER
networks [96], so-called Passive Optical Networks (PONs), This section surveys mechanisms for controlling physi-
in particular, Ethernet PONs (EPONs) and Gigabit PONs cal layer aspects of the optical (photonic) communication
(GPONs) [97], [98], have been widely studied. A PON has infrastructure through SDN. Enabling the SDN control down
typically an inverse tree structure with a central Optical to the photonic level operation of optical communications
Line Terminal (OLT) connecting multiple distributed Optical allows for flexible adaptation of the photonic components sup-
Network Units (ONUs; also referred to as Optical Network porting optical networking functionalities [33], [205]–[207].
Terminals, ONTs) to metro networks. In the downstream As illustrated in Fig. 3, this section first surveys transmit-
(OLT to ONUs) direction, the OLT broadcasts transmissions. ters and receivers (collectively referred to as transceivers or
However, in the upstream (ONUs to OLT) direction, the transponders) that permit SDN control of the optical sig-
transmissions of the distributed ONUs need to be coordi- nal transmission characteristics, such as modulation format.
nated to avoid collisions on the shared upstream wavelength We also survey SDN controlled space division multiplex-
channel. Typically, a cyclic polling based Medium Access ing (SDM), which provides an emerging avenue for highly
Control (MAC) protocol, e.g., based on the MultiPoint Control efficient optical transmissions. Then, we survey SDN con-
Protocol (MPCP, IEEE 802.3ah), is employed. The ONUs trolled optical switching, covering first switching elements and
2744 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 18, NO. 4, FOURTH QUARTER 2016

Fig. 3. Classification of physical infrastructure layer SDON studies.

then overall switching paradigms, such as converged packet Through adjusting the direct current bias voltages and ampli-
and circuit switching. Finally, we survey cognitive photonic tudes of drive signals the combination of MZMs can vary
communication infrastructures that monitor the optical signal the amplitude and phase of the generated optical signal [214].
quality. The optical signal quality information can be used to Thus, modulation formats ranging from Binary Phase Shift
dynamically control the transceivers as well as the filters in Keying (BPSK) to Quadrature Phase Shift Keying (QPSK)
switching elements. as well as 8 and 16 quadrature amplitude modulation [209]
can be generated. The amplitudes and bias voltages of the
drive signals can be signaled through an SDN OpenFlow con-
A. Transceivers trol plane to achieve the different modulation formats. The
Software defined optical transceivers are optical transmitters corresponding flexible receiver consists of a polarization fil-
and receivers that can be flexibly configured by SDN to trans- ter that feeds four parallel photodetectors, each followed by
mit or receive a wide range of optical signals [208]. Generally, an Analog-to-Digital Converter (ADC). The outputs of the
software defined optical transceivers vary the modulation four parallel ADCs are then processed with DSP techniques
format [209] of the transmitted optical signal by adjusting to automatically (without SDN control) detect the modula-
the transmitter and receiver operation through Digital Signal tion format. Experiments in [143] and [144] have evaluated
Processing (DSP) techniques [210]–[212]. These transceivers the bit error rates and transmission capacities of the dif-
have evolved in recent years from Bandwidth Variable ferent modulation formats and have demonstrated the SDN
Transceivers (BVTs) generating a single signal flow to slice- control.
able multi-flow BVTs. Single-flow BVTs permit SDN control b) Single-flow BVTs for PONs: Flexible optical net-
to adjust the transmission bandwidth of the single gener- working with real-time bandwidth adjustments is also highly
ated signal flow. In contrast, sliceable multi-flow BVTs allow desirable for PON access and metro networks, albeit the BVT
for the independent SDN control of multiple communication technologies for access and metro networks should have low
traffic flows generated by a single BVT. cost and complexity [145]. Iiyama et al. [146] have devel-
1) Single-Flow Bandwidth Variable Transceivers (BVTs): oped a DSP based approach that employs SDN to coordinate
Software defined optical transceivers have initially been exam- the downstream PON transmission of On-Off Keying (OOK)
ined in the context of adjusting a single optical signal modulation [147] and Quadrature Amplitude Modulation
flow for flexible WDM networking [140]–[142]. The goal (QAM) [148] signals. The OOK-QAM-SDN scheme involves
has been to make the photonic transmission characteris- a novel multiplexing method, wherein all the data are simul-
tics of a given transmitter fully programmable. We pro- taneously sent from the OLT to the ONUs and the ONUs
ceed to review a representative single-flow BVT design for filter the data they need. The experimental setup in [146] also
general optical mesh networks in detail and then summa- demonstrated digital software ONUs that concurrently trans-
rize related single-flow BVTs for PONs and data center mit data by exploiting the coexistence of OOK and QAM. The
networks. OOK-QAM-SDN evaluations demonstrated the control of the
a) Mach-Zehnder modulator based flexible transmitter: receiving sensitivity which is very useful for a wide range of
Choi et al. [143] and Liu et al. [144] have demonstrated transmission environments.
a flexible transmitter based on Mach-Zehnder Modulators In a related study, Vacondio et al. [149] have examined
(MZMs) [213] and a corresponding flexible receiver for SDN Software-Defined Coherent Transponders (SDCT) for TDMA
control in a general mesh network. The flexible transceiver PON access networks. The proposed SDCT digitally processes
employs a single dual-drive MZM that is fed by two binary the burst transmissions to achieve improved burst mode trans-
electric signals as well as a parallel arrangement of two missions according to the distance of a user from the OLT.
MZMs which are fed by two additional electrical signals. The performance results indicate that the proposed flexible
THYAGATURU et al.: SDONs: A COMPREHENSIVE SURVEY 2745

Fig. 4. Illustration of DSP reconfigurable ONU and OLT designs [151]: Each ONU passes the electrical Optical OFDM signal [150] through a Shaping
Filter (SF) that is SDN-configured by the DSP controller, followed by a Digital-to-Analog Converter (DAC) and Intensity Modulator (IM) to generate the
optical signal. The centralized SDN controller configures the corresponding OLT Matching Filter (MF) and ensures that all ONU filters are orthogonal.

approach more than doubles the average transmission capacity PSK and OOK signal and thus providing a higher transmission
per user compared to a static approach. bit rate.
Bolea et al. [150], [151] have recently developed low- 2) Sliceable Multi-Flow Bandwidth Variable Transceivers:
complexity DSP reconfigurable ONU and OLT designs for Whereas the single-flow transceivers surveyed in
SDN-controlled PON communication. The proposed commu- Section III-A1 generate a single optical signal flow,
nication is based on carrierless amplitude and phase mod- parallelization efforts have resulted in multi-flow transceivers
ulation [215] enhanced with optical Orthogonal frequency (transponders) [153]. Multi-flow transceivers can generate
Division Multiplexing (OFDM) [150]. The different OFDM multiple parallel optical signal flows and thus form the
channels are manipulated through DSP filtering. As illustrated infrastructure basis for network virtualization.
in Fig. 4, the ONU consists of a DSP controller that controls a) Encoder based programmable transponder:
the filter coefficients of the shaping filter. The filter output Sambo et al. [154], [155] have developed an SDN-
is then passed through a Digital-to-Analog Converter (DAC) programmable bandwidth-variable multi-flow transmitter
and intensity modulator for electric-optical conversion. At the and corresponding SDN-programmable multi-flow band-
OLT, a photo diode converts the optical signal to an elec- width variable receiver, referred to jointly as programmable
trical signal, which then passes through an Analog-to-Digital bandwidth-variable transponder. The transmitter mainly
Converter (ADC). The SDN controlled OLT DSP controller consists of a programmable encoder and multiple parallel
sets the filter coefficients in the matching filter to correspond Polarization-Multiplexing Quadrature Phase Shift Keying
to the filtering in the sending ONU. The OLT DSP controller (PM-QPSK [209]) laser transmitters, whose signals are
is also responsible for ensuring the orthogonality of all the multiplexed by a coupler. The encoder is SDN-controlled to
ONU filters in the PON. The performance evaluations in [151] implement Low-Density Parity-Check (LDPC) coding [216]
indicate that the proposed DSP reconfigurable ONU and OLT with different code rates. At the receiver, the SDN control
system achieves ONU signal bitrates around 3.7 Gb/s for eight sets the local oscillators and LDPC decoder. The developed
ONUs transmitting upstream over a 25 km PON. The perfor- transponder allows the setting of the number of subcarriers,
mance evaluations also illustrate that long DSP filter lengths, the subcarrier bitrate, and the LDPC coding rate through
which increase the filter complexity, improve performance. SDN. Related frequency conversion and defragmentation
c) Single-flow BVTs for data center networks: issues have been examined in [217]. In [156], a low-cost
Malacarne et al. [152] have developed a low-complexity version of the SDN programmable transponder with a multi-
and low-cost bandwidth adaptable transmitter for data cen- wavelength source has been developed. The multiwavelength
ter networking. The transmitter can multiplex Amplitude Shift source is based on a micro-ring resonator [218] that generates
Keying (ASK), specifically On-Off Keying (OOK), and Phase multiple signal carriers with only a single laser. Automated
Shift Keying (PSK) on the same optical carrier signal without configuration procedures for the comprehensive set of trans-
any special synchronization or temporal alignment mechanism. mission parameters, including modulation format, coding
In particular, the transmitter design [152] uses the OOK elec- configuration, and carriers have been explored in [157].
tronic signal to drive a Mach-Zehnder Modulator (MZM) that b) DSP based sliceable BVT: Moreolo et al. [158] have
is fed by the optical pulse modulated signal. SDN control can developed an SDN controlled sliceable BVT based on adap-
activate (or de-activate) the OOK signal stream, i.e., adapt tive Digital Signal Processing (DSP) of multiple parallel signal
from transmitting only the PSK signal to transmitting both the subcarriers. Each subcarrier is fed by a DSP module that
2746 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 18, NO. 4, FOURTH QUARTER 2016

Node (ARN) that directly connects via a distribution fiber seg-

ment, a passive remote node, and a trunk fiber segment to the
core (backbone) network, bypassing the conventional metro
network. The ARN is based on an SDN controlled S-BVT
to optimize the modulation format. With the modulation for-
mat optimization, the ARN can optimize the transmission
capacity for the given distance (via the distribution and trunk
fiber segments) to the core network. The evaluations in [161]
demonstrate good bit error rate performance of representative
HYDRA scenarios with a 200 km trunk fiber segment and dis-
tribution fiber lengths up to 100 km. In particular, distribution
fiber lengths up to around 70 km can be supported without
Forward Error Correction (FEC), whereas distribution fiber
lengths above 70 km would require standard FEC. The con-
solidation of the access and metro network infrastructure [219]
achieved through the optimized S-BVT transmissions can
significantly reduce the network cost and power consumption.
Fig. 5. Illustration of Subcarrier and Modulator Pool Based Virtualizable
Bandwidth Variable Transceiver (V-BVT) [159]: Through SDN control, the B. Space Division Multiplexing (SDM)-SDN
V-BVT Manager composes virtual transceivers by combining subcarriers from
the optical subcarriers pool with modulators from the optical modulators pool. Amaya et al. [162], [163] have demonstrated SDN control of
Space Division Multiplexing (SDM) [220] in optical networks.
More specifically, Amaya et al. [162], [163] employ SDN to
control the physical layer so as to achieve a bandwidth-flexible
configures the modulation format, including the bit rate set-
and programmable SDM optical network. The SDN control
ting, and the power level of the carrier by adapting a gain
can perform network slicing, resulting in sliceable superchan-
coefficient. The output of the DSP module is then passed
nels. A superchannel consists of multiple spatial carriers to
through digital to analog conversion that drives laser sources.
support dynamic bandwidth and QoS provisioning.
The parallel flows can be combined with a wavelength selec-
Galve et al. [164] have built on the flexible SDN con-
tive switch; the combined flow can be sliced into multiple
trolled SDM communication principles to develop a recon-
distinct sub-flows for distinct destinations. The functionality
figurable Radio Access Network (RAN). The RAN connects
of the developed DSP based BVT has been verified for a
the BaseBand processing Units (BBUs) in a shared central
metropolitan area network with links reaching up to 150 km.
office with the corresponding distributed Remote Radio Heads
c) Subcarrier and modulator pool based virtualizable
(RRHs) located at Base Stations (BSs). A multicore fiber oper-
BVT: Ou et al. [159], [160] have developed a Virtualizable
ated with SDM [220] connects the RRHs to the BBUs in
BVT (V-BVT) based on a combination of an optical subcar-
the central office. Galve et al. [164] introduce a radio over
riers pool with an independent optical modulators pool, as
fiber operation mode where SDN controlled switching maps
illustrated in Fig. 5. The emphasis of the design is on imple-
the subcarriers dynamically to spatial output ports. A comple-
menting Virtual Optical Networks (VONs) at the transceiver
mentary digitized radio over fiber operating mode maintains
level. The optical subcarriers pool contains multiple opti-
a BBU pool. Virtual BBUs are dynamically allocated to the
cal carriers, whereby channel spacing and central frequency
cores of the SDM operated multicore fiber.
(wavelength channel) can be selected. The optical modulators
pool contains optical modulators that can generate a wide vari-
ety of modulation formats. The SDN control interacts with a C. SDN-Controlled Switching
V-BVT Manager that implements a virtualization algorithm. 1) Switching Elements:
The virtualization algorithm generates a transceiver slice by a) ROADM: The Reconfigurable Optical Add-Drop
combining a particular set of subcarriers (with specific num- Multiplexer (ROADM) is an important photonic switching
ber of subcarriers, channel spacing, and central frequencies) device for optical networks. Through wavelength selective
from the optical subcarriers pool with a particular modulation optical switches, a ROADM can drop (or add) one or mul-
(with specific number of modulators and modulation formats) tiple wavelength channels carrying optical data signals from
from the optical modulators pool. The evaluations in [159] (to) a fiber without requiring the conversion of the optical
have evaluated the proposed V-BVT in a network testbed with signal to electric signals [221]. The ROADM thus provides
path lengths up to 200 km with 20 GHz channel spacing and an elementary switching functionality in the optical wave-
a variety of modulation formats, including BPSK as well as length domain. Initial ROADM based node architectures for
16QAM and 32QAM. cost-effectively supporting flexible SDN networks have been
d) S-BVT based hybrid long-reach fiber access network presented in [165]. Conventional ROADM networks have typ-
(HYDRA): HYDRA [161] is a novel hybrid long-reach fiber ically statically configured wavelength channels that transport
access network architecture based on sliceable BVTs. HYDRA traffic along a pre-configured route. Changes of wavelength
supports low-cost end-user ONUs through an Active Remote channels or routes in the statically configured networks incur
THYAGATURU et al.: SDONs: A COMPREHENSIVE SURVEY 2747

presently high operational costs due to required physical inter- d) Optical white box: Nejabati et al. [173] have proposed
ventions and are therefore typically avoided. New ROADM an optical white box switch design as a building block for
node designs allow changes of wavelength channels and routes a completely softwarized optical network. The optical white
through a management control plane. Due to these two flexi- box design combines a programmable backplane with pro-
bility dimensions (wavelength and route), these new ROADM grammable switching node elements. More specifically, the
nodes are referred to as “colorless” and “directionless”. backplane consists of two slivers, namely an optical back-
First designs for such colorless and directionless ROADM plane sliver and an electronic backplane sliver. These slivers
nodes have been outlined in [165] and further elaborated are set up to allow for flexible arbitrary connections between
in [166] and [167]. In addition to the colorless and direc- the switch node elements. The switch node elements include
tionless properties, the contentionless property has emerged programmable interfaces that build on SDN-controlled BVTs
for ROADMs [142]. Contentionless ROADM operation means (see Section III-A), protocol agnostic switching, and DSP ele-
that any port can be routed on any wavelength (color) in ments. The protocol agnostic switching element is envisioned
any direction without causing resource contention. Designs for to support both wavelength channel and time slot switching
such Colorless-Directionless-Contentionless (CDC) ROADMs in the optical backplane as well as programmable switching
have been proposed in [168] and [169]. In general, the with a high-speed packet processor in the electronic backplane.
ROADM designs consist of an express bank that intercon- The DSP elements support both the network processing and
nects the input and output ports coming from/leading to other the signal processing for executing a wide range of network
ROADMs, and an add-drop bank that connects the express functions. A prototype of the optical white box has been built
bank with the local receivers for dropped wavelength chan- with only a optical backplane sliver consisting of a 192 × 192
nels or transmitters for added wavelength channels. The recent optical space switch. Experiments have indicated that the cre-
designs have focused on the add-drop bank and explored ation of a virtual switching node with the OpenDayLight SDN
different arrangements of wavelength selective switches and controller takes roughly 400 ms.
multicast switches to provide add-drop bank functionality with e) GPON virtual switch: Lee et al. [174] have developed
the CDC property [168], [169]. a GPON virtual switch design that makes the GPON fully
Garrich et al. [170] have recently designed and demon- programmable similar to a conventional OpenFlow switch.
strated a CDC ROADM with an add-drop bank based on Preliminary steps towards the GPON virtual switch design
an Optical Cross-Connect (OXC) backplane [222]. The OXC have been taken by Gu et al. [175] who developed com-
backplane allows for highly flexible add/drop configurations ponents for SDN control of a PON in a data center and
implemented through SDN control. The backplane based Amokrane et al. [176], [177] who developed a module for
ROADM has been analytically compared with prior designs mapping OpenFlow flow control requests into PON con-
based on wavelength selective and multicast switches and has figuration commands. Lee et al. [174] have expanded on
been shown to achieve higher flexibility and lower losses. this groundwork to abstract the entire GPON into a virtual
An experimental evaluation has tested the backplane based OpenFlow switch. More specifically, Lee et al. [174] have
ROADM for a metropolitan area mesh network extending over comprehensively designed a hardware architecture and a soft-
100 km with an aggregate traffic load of close to 9 Tb/s. ware architecture to allow SDN control to interface with the
b) Open transport switch (OTS): The Open Transport virtual GPON as if it were a standard OpenFlow switch. The
Switch (OTS) [171] is an OpenFlow-enabled optical virtual experimental performance evaluation of the designed GPON
switch design. The OTS design abstracts the details of the virtual switch measured response times for flow entry modi-
underlying physical switching layer (which could be packet fications from an ONU port (where a subscriber connects to
switching or circuit switching) to a virtual switch element. The the virtual GPON switch) to an SDN external port around
OTS design introduces three agent modules (discovery, con- 0.6 ms, which compares to 0.2 ms for a corresponding flow
trol, and data plane) to interface with the physical switching entry modification in a conventional OFsoftswitch and 1.7 ms
hardware. These agent modules are controlled from an SDN in a EdgeCore AS4600 switch. In a related study on SDN con-
controller through extended OpenFlow messages. Performance trolled switching in a PON, Yeh et al. [178] have designed an
measurements for an example testbed network setup indicate ONU with an optical switch that selects OFDM subchannels
that the circuit path computation latencies on the order of 2–3 s in a TWDM-PON. The switch in the ONU allows for flexi-
that can be reduced through faster processing in the controller. ble dynamic adaption of the downstream bandwidth through
c) Logical xBar: The logical xBar [172] has been defined SDN. Gu et al. [179] have examined the flexible SDN con-
to represent a programmable switch. An elementary (small) trolled re-arrangement of ONUs to OLTs so as to efficiently
xBar could consist of a single OpenFlow switch. Multiple support PON service with network coding [224].
small xBars can be recursively merged to form a single large f) Flexi access network node: A flexi-node for an access
xBar with a single forwarding table. The xBar concept envi- network that flexibly aggregates traffic flows from a wide range
sions that xBars are the building blocks for forming large of networks, such as local area networks and base stations
networks. Moreover, labels based on SDN and MPLS are envi- of wireless networks has been proposed in [180]. The flexi-
sioned for managing the xBar data plane forwarding. The xBar node design is motivated by the shortcomings of the currently
concepts have been further advanced in the Orion study [223] deployed core/metro network architectures that attempt to con-
to achieve low computational complexity of the SDN control solidate the access and metro networks. This consolidation
plane. forces all traffic in the access network to traverse the metro
2748 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 18, NO. 4, FOURTH QUARTER 2016

network, even if the traffic is destined to destination nodes performance results for throughput (in bit/s and packets/s) to
in the coverage area of an access network. In contrast, the demonstrate the feasibility of the proposed unified OpenFlow
proposed flexi-node encompasses electrical and optical for- switching network.
warding capabilities that can be controlled through SDN. The b) R-LR-UFAN: The Reconfigurable Long-Reach
flexi-node can thus serve as an effective aggregation node in UltraFlow Access Network (R-LR-UFAN) [188], [189]
access-metro networks. Traffic that is destined to other nodes provides flexible dual-mode transport service based on either
in the coverage area of an access network can be sent directly the Internet Protocol (IP) or Optical Flow Switching (OFS).
to the access network. OFS [225] provides dedicated end-to-end network paths
Kondepu et al. [181] have similarly presented an SDN through purely optical switching, i.e., there is no electronic
based PON aggregation node. In their architecture, multi- processing or buffering at intermediate network nodes. The
ple ONUs communicate with the SDN controller within the R-LR-UFAN architecture employs multiple feeder fibers to
aggregation node to request the scheduling of upstream trans- form subnets within the network. UltraFlow coexists alongside
mission resources. ONUs are then serviced by multiple Optical the conventional PON OLT and ONUs. The R-LR-UFAN
Service Units (OSUs) which exist within the aggregation node introduces new entities, namely the Optical Flow Network
alongside with the SDN controller. OSUs are then configured Unit (OFNU) and the SDN-controlled Optical Flow Line
by the controller based on Time and Wavelength Division Terminal (OFLT). A Quasi-PAssive Reconfigurable (QPAR)
Multiplexed (TWDM) PON. The OSUs step between normal node [190] is introduced between the OFNU and OFLT. The
and sleep-mode depending on the traffic loads, thus saving QPAR node can re-route intra PON traffic between OFNUs
power. without having to pass through the OLFTs. The optically
2) Switching Paradigms: rerouted intra-PON channels can be used for communication
a) Converged packet-circuit switching: Hybrid packet- between wireless base stations supporting inter cell device-
circuit optical network infrastructures controlled by SDN have to-device communication. The testbed evaluations indicate
been explored in a few studies. Das et al. [182] have described that for an intra-PON traffic ratio of 0.3, the QPAR strategy
how to unify the control and management of circuit- and achieves power savings up to 24%.
packet-switched networks using OpenFlow. Since packet- and c) Flexi-grid: The principle of flexi-grid (elastic) optical
circuit-switched networking are extensively employed in opti- networking [31], [87]–[92], [226] has been explored in several
cal networks, examining their integration is an important SDN infrastructure studies. Generally, flexi-grid networking
research direction. Das et al. [182] have given a high-level strives to enhance the efficiency of the optical transmissions
overview of a flow abstraction for each type of switched net- by adapting physical (photonic) transmission parameters, such
work and a common control paradigm. In their follow-up as modulation format, symbol rate, number and spacing of
work, Das et al. [183] have described how a packet and cir- subcarrier wavelength channels, as well as the ratio of for-
cuit switching network can be implemented in the context of ward error correction to payload. Flexi-grid transmissions
an OpenFlow-protocol based testbed. The testbed is a stan- have become feasible with high-capacity flexible transceivers.
dard Ethernet network that could generally be employed in Flexi-grid transmissions use narrower frequency slots (e.g.,
any access network with Time Division Multiplexing (TDM). 12.5 GHz) than classical Wavelength Division Multiplexing
Veisllari et al. [184] studied packet/circuit hybrid optical long- (WDM, with typically 50 GHz frequency slots for WDM)
haul metro access networks. Although Veisllari et al. [184] and can flexibly form optical transmission channels that span
indicated that SDN can be used for load balancing in multiple contiguous frequency slots.
the proposed packet/circuit network, no detailed study of Cvijetic [191] has proposed a hierarchical flexi-grid infras-
such an SDN-based load balancing has been conducted tructure for multiservice broadband optical access utilizing
in [184]. Related switching paradigms that integrate SDN centralized software-reconfigurable resource management and
with Generalized Multiple Protocol Label Switching (GMPLS) digital signal processing. The proposed flexi-grid infras-
have been examined in [185] and [186], while data center tructure incorporates mobile backhaul, as well as SDN
specific aspects have been surveyed in [130]. controlled transceivers (Section III-A). In follow-up work,
Cerroni et al. [187] have further developed the con- Cvijetic et al. [192] have designed a dynamic flexi-grid
cept of unifying circuit- and packet-switching networks with optical access and aggregation network. They employ SDN
OpenFlow, which was initiated by Das et al. [182], [183]. to control tunable lasers in the OLT for flexible down-
The unification is accomplished with SDN on the net- stream transmissions. Flexi-grid wavelength selective switches
work layer and can be used in core networks. Specifically, are controlled through SDN to dynamically tune the pass-
Cerroni et al. [187] have described an extension of the band for the upstream transmissions arriving at the OLT.
OpenFlow flow concept to support hybrid networks. OpenFlow Cvijetic et al. [192] obtained good results for the upstream and
message format extensions to include matching rules and flow downstream bit error rate and were able to provide 150 Mb/s
entries have also been provided. The matching rules can repre- per wireless network cell.
sent different transport functions, such as a channel on which a Oliveira et al. [193] have demonstrated a testbed for
packet is received in optical circuit-switched WDM networks, a Reconfigurable Flexible Optical Network (RFON), which
time slots in TDM networks, or transport class services (such was one of the first physical layer SDN-based testbeds.
as guaranteed circuit service or best effort packet service). The RFON testbed is comprised of 4 ROADMs with flexi-
Cerroni et al. [187] have presented a testbed setup and reported grid Wavelength Selective Switching (WSS) modules, optical
THYAGATURU et al.: SDONs: A COMPREHENSIVE SURVEY 2749

amplifiers, optical channel monitors and supervisor boards. Detrimental non-ideal filtering effects accumulate and impair
The controller daemon implements a node abstraction layer the OSNR [202]. At the same time, Erbium Doped Fiber
and provides configuration details for an overall view of the Amplifiers (EDFAs) [234] are widely deployed in optical net-
network. Also, virtualization of the GMPLS control plane with works to boost optical signal power that has been depleted
topology discovery and Traffic Engineering (TE)-link instanti- through attenuation in fibers and ROADMs. However, depend-
ation have been incorporated. Instead of using OpenFlow, the ing on their operating points, EDFAs can introduce significant
RFON testbed uses the controller language YANG [227] to noise. Moura et al. [200], [201] have explored SDN based
obtain the topology information and collect monitoring data adaptation strategies for EDFA operating points to increase
for the lightpaths. the OSNR. In a complementary study, Paolucci et al. [202]
Zhao et al. [194] have presented an architecture with have exploited SDN control to reduce the detrimental filter-
OpenFlow-based optical interconnects for intra-data center ing effects. Paolucci group wavelength channels that jointly
networking and OpenFlow-based flexi-grid optical networks traverse a sequence of filters at successive switching nodes.
for inter-data center networking. Zhao et al. [194] focus on the Instead of passing these wavelength channels through individ-
SDN benefits for inter-data center networking with heteroge- ual (per-wavelength channel) filters, the group of wavelength
neous networks. The proposed architecture includes a service channels is jointly passed through a superfilter that encom-
controller, an IP controller, and an optical controller based on passes all grouped wavelength channels. This joint filtering
the Father Network Operating System (F-NOX) [228], [229]. significantly improves the OSNR.
The performance evaluations in [194] include results for While the studies [200]–[202] have focused on either the
blocking probability, release latency, and bandwidth spectrum EDFA or the filters, Carvalho et al. [203] and Wang et al. [204]
characteristics. have jointly considered the EDFA and filter control. More
specifically, the EDFA gain and the filter attenuation (and
signal equalization) profile were adapted to improve the
D. Optical Performance Monitoring OSNR. Carvalho et al. [203] propose and evaluate a specific
1) Cognitive Network Infrastructure: A Cognitive joint EDFA and filter optimization approach that exploits the
Heterogeneous Reconfigurable Dynamic Optical Network global perspective of the SDN controller. The global opti-
(CHRON) architecture has been outlined in [195]–[197]. mization achieves ONSR improvements close to 5 dB for a
CHRON senses the current network conditions and adapts testbed consisting of four ROADMs with 100 km fiber links.
the network operation accordingly. The three main com- Wang et al. [204] explore different combinations of EDFA
ponents of CHRON are monitoring elements, software gain control strategies and filter equalization strategies for a
adaptable elements, and cognitive processes. The monitoring simulated network with 14 nodes and 100 km fiber links. They
elements observe two main types of optical transmission find mutual interactions between the EDFA gain control and
impairments, namely non-catastrophic impairments and the filter equalization control as well as an additional wave-
catastrophic impairments. Non-catastrophic impairments length assignment module. They conclude that global SDN
include the photonic impairments that degrade the Optical control is highly useful for synchronizing the EDFA gain and
Signal to Noise Ratio (OSNR), such as the various forms of filter equalization in conjunction with wavelength assignments
dispersion, cross-talk, and non-linear propagation effects, but so as to achieve improved OSNR.
do not completely disrupt the communication. In contrast, a
catastrophic impairment, such as a fiber cut or malfunctioning
switch, can completely disrupt the communication. Advances E. Infrastructure Layer: Summary and Discussion
in optical performance monitoring allow for in-band OSNR The research to date on the SDN controlled infrastructure
monitoring [230]–[233] at midpoints in the communication layer has resulted in a variety of SDN controlled transceivers
path, e.g., at optical amplifiers and ROADMs. as well as a few designs of SDN controlled switching elements.
The cognitive processes involve the collection of the mon- Moreover, the SDN control of switching paradigms and opti-
itoring information in the controller, executing control algo- cal performance monitoring have been examined. The SDN
rithms, and instructing the software adaptable components to infrastructure studies have paid close attention to the physi-
implement the control decisions. SDN can provide the frame- cal (photonic) communication aspects. Principles of isolation
work for implementing these cognitive processes. Two main of control plane and data plane with the goals of simpli-
types of software adaptable components have been considered fying network management and making the networks more
so far [198], [199], namely control of transceivers and control flexible have been explored. The completed SDN infrastruc-
of wavelength selective switches/amplifiers. For transceiver ture layer studies have indicated that the SDN control of
control, the cognitive control adjusts the transmission param- the infrastructure layer can reduce costs, facilitate flexible
eters. For instance, transmission bit rates can be adjusted reconfigurable resource management, increase utilizations, and
through varying the modulation format or the number of signal lower latency. However, detailed comprehensive optimizations
carriers in multicarrier communication (see Section III-A). of the infrastructure components and paradigms that mini-
2) Wavelength Selective Switch/Amplifier Control: In gen- mize capital and operational expenditures are an important
eral, ROADMs (see Section III-C1a) employ wavelength area for future research. Also, further refinements of the opti-
selective switches based on filters to add or drop wave- cal components and switching paradigms are needed to ease
length channels for routing through an optical network. the deployment of SDONs and make the networks operating
2750 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 18, NO. 4, FOURTH QUARTER 2016

Fig. 6. Classification of SDON control layer studies.

on the SDON infrastructures more efficient. Moreover, the for retro-fitting non-SDN optical network elements so that
cost reduction of implementations, easy adoption by network they can be controlled by OpenFlow. The retro-fitting typically
providers, flexible upgrades to adopt new technologies, and involves the insertion of an abstraction layer into the network
reduced complexity require thorough future research. elements. The abstraction layer makes the optical hardware
Most SDON infrastructure studies have focused on a par- controllable by OpenFlow. The retro-fitting studies would also
ticular network component or networking aspect, e.g., a fit into Section III as the abstraction layer is inserted into
transceiver or the hybrid packet-circuit switching paradigm, the network elements; however, the abstraction mechanisms
or a particular application context, e.g., data center net- closely relate to the OpenFlow extensions for optical network-
working. Future research should comprehensively examine ing and we include the retro-fitting studies therefore in this
SDON infrastructure components and paradigms to optimize control layer section. We then survey the various SDN con-
their interactions for a wide set of networking scenarios and trol mechanisms for operational aspects of optical networks,
application contexts. including the control of tandem networks that include opti-
The SDON infrastructure studies to date have primar- cal segments. Lastly, we survey SDON controller performance
ily focused on the optical transmission medium. Future analysis studies.
research should explore complementary infrastructure com-
ponents and paradigms to support transmissions in hybrid A. SDN Control of Optical Infrastructure Components
fiber-wireless and other hybrid fiber-X networks, such as
1) Controlling Optical Transceivers With OpenFlow:
fiber-Digital Subscriber Line (DSL) or fiber-coax cable
Recent generations of optical transceivers utilize digital
networks [111], [235], [236]. Generally, the flexible SDN con-
signal processing techniques that allow many parame-
trol can be very advantageous for hybrid networks composed
ters of the transceiver to be software controlled (see
of heterogeneous network segments. The OpenFlow protocol
Sections III-A1 and III-A2). These parameters include mod-
can facilitate the topology abstraction of the heterogeneous
ulation scheme, symbol rate, and wavelength. Yu et al. [237]
physical transmission media, which in turn facilitates control
and Chen et al. [238] proposed adding a “modulation format”
and optimization at the higher network protocol layers.
field to the OpenFlow cross-connect table entries to support
this programmable feature of some software defined optical
IV. SDN C ONTROL L AYER transceivers.
This section surveys the SDON studies that are focused Ji et al. [239] created a testbed that places super-channel
on applying the SDN principles at the SDN control layer to optical transponders and optical amplifiers under SDN con-
control the various optical network elements and operational trol. An OpenFlow extension is proposed to control these
aspects. The main challenges of SDON control include exten- devices. The modulation technique and FEC code for each
sions of the OpenFlow protocol for specifically controlling the optical subcarrier of the super-channel transponder and the
optical transmission and switching components surveyed in optical amplifier power level can be controlled via OpenFlow.
Section III and for controlling the optical spectrum as well as Ji et al. [239] do not discuss this explicitly but the transpon-
for controlling optical networks spanning multiple optical net- der subcarriers can be treated as OpenFlow switch ports
work tiers (see Section II-D2). As illustrated in Fig. 6, we first that can be configured through the OpenFlow protocol via
survey SDN control mechanisms and frameworks for control- port modification messages. It is unclear in [239] how the
ling infrastructure layer components, namely transceivers as amplifiers would be controlled via OpenFlow. However, doing
well as optical circuit, packet, and burst switches. More specif- so would allow the SDN controller to adaptively modify
ically, we survey OpenFlow extensions for controlling the amplifiers to compensate for channel impairments while min-
optical infrastructure components. We then survey mechanisms imizing energy consumption. Ji et al. [239] have established
THYAGATURU et al.: SDONs: A COMPREHENSIVE SURVEY 2751

a testbed demonstrating the placement of transponders and

EDFA optical amplifiers under SDN control.
Liu et al. [34] propose configuring optical transponder oper-
ation via flow table entries with new transponder specific fields
(without providing details). They also propose capturing fail-
ure alarms from optical transponders and sending them to the
SDN controller via OpenFlow Packet-In messages. These mes-
sages are normally meant to establish new flow connections.
Alternatively, a new OpenFlow message type could be cre-
ated for the purpose of capturing failure alarms [34]. With
failure alarm information, the SDN controller can implement
protection switching services.
2) Controlling Optical Circuit Switches With OpenFlow: Fig. 7. Traditional non-SDN network elements can be retro-fitted for con-
Circuit switching can be enabled by OpenFlow by adding new trol by an SDN controller using OpenFlow using a hardware abstraction
circuit switching flow table entries [182], [183], [240], [243]. layer [34], [247]–[250].
The OpenFlow circuit switching addendum [241] discusses
the addition of cross-connect tables for this purpose. These
entry to the electrical domain for forwarding to the SDN con-
cross-connect tables are configured via OpenFlow messages
troller, and (iii) a wavelength identifier extension to the flow
inside the circuit switches. According to the addendum, a
table entries. To compensate for either the lack of any opti-
cross-connect table entry consists of the following fields to
cal buffering or limited optical buffering, an SDN controller,
identify the input:
with its global view, can provide more effective means to
• Input Port
resolve contention that would lead to packet loss in optical
• Input Wavelength
packet switches. Specifically, Cao et al. [244] suggest to select
• Input Time Slot
the path with the most available resources among multiple
• Virtual Concatenation Group
available paths between two nodes. Paths can be re-computed
and the following fields to identify the output:
periodically or on-demand to account for changes in traffic
• Output Port
conditions. Monitoring messages can be defined to keep the
• Output Wavelength
SDN controller updated of network traffic conditions.
• Output Time Slot
Engineers with Japan’s National Institute of Information and
• Virtual Concatenation Group
Communications Technology [245] have created an optical
These cross-connect tables cover circuit switching in space,
circuit and packet switched demonstration system in which
fixed-grid wavelength, and time.
the packet portion is SDN controlled. The optical circuit
Channegowda et al. [33], [242] extend the capabilities of
switching is implemented with Wavelength Selective Switches
the OpenFlow circuit switching addendum to support flexible
(WSSs) and the optical packet switching is implemented with
wavelength grid optical switching. Specifically, the wave-
an Semiconductor Optical Amplifier (SOA) switch.
length identifier specified in the circuit switching addendum
OpenFlow flow tables can also be used to configure optical
to OpenFlow is replaced with two fields: center frequency,
burst switching devices [34]. When there is no flow table entry
and slot width. The center frequency is an integer specifying
for a burst of packets, the optical burst switching device can
the multiple of 6.25 GHz the center frequency is away from
send the Burst Header Packet (BHP) to the SDN controller
193.1 Thz and the slot width is a positive integer specifying
to process the addition of the new flow to the network [34]
the spectral width in multiples of 12.5 GHz.
rather than the first packet in the burst.
An SDN controlled optical network testbed at the University
of Bristol has been established to demonstrate the OpenFlow
extensions for flexible grid DWDM [33]. The testbed consists B. Retro-Fitting Devices to Support OpenFlow
of both fixed-grid and flexible-grid optical switching devices. An abstraction layer can be used to turn non-SDN opti-
South Korea Telekom has also built an SDN controlled optical cal switching devices into OpenFlow controllable switching
network testbed [277]. devices [33], [34], [242], [244], [246]. As illustrated in Fig. 7,
3) Controlling Optical Packet and Burst Switches With the abstraction layer provides a conversion layer between
OpenFlow: OpenFlow flow tables can be utilized in opti- OpenFlow configuration messages and the optical switch-
cal packet switches for expressing the forwarding table and ing devices’ native management interface, e.g., the Simple
its computation can be offloaded to an SDN controller. This Network Management Protocol (SNMP), the Transaction
offloading can simplify the design of highly complex optical Language 1 (TL1) protocol, or a proprietary (vendor-specific)
packet switches [244]. API. Additionally, a virtual OpenFlow switch with virtual
Cao et al. [244] extend the OpenFlow protocol to work interfaces that correspond to physical switching ports on
with Optical Packet Switching (OPS) devices by creating: the non-SDN switching device completes the abstraction
(i) an abstraction layer that converts OpenFlow configura- layer [34], [247]–[250]. When a flow entry is added between
tion messages to the native OPS configuration, (ii) a process two virtual ports in the virtual OpenFlow switch, the abstrac-
that converts optical packets that do not match a flow table tion layer uses the switching devices’ native management
2752 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 18, NO. 4, FOURTH QUARTER 2016

controller and OLTs. However, Khalili et al. [252] identify

ONU registration policy and coarse timescale DBA policy as
functions that operate at timescales that allow effective offload-
ing to an SDN controller. Yan et al. [256] further identify OLT
and ONU power control for energy savings as a function that
can be effectively offloaded to an SDN controller.
There is also a movement to use PONs in edge networks
to provide connectivity inside a multitenant building or on a
campus with multiple buildings [254], [255]. The use of PONs
in this edge scenario requires rapid re-provisioning from the
OLT. A software controlled PON can provide this needed rapid
reprovisioning [254], [255].
Kanonakis et al. [257] propose leveraging the broad per-
spective that SDN can provide to perform dynamic bandwidth
allocation across several Virtual PONs (VPONs). The VPONs
Fig. 8. Non-SDN OLTs can be retro-fitted for control by an SDN controller are separated on a physical PON by the wavelength bands
using OpenFlow [251]. that they utilize. Bandwidth allocation is performed at the
granularity of OFDMA subcarriers that compose the optical
spectrum.
interface to add the flow entry between the two corresponding 2) SDN Control of Optical Spectrum Defragmentation: In
physical ports. a departure from the fixed wavelength grid (ITU-T G.694.1),
A non-SDN PON OLT can be supplemented with a two- elastic optical networking allows flexible use of the optical
port OpenFlow switch and a hardware abstraction layer that spectrum. This flexibility can permit higher spectral effi-
converts OpenFlow forwarding rules to control messages ciency by avoiding consuming an entire fixed-grid wavelength
understood by the non-SDN OLT [251]. Fig. 8 illustrates this channel when unnecessary and avoiding unnecessary guard
OLT retro-fit for SDN control via OpenFlow. In this way the bands in certain circumstances [93]. However, this flexibility
PON has its switching functions controlled by OpenFlow. causes fragmentation of the optical spectrum as flexible grid
lightpaths are established and terminated over time.
C. SDN Control of Optical Network Operation Spectrum fragmentation leads to the circumstance in which
1) Controlling Passive Optical Networks With OpenFlow: there is enough spectral capacity to satisfy a demand but
An SDN controlled PON can be created by upgrading OLTs that capacity is spread over several fragments rather than
to SDN-OLTs that can be controlled using a Southbound being consolidated in adjacent spectrum as required. If the
Interface, such as OpenFlow [252], [253]. A centralized PON fragmentation is not counter-acted by a periodic defragmenta-
controller, potentially executing in a data center, controls tion process than overall spectral utilization will suffer. This
one or more SDN-OLTs. The advantage of using SDN is resource fragmentation problem appears in computer systems
the broadened perspective of the PON controller as well as in main memory and long term storage. In those contexts the
the potentially reduced cost of the SDN-OLT compared to a problem is typically solved by allowing the memory to be
non-SDN OLT. allocated using non-adjacent segments. Memory and storage
Parol and Pawlowski [254], [255] define OpenFlowPLUS is partitioned into pages and blocks, respectively. The alloca-
to extend the OpenFlow SBI for GPON. OpenFlowPLUS tions of pages to a process or blocks to a file do not need
extends SDN programmability to both OLT and ONU devices to be contiguous. With communication spectrum this would
whereby each act as an OpenFlow switch through a pro- mean combining multiple small bandwidth channels through
grammable flow table. Non-switching functions (e.g., ONU inverse multiplexing to create a larger channel [258].
registration, dynamic bandwidth allocation) are outside the An SDN controller can provide a broad network per-
scope of OpenFlowPLUS. OpenFlowPLUS extends OpenFlow spective to empower the periodic optical spectrum defrag-
by channeling OpenFlow messages through the GPON ONU mentation process to be more effective [258]. In general,
Management and Control Interface (OMCI) control chan- optical spectrum defragmentation operations can reduce light-
nel and adding PON specific action instructions to flow path blocking probabilities from 3% [237] up to as much as
table entries. The PON specific action instructions defined in 75% [238], [259]. Multicore fibers provide additional spec-
OpenFlowPLUS are: tral resources through additional transmission cores to permit
• (new gpon action type): map matching packets to PON quasi-hitless defragmentation [260].
specific traffic identifiers, e.g., GPON Encapsulation 3) SDN Control of Tandem Networks:
Method (GEM) ports and GPON Traffic CONTainers a) Metro and access: Wu et al. [261] and
(T-CONTs) Zhao et al. [262] propose leveraging the broad perspec-
• (output action type): activate PON specific framing of tive that SDN can provide to improve bandwidth allocation.
matching packets Two cooperating stages of SDN controllers: (i) access stage
Many of the OLT functions operate at timescales that are that controls each SDN OLT individually, and (ii) metro
problematic for the controller due to the latency between the stage that controls global bandwidth allocation strategy,
THYAGATURU et al.: SDONs: A COMPREHENSIVE SURVEY 2753

can coordinate bandwidth allocation across several physical network bandwidth. Bandwidth allocation that utilizes the
PONs [261], [262]. The bandwidth allocation is managed broad perspective that SDN can provide is critical for rea-
cooperatively among the two stages of SDN controllers to sonable VM migration latencies without sacrificing network
optimize the utilization of the access and metro network bandwidth utilization.
bandwidth. Simulation experiments indicate a 40% increase e) Internet of things: Wang et al. [268] examine
in network bandwidth utilization as a result of the global another use case for SDN bandwidth allocation across net-
coordination compared to operating the bandwidth allocation work segments: the Internet of Things (IoT). Specifically,
only within the individual PONs [261], [262]. Wang et al. [268] have developed a Dynamic Bandwidth
b) Access and wireless: Bojic et al. [263] expand on the Allocation (DBA) protocol that exploits SDN control for mul-
concept of SDN controlled OFDMA enabled VPONs [257] to ticasting and suspending flows. This DBA protocol is studied
provide mobile backhaul service. The backhaul service can be in the context of a virtualized WDM optical access net-
provided for wireless small-cell sites (e.g., micro and femto work that provides IoT services through the distributed ONUs
cells) that utilize millimeter wave frequencies. Each small- to individual devices. The SDN controller employs multi-
cell site contains an OFDMA-PON ONU that provides the casting and flow suspension to efficiently prioritize the IoT
backhaul service through the access network over a VPON. service requests. Multicasting allows multiple requests to share
An SDN controller is utilized to assign bandwidth to each resources in the central nodes that are responsible for process-
small-cell site through OFDMA subcarrier assignment in a ing a prescribed wavelength in the central office (OLT). Flow
VPON to the constituent ONU. The SDN controller leverages suspension allows high-priority requests (e.g., an emergency
its broad view of the network to provide solutions to the joint call) to suspend ongoing low-priority traffic flows (e.g., rou-
bandwidth allocation and routing across several network seg- tine meter readings). Performance results for a real-time SDN
ments. With this broad perspective of the network, the SDN controller implementation indicate that the proposed band-
controller can make globally rather than just locally optimal width (resource) allocation with multicast and flow suspension
bandwidth allocation and routing decisions. Efficient optimiza- can improve several key performance metrics, such as request
tion algorithms, such as genetic algorithms, can be used to serving ratio, revenue, and delays by 30–50% [268].
provide computationally efficient competitive solutions, miti-
gating computational complexity issues associated with opti-
mization for large networks. Additionally, network partitioning D. Hybrid SDN-GMPLS Control
with an SDN controller for each partition can be used to mit- 1) Generalized MultiProtocol Label Switching (GMPLS):
igate unreasonable computational complexity that arises when Prior to SDN, MultiProtocol Label Switching (MPLS) offered
scaling to large networks. Tanaka and Cvijetic [264] presented a mechanism to separate the control and data planes through
one such optimization formulation for maximizing throughput. label switching. With MPLS, packets are forwarded in a
Costa-Requena et al. [265] described a proof-of-concept connection-oriented manner through Label Switched Paths
LTE testbed they have constructed whereby the network (LSPs) traversing Label Switching Routers (LSRs). An entity
consists of software defined base stations and various net- in the network establishes an LSP through a network of LSRs
work functions executing on cloud resources. The testbed is for a particular class of packets and then signals the label-
described in broad qualitative terms, no technical details are based forwarding table entries to the LSRs. At each hop along
provided. There was no mathematical or experimental analysis an LSP, a packet is assigned a label that determines its forward-
provided. ing rule at the next hop. At the next hop, that label determines
c) Access, metro, and core: Slyne and Ruffini [266] pro- that packet’s output port and label for the next hop; the process
vide a use case for SDN switching control across network repeats until the packet reaches the end of the LSP. Several sig-
segments: use Layer 2 switching across the access, metro, and nalling protocols for programming the label-based forwarding
core networks. Layer 2 (e.g., Ethernet) switching does not table entries inside LSRs have been defined, e.g., through the
scale well due to a lack of hierarchy in its addresses. That lack Resource Reservation Protocol (RSVP). Generalized MPLS
of hierarchy does not allow for switching rules on aggregates (GMPLS) extends MPLS to offer circuit switching capabil-
of addresses thereby limiting the scaling of these networks. ity. Although never commercially deployed [34], GMPLS and
Slyne and Ruffini [266] propose using SDN to create hierar- a centralized Path Computation Element (PCE) [278]–[281]
chical pseudo-MAC addresses that permit a small number of have been considered for control of optical networks.
flow table entries to configure the switching of traffic using 2) Path Computation Element (PCE): A PCE is a concept
Layer 2 addresses across network segments. The pseudo-MAC developed by the IETF (see RFC 4655) to refer to an entity
addresses encode information about the device location to per- that computes network paths given a topology and some crite-
mit simple switching rules. At the entry of the network, flow ria. The PCE concept breaks the path computation action from
table entries are set up to translate from real (non-hierarchical) the forwarding action in switching devices. A PCE could be
MAC addresses to hierarchical pseudo-MAC addresses. The distributed in every switching element in a network domain
reverse takes place at the exit point of the network. or there could be a single centralized PCE for an entire net-
d) DC virtual machine migration: Mandal et al. [267] work domain. The network domain could be an area of an
provided a cloud computing use case for SDN bandwidth allo- Autonomous System (AS), an AS, a conglomeration of sev-
cation across network segments: Virtual Machine (VM) migra- eral ASes, or just a group of switching devices relying on one
tion between data centers. VM migrations require significant PCE. Some of an SDN controller’s functionality falls under
2754 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 18, NO. 4, FOURTH QUARTER 2016

the classification of a centralized PCE. However, the PCE con-

cept does not include the external configuration of forwarding
tables. Thus, a centralized PCE device does not necessarily
have a means to configure the switching elements to provision
a computed path.
When the entity requesting path computation is not co-
located with the PCE, a PCE Communication Protocol (PCEP)
is used over TCP port 4189 to facilitate path computation
requests and responses. The PCEP consists of the following
message types:
• Session establishment messages (open, keepalive, close)
• PCReq – Path computation request
• PCRep – Path computation reply
• PCNtf – event notification Fig. 9. Hybrid GMPLS/PCE and SDN network control: (a) SDN controller
• PCErr – signal a protocol error utilizes a PCE to control a portion of the network [185], [258] through the
The path computation request message must include the end NETCONF protocol or a proprietary command line interface (CLI). (b) SDN
controller utilizes the path computation ability of the PCE [246], [269], [270]
points of the path and can optionally include the requested and controls network through OpenFlow protocol.
bandwidth, the metric to be optimized in the path computa-
tion, and a list of links to be included in the path. The Path
computation reply includes the computed path expressed in using an SBI protocol, such as OpenFlow [246], [269], [270];
the Explicit Route Object format (see RFC 3209) or an indi- see illustration b) in Fig. 9.
cation that there is no path. See RFC 5440 for more details
on PCEP.
A PCE has been proposed as a central entity to manage a E. SDN Performance Analysis
GMPLS-enabled optical circuit switched network. Specifically, 1) SDN vs. GMPLS: Liu et al. [271] provided a qual-
the PCE maintains the network topology in a structure called itative comparison of GMPLS, GMPLS/PCE, and SDN
the Traffic Engineering Database (TED). The traffic engi- OpenFlow for control of wavelength switched optical net-
neering modifier (see RFC 2702) signifies that the path works. Liu et al. [271] noted that there is an evolution of cen-
computations are made to relieve congestion that is caused by tralized control from GMPLS to GMPLS/PCE to OpenFlow.
the sub-optimal allocation of network resources. This modifier Whereas GMPLS offers distributed control, GMPLS/PCE
is used extensively in discussions of MPLS/GMPLS because is commonly regarded as having centralized path compu-
their use case is for traffic engineering; in acronym form the tation but still distributed provisioning/configuration; while
modifier is TE (e.g., TE LSP, RSVP-TE). OpenFlow centralizes all of the network control. In our dis-
If the PCE is stateful with complete control over its network cussion in Section IV-D we noted that a stateful PCE with
domain, it will also maintain an LSP database recording the instantiation capabilities centralizes all network control and
provisioned GMPLS lightpaths. A lightpath request can be sent is therefore very similar to SDN. Liu et al. [271] have also
to the PCE, it will use the topology and LSP database to find pointed out that GMPLS/PCE is more technically mature com-
the optimal path and then configure the GMPLS-controlled pared to OpenFlow with IETF RFCs for GMPLS (see RFC
optical circuit switching nodes using NETCONF (see RFC 3471) and PCE (see RFC 4655) that date back to 2003 and
6241) or proprietary command line interfaces (CLIs) [258]. 2006, respectively. SDN has just recently, in 2014, received
This stateful PCE with instantiation capabilities (capabilities standardization attention from the IETF (see RFC 7149).
to provision lightpaths) operates similarly to an SDN con- A comparison of GMPLS and OpenFlow has been con-
troller. For that reason, GMPLS with a centralized stateful ducted by Zhao et al. [272] for large-scale optical networks.
PCE with instantiation capabilities can provide a baseline for Two testbeds were built, based on GMPLS and on OpenFlow,
performance analysis of an SDN controller as well as provide respectively. Performance metrics, such as blocking proba-
a mechanism to be blended with an SDN controller for hybrid bility, wavelength utilization, and lightpath setup time were
control [33], [242], [246]. evaluated for a 1000 node topology. The results indicated that
3) Approaches to Hybrid SDN-GMPLS Control: Hybrid GMPLS gives slightly lower blocking probability. However,
GMPLS/PCE and SDN control can be formed by allowing an OpenFlow gives higher wavelength utilization and shorter
SDN controller to leverage a centralized PCE to control a por- average lightpath setup time. Thus, the results suggest that
tion of the infrastructure using PCEP as the SBI [185], [258]; OpenFlow is overall advantageous compared to GMPLS in
see illustration a) in Fig. 9. The SDN controller builds higher large-scale optical networks.
functionality above what the PCE provides and can possibly Cvijetic et al. [273] conducted a numerical analysis to
control a large network that utilizes several PCEs as well as compare the computed shortest path lengths for non-SDN,
OpenFlow controlled network elements. partial-SDN, and full-SDN optical networks. A full-SDN net-
Alternatively, the SDN controller can leverage a PCE for its work enables path lengths that are approximately a third of
path computation abilities with the SDN controller handling those computed on a non-SDN network. These path lengths
the configuration of the network elements to establish a path can also translate into an energy consumption measure, with
THYAGATURU et al.: SDONs: A COMPREHENSIVE SURVEY 2755

shortest paths resulting in reduced energy consumption. An but provide reliability of the network control under failure
SDN controlled network can result in smaller computed short- of the ring network. In a multiwavelength ring network, a
est paths that translates to smaller network latency and energy separate wavelength can be allocated to carry the control
consumption [273]. traffic. Sanchez et al. [275] focused on a Tunable Transceiver
Experiments conducted on the testbed described in [242] Fixed Receiver (TTFR) WDM ring node architecture. In this
show a 4% reduction in lightpath blocking probability using architecture each node receives data on a home wavelength
SDN OpenFlow compared to GMPLS for lightpath provision- channel and has the capability to transmit on any of the
ing. The same experiments show that lightpath setup times can available wavelengths to reach any other node. The addition
be reduced to nearly half using SDN OpenFlow compared to of the out-of-band control channel on a separate wavelength
GMPLS. Finally, the experiments show that an Open vSwitch requires each node to have an additional fixed receiver, thereby
based controller can process about three times the number of increasing cost. Sanchez et al. [275] identified a clear trade-
flows per second as a NOX [228] based controller. off between cost and reliability when comparing the four
2) SDN Controller Flow Setup: Veisllari et al. [274] eval- architectures.
uated the use of SDN to support both circuit and packet 4) Clustered SDN Control: Penna et al. [276] described
switching in a metropolitan area ring network that intercon- partitioning a wavelength-switched optical network into
nects access network segments with a backbone network. This administrative domains or clusters for control by a single SDN
network is assumed to be controlled by a single SDN con- controller. The clustering should meet certain performance cri-
troller. The objective of the study [274] was to determine the teria for the SDN controller. To permit lightpath establishment
effect of packet service flow size on the required SDN con- across clusters, an inter-cluster lightpath establishment proto-
troller flow service time to meet stability conditions at the col is established. Each SDN controller provides a lightpath
controller. Toward this end, Veisllari et al. [274] produced a establishment function between any two points in its associ-
mean arrival rate function of new packet and circuit flows at ated cluster. Each SDN controller also keeps a global view
that controller. This arrival rate function was visualized by of the network topology. When an SDN controller receives
varying the length of short-lived (“mice”) flows, the fraction a lightpath establishment request whose computed path tra-
of long-lived (“elephant”) flows, and the volume of traffic con- verses other clusters, the SDN controller requests lightpath
sumed by “elephant” flows. Veisllari et al. [274] discovered, establishment within those clusters via a WBI.
through these visualizations, that the length of “mice” flows The formation of clusters can be performed such that for
is the dominating parameter in this model. a specified number of clusters the average distance to each
Veisllari et al. [274] translated the arrival rate function anal- SDN controller is minimized [276]. The lightpath establish-
ysis to an analysis of the ring MAN network dimensions that ment time decreases exponentially as the number of clusters
can be supported by a single SDN controller. The current state- increases.
of-the-art Beacon controller can handle a flow request every
571 ns. Assuming mice flows sizes of 20 kB and average
circuit lifetimes of 1 second, as the fraction of packet traffic F. Control Layer: Summary and Discussion
increases from 0.1 to 0.9, the network dimension supported A very large body of literature has explored how to expand
by a single Beacon SDN controller decreases from 14 nodes the OpenFlow protocol to support various optical network
with 92 wavelengths per node to 5 nodes with 10 wavelengths technologies (e.g., optical circuit switching, optical packet
per node. switching, passive optical networks). A significant body of
Liu et al. [34] use a multinational (Japan, China, Spain) literature has investigated methodologies for retro-fitting non-
NOX:OpenFlow controlled four-wavelength optical circuit and SDN network elements for OpenFlow control as well as
burst switched network to study path setup/release times as integrating SDN/OpenFlow with the GMPLS/PCE control
well as path restoration times. The optical transponders that framework. A variety of SDN controller use cases have been
can generate failure alarms were also under NOX:OpenFlow identified that motivate the benefits of the centralized network
control and these alarms were used to trigger protec- control made possible with SDN (e.g., bandwidth alloca-
tion switching. The single SDN controller was located in the tion over large numbers of subscribers, controlling tandem
Japanese portion of the network. The experiments found the networks).
path setup time to vary from 250–600 ms and the path release However, analyzing the performance of SDN controllers for
times to vary from 130–450 ms. Path restoration times var- optical network applications is still in a state of infancy. It
ied from 250–500 ms. Liu et al. [34] noted that the major will be important to understand the connection between the
contributing factor to these times was the OpenFlow message implementation of the SDN controller (e.g., processor core
delivery time [34]. architecture, number of threads, operating system) and the net-
3) Out of Band Control: Sanchez et al. [275] have qual- work it can effectively control (e.g., network traffic volume,
itatively compared four SDN controlled ring metropolitan network size) to meet certain performance objectives (e.g.,
network architectures. The architectures vary in whether the maximum flow setup time). At present there are not enough
SDN control traffic is carried in-band with the data traffic relevant studies to gain an understanding of this connection.
or out-of-band separately from the data traffic. In a single With this understanding network service providers will be able
wavelength ring network, out-of-band control would require to partition their networks into control domains in a manner
a separate physical network that would come at a high cost, that meets their performance objectives.
2756 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 18, NO. 4, FOURTH QUARTER 2016

Fig. 10. Classification of SDON virtualization studies.

V. V IRTUALIZATION the static allocation. A similar strategy for flexibly employing

This section surveys control layer mechanisms for vir- different dynamic bandwidth allocation modules for different
tualizing SDONs. As optical infrastructures have typically groups of ONU queues has been examined in [285].
high costs, creating multiple VONs over the optical network Similar OFDMA based slicing strategies for sup-
infrastructure is especially important for access networks, porting cloud computing have been examined by
where the costs need to be amortized over relatively few Jinno and Tsukishima [286]. Zhou et al. [287] have
users. Throughout, accounting for the specific optical trans- explored a FlexPON with similar virtualization capabilities.
mission and signal propagation characteristics is a key chal- The FlexPON employs OFDM for adaptive transmissions.
lenge for SDON virtualization. Following the classification The isolation of different VPONs is mainly achieved through
structure illustrated in Fig. 10, we initially survey virtual- separate MAC processing. The resulting VPONs allow for
ization mechanisms for access networks and data center net- flexible port assignments in ONUs and OLT, which have been
works, followed by virtualization mechanisms for optical core demonstrated in a testbed [287].
networks. 2) FiWi Access Network Virtualization:
a) Virtualized FiWi network: Dai et al. [288], [289] and
A. Access Networks He et al. [290] have examined the virtualization of FiWi
1) OFDMA Based PON Access Network Virtualization: networks [331], [332] to eliminate the differences between
Wei et al. [282]–[284] have developed a link virtualization the heterogeneous segments (fiber and wireless). The vir-
mechanism that can span from optical access to backbone tualization provides a unified homogenous (virtual) view
networks based on Orthogonal Frequency Division Multiple of the FiWi network. The unified network view simpli-
Access (OFDMA). Specifically, for access networks, a Virtual fies flow control and other operational algorithms for traf-
PON (VPON) approach based on multicarrier OFDMA over fic transmissions over the heterogeneous network segments.
WDM has been proposed. Distinct network slices (VPONs) In particular, a virtual resource manager operates the het-
utilize distinct OFDMA subcarriers, which provide a level of erogeneous segments. The resource manager permits mul-
isolation between the VPONs. Thus, different VPONs may tiple routes from a given source node to a given desti-
operate with different MAC standards, e.g., as illustrated in nation node. Load balancing across the multiple paths has
Fig. 11(a), VPON A may operate as an Ethernet PON (EPON) been examined in [291] and [292]. Simulation results indi-
while VPON B operates as a Gigabit PON (GPON). In addi- cate that the virtualized FiWi network with load balancing
tion, virtual MAC queues and processors are isolated to store significantly reduces packet delays compared to a conven-
and process the data from multiple VPONs, thus creating vir- tional FiWi network. An experimental OpenFlow switch
tual MAC protocols, as illustrated in Fig. 11(b). The OFDMA testbed of the virtualized FiWi network has been presented
transmissions and receptions are processed in a DSP mod- in [293]. Testbed measurements demonstrate the seamless
ule that is controlled by a central SDN control module. The networking across the heterogeneous fiber and wireless net-
central SDN control module also controls the different virtual works segments. Measurements for nodal throughput, link
MAC processes in Fig. 11(b), which feed/receive data to/from bandwidth utilization, and packet delay indicate perfor-
the DSP module. Additional bandwidth partitioning between mance improvements due to the virtualized FiWi network-
VPONs can be achieved through Time Division Multiple ing approach. Moreover, the FiWi testbed performance is
Access (TDMA). Simulation studies compared a static allo- measured for a video service scenario indicating that the
cation of subcarriers to VPONs with a dynamic allocation virtualized FiWi networking approach improves the Quality
based on traffic demands. The dynamic allocation achieved of Experience (QoE) [333], [334] of the video streaming.
significantly higher numbers of supported VPONs on a given A mathematical performance model of the virtualized FiWi
network infrastructure as well as lower packet delays than network has been developed in [293].
THYAGATURU et al.: SDONs: A COMPREHENSIVE SURVEY 2757

Fig. 11. Illustration of OFDMA based virtual access network [282]: The different VPONs operate on isolated OFDMA sub carriers allowing different MAC
standards, such as EPON and GPON, to operate on the same physical PON infrastructure, as illustrated in part (a). A central SDN control module controls
the OFDMA transmissions and receptions as well as the virtual MAC processing, see part (b).

b) WiMAX-VPON: WiMAX-VPON [294], [295] is a 2) Cloudnets: Cloudnets [335]–[340] exploit network vir-
Layer-2 Virtual Private Network (VPN) design for FiWi access tualization for pooling resources among distributed data cen-
networks. WiMAX-VPON executes a common MAC proto- ters. Cloudnets support the migration of virtual machines
col across the wireless and fiber network segments. A VPN across networks to achieve resource pooling. Cloudnet
based admission control mechanism in conjunction with a designs can be supported through optical networks [341].
VPN bandwidth allocation ensures per-flow Quality of Service Kantarci and Mouftah [300] have examined designs for a
(QoS). Results from discrete event simulations demonstrate virtual cloud backbone network that interconnects distributed
that the proposed WiMAX-VPON achieves favorable perfor- backbone nodes, whereby each backbone node is associated
mance. Also, Dhaini et al. [294], [295] demonstrate how the with one data center. A network resource manager period-
WiMAX-VPON design can be extended to different access ically executes a virtualization algorithm to accommodate
network types with polling-based wireless and optical medium traffic demands through appropriate resource provisioning.
access control. Kantarci and Mouftah [300] have developed and evaluated
algorithms for three provisioning objectives: minimize the
B. Data Centers outage probability of the cloud, minimize the resource pro-
1) LIGHTNESS: LIGHTNESS [296]–[299] is a European visioning, and minimize a tradeoff between resource saving
research project examining an optical Data Center Network and cloud outage probability. The range of performance char-
(DCN) capable of providing dynamic, programmable, and acteristics for outage probability, resource consumption, and
highly available DCN connectivity services. Whereas conven- delays of the provisioning approaches have been evaluated
tional DCNs have rigid control and management platforms, through simulations. The outage probability of optical cloud
LIGHTNESS strives to introduce flexible control and manage- networks has been reduced in [301] through optimized service
ment through SDN control. The LIGHTNESSS architecture re-locations.
comprises server racks that are interconnected through optical Several complementary aspects of optical cloudnet networks
packet switches, optical circuit switches, and hybrid Top-of- have recently been investigated. A multilayer network archi-
the-Rack (ToR) switches. The server racks and switches are all tecture with an SDN based network management structure
controlled and managed by an SDN controller. LIGHTNESS for cloud services has been developed in [302]. A dynamic
control consists of an SDN controller above the optical phys- variation of the sharing of optical network resources for
ical layer and OpenFlow agents that interact with the optical intra- and inter-data center networking has been examined
network and server elements. The SDN controller in coop- in [303]. The dynamic sharing does not statically assign opti-
eration with the OpenFlow-agents provides a programmable cal network resources to virtual optical networks; instead,
data plane to the virtualization modules. The virtualization cre- the network resources are dynamically assigned according
ates multiple Virtual Data Centers (VDCs), each with its own to the time-varying traffic demands. An SDN based opti-
virtual computing and memory resources, as well as virtual cal transport mode for data center traffic has been explored
networking resources, based on a given physical data center. in [304]. Virtual machine migration mechanisms that take the
The virtualization is achieved through a VDC planner module characteristics of renewable energy into account have been
and an NFV application that directly interact with the SDN examined in [305] while general energy efficiency mechanisms
controller. The VDC planner composes the VDC slices through for optically networked could computing resources have been
mapping of the VDC requests to the physical SDN-controlled examined in [306].
switches and server racks. The VDC slices are monitored by
the NFV application, which interfaces with the VDC plan- C. Metro/Core Networks
ner. Based on monitoring data, the NFV application and VDC 1) Virtual Optical Network Embedding: Virtual opti-
planner may revise the VDC composition, e.g., transition from cal network embedding seeks to map requests for vir-
optical packet switches to optical circuit switches. tual optical networks to a given physical optical network
2758 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 18, NO. 4, FOURTH QUARTER 2016

infrastructure (substrate). A virtual optical network consists Gong and Zhu [311] have considered flexi-grid networks with
of both a set of virtual nodes and a set of interconnect- a similar overall strategy of node mapping followed by link
ing links that need to be mapped to the network substrate. mapping as Zhang et al. [309]. Based on the local resource
This mapping of virtual networks consisting of both network constraints at each node, Gong and Zhu [311] have formed a
nodes and links is fundamentally different from the extensively layered auxiliary graph for the node mapping. The link map-
studied virtual topology design for optical wavelength routed ping is then solved with a shortest path routing approach.
networks [342], which only considered network links (and did Wang et al. [312] have examined an embedding approach
not map nodes). Virtual network embedding of both nodes and based on candidate mapping patterns that could provide the
link has already been extensively studied in general network requested resources. The VON is then embedded according
graphs [72], [343]. However, virtual optical network embed- to a shortest path routing. Pages et al. [298] have considered
ding requires additional constraints to account for the special embeddings that minimize the required optical transponders.
optical transmission characteristics, such as the wavelength c) Survivable embedding: Survivability of a virtual opti-
continuity constraint and the transmission reach constraint. cal network, i.e., its continued operation in the face of physical
Consequently, several studies have begun to examine vir- node or link failures, is important for many applications that
tual network embedding algorithms specifically for optical require dependable service. Hu et al. [313] developed an
networks. embedding that can survive the failure of a single physical
a) Impairment-aware embedding: Peng et al. [307], node. Ye et al. [314] have examined the embedding of virtual
[308] have modeled the optical transmission impairments to optical networks so as to survive the failure of a single physi-
facilitate the embedding of isolated VONs in a given under- cal node or a physical link. Specifically, Ye et al. [314] ensure
lying physical network infrastructure. Specifically, they model that each virtual node request is mapped to a primary physi-
the physical (photonic) layer impairments of both single-line cal node as well as a distinct backup physical node. Similarly,
rate and mixed-line rates [344]. Peng et al. [308] con- each virtual link is mapped to a primary physical route as
sider intra-VON impairments from Amplified Spontaneous well as a node-disjoint backup physical route. Ye et al. [314]
Emission (ASE) and inter-VON impairments from non-linear mathematically formulate an optimization problem for the
impairments and four wave mixing. These impairments are survivable embedding and then propose a Parallel Virtual
captured in a Q-factor [345], [346], which is considered in Infrastructure (VI) Mapping (PAR) algorithm. The PAR algo-
the mapping of virtual links to the underlying physical link rithm finds distinct candidate physical nodes (with the highest
resources, such as wavelengths and wavebands. remaining resources) for each virtual node request. The can-
b) Embedding on WDM and flexi-grid networks: didate physical nodes are then jointly examined with pairs of
Zhang et al. [309] have considered the embedding of overall shortest node-disjoint paths. The evaluations in [314] indicate
virtual networks encompassing both virtual nodes and virtual that the parallel PAR algorithm reduces the blocking proba-
links. Zhang et al. [309] have considered both conventional bilities of virtual network requests by 5–20% compared to a
WDM networks as well as flexi-grid networks. For each net- sequential algorithm benchmark. A limitation of the survivable
work type, they formulate the virtual node and virtual link embedding [314] is that it protects only from a single link or
mapping as a mixed integer linear program. Concluding that node failure. As the optical infrastructure is expected to pen-
the mixed integer linear program is NP-hard, heuristic solution etrate deeper in the access network deployments (e.g., mobile
approaches are developed. Specifically, the overall embedding backhaul), it will become necessary to consider multiple fail-
(mapping) problem is divided into a node mapping problem ure points. Similar survivable network embedding algorithms
and a link mapping problem. The node mapping problem is that employ node-disjoint shortest paths in conjunction with
heuristically solved through a greedy MinMapping strategy specific cost metrics for node mappings have been investigated
that maps the largest computing resource demand to the node by Xie et al. [315] and Chen et al. [316]. Jiang et al. [317]
with the minimum remaining computing capacity (a comple- have examined a solution variant based on maximum-weight
mentary MaxMapping strategy that maps the largest demand maximum clique formation.
to the node with the maximum remaining capacity is also con- The studies [318]–[320] have examined so-called bandwidth
sidered). After the node mapping, the link mapping problem is squeezed restoration for virtual topologies. With bandwidth
solved with an extended grooming graph [347]. Comparisons squeezing, the back-up path bandwidths of the surviving vir-
for a small network indicate that the MinMapping strategy tual topologies are generally lower than the bandwidths on the
approaches the optimal mixed integer linear program solution working paths.
quite closely; whereas the MaxMapping strategy gives poor Survivable virtual topology design in the context of multido-
results. The evaluations also indicate that the flexi-grid net- main optical networks has been studied by Hong et al. [321].
work requires only about half the spectrum compared to an Hong et al. [321] focused on minimizing the total network
equivalent WDM network for several evaluation scenarios. link cost for a given virtual traffic demand. A heuristic algo-
The embedding of virtual optical networks in the con- rithm for partition and contraction mechanisms based on cut
text of elastic flexi-grid optical networking has been fur- set theory has been proposed for the mapping of virtual links
ther examined in several studies. For a flexi-grid network onto multidomain optical networks. A hierarchical SDN con-
based on OFDM [31], Zhao et al. [310] have compared a trol plane is split between local controllers that to manage
greedy heuristic that maps requests in decreasing order of individual domains and a global controller for the overall man-
the required resources with an arbitrary first-fit benchmark. agement. The partition and contraction mechanisms abstract
THYAGATURU et al.: SDONs: A COMPREHENSIVE SURVEY 2759

inter- and intra-domain information as a method of contraction. wavelength channels. The evaluation results in [324] indicate
Survivability conditions are ensured individually for inter- and that the energy optimized embedding significantly reduces the
intra-domains such that survivability is met for the entire net- overall energy consumption for low to moderate loads on
work. The evaluations in [321] demonstrate successful virtual the physical infrastructure; for high loads, when all physical
network mapping at the scale required by commercial Internet resources need to be utilized, there are no significant savings.
service providers and infrastructure providers. Across the entire load range, the energy optimized embed-
d) Dynamic embedding: The embedding approaches sur- ding saves on average 20% energy compared to the benchmark
veyed so far have mainly focused on the offline embedding of a minimizing the wavelength channels.
static set of virtual network requests. However, in the ongoing Chen [325] has examined a similar energy-efficient virtual
network operation the dynamic embedding of modifications optical network embedding that considers primary and link-
(upgrades) of existing virtual networks, or the addition of new disjoint backup paths, similar to the survivable embeddings
virtual networks are important. Ye et al. [322] have examined in Section V-C1c. More specifically, virtual link requests are
a variety of strategies for upgrading existing virtual topologies. mapped in decreasing order of their bandwidth requirements
Ye et al. [322] have considered both scenarios without advance to the shortest physical transmission distance paths, i.e., the
planning (knowledge) of virtual network upgrades and scenar- highest virtual bandwidth demands are allocated to the short-
ios that plan ahead for possible (anticipated) upgrades. For est physical paths. Evaluations indicate that this link mapping
both scenarios, a divide-and-conquer strategy and an integrate- approach roughly halves the power consumption compared to
and-cooperate strategy are examined. The divide-and-conquer a random node mapping benchmark. Further studies focused
strategy sequentially maps all the virtual nodes and then the on energy savings have examined virtual link embeddings that
virtual links. In contrast, the integrate-and-cooperate strategy maximize the usage of nodes with renewable energy [326] and
jointly considers the virtual node and virtual link mappings. the traffic grooming [327] onto sliceable BVTs [349].
Without advance planning, these strategies are applied sequen- 2) Hypervisors for VONs: The operation of VONs over a
tially, as the virtual network requests arrive over time, whereas, given underlying physical (substrate) optical network requires
with planning, the initial and upgrade requests are jointly an intermediate hypervisor. The hypervisor presents the phys-
considered. Evaluation results indicate that the integrate-and- ical network as multiple isolated VONs to the corresponding
cooperate strategy slightly increases a revenue measure and VON controllers (with typically one VON controller per
request acceptance ratio compared to the divide-and-conquer VON). In turn, the hypervisor intercepts the control messages
strategy. The results also indicate that planning has the poten- issued by a VON controller and controls the physical network
tial to substantially increase the revenue and acceptance to effect the control actions desired by the VON controller for
ratio. In a related study, Zhang et al. [323] have examined the corresponding VON.
embedding algorithms for virtual network requests that arrive Towards the development of an optical network hypervisor,
dynamically to a multilayer network consisting of electrical de Siquera et al. [328] have developed a SDN-based controller
and optical network substrates. for an optical transport architecture. The controller implements
e) Energy-efficient embedding: Motivated by the grow- a virtualized GMPLS control plane with offloading to facilitate
ing importance of green networking and information tech- the implementation of hypervisor functionalities, namely the
nology [348], a few studies have begun to consider the creation optical virtual private networks, optical network slic-
energy efficiency of the embedded virtual optical networks. ing, and optical interface management. A major contribution
Nonde et al. [324] have developed and evaluated mecha- of de Siquera et al. [328] is a Transport Network Operating
nisms for embedding virtual cloud networks so as to minimize System (T-NOS), which abstracts the physical layer for the
the overall power consumption, i.e., the aggregate of the controller and could be utilized for hypervisor functionalities.
power consumption for communication and computing (in OpenSlice [329] is a comprehensive OpenFlow-based
the data centers). Nonde et al. [324] have incorporated the hypervisor that creates VONs over underlying elastic optical
power consumption of the communication components, such networks [90], [91]. OpenSlice dynamically provisions end-to-
as transponders and optical switches, as well as the power end paths and offloads IP traffic by slicing the optical commu-
consumption characteristics of data center servers into a math- nications spectrum. The paths are set up through a handshake
ematical power minimization model. Nonde et al. [324] then protocol that fills in cross-connection table entries. The con-
develop a real-time heuristic for energy-optimized virtual trol messages for slicing the optical communications spectrum,
network embedding. The heuristic strives to consolidate com- such as slot width and modulation format, are carried in
puting requests in the physical nodes with the least residual extended OpenFlow protocol messages. OpenSlice relies on
computing capacity. This consolidation strategy is motivated special distributed network elements, namely bandwidth vari-
by the typical power consumption characteristic of a compute able wavelength cross-connects [350] and multiflow optical
server that has a significant idle power consumption and then transponders [153] that have been extended for control through
grows linearly with increasing computing load; thus a fully the extended OpenFlow messages. The OpenSlice evaluation
loaded server is more energy-efficient than a lightly loaded includes an experimental demonstration. The evaluation results
server. The bandwidth demands are then routed between the include path provisioning latency comparisons with a GMPLS-
nodes according to a minimum hop algorithm. The energy based control plane and indicate that OpenFlow outperforms
optimized embedding is compared with a cost optimized GMPLS for paths with more than three hops. OpenSlice exten-
embedding that only seeks to minimize the number of utilized sion and refinements to multilayer and multidomain networks
2760 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 18, NO. 4, FOURTH QUARTER 2016

are surveyed in Section VII. An alternate centralized Optical well as low-complexity hypervisor deployment, operation, and
FlowVisor that does not require extensions to the distributed maintenance.
network elements has been investigated in [330].
VI. SDN A PPLICATION L AYER
D. Virtualization: Summary and Discussion
In the SDN paradigm, applications interact with the con-
The virtualization studies on access networks [282]–[284],
trollers to implement network services. We organize the survey
[286]–[295] have primarily focused on exploiting and manipu-
of the studies on application layer aspects of SDONs according
lating the specific properties of the optical physical layer (e.g.,
to the main application categories of quality of service (QoS),
different OFDMA subcarriers) and MAC layer (e.g., polling
access control and security, energy efficiency, and failure
based MAC protocol) of the optical access networks for vir-
recovery, as illustrated in Fig. 12.
tualization. In addition, to virtualization studies on purely
optical PON access networks, two sets of studies, namely
sets [288]–[293] and WiMAX-VPON [294], [295] have exam- A. QoS
ined virtualization for two forms of FiWi access networks. 1) Long-Term QoS (Time-Aware SDN): Data Center (DC)
Future research needs to consider virtualization of a wider networks move data back and forth between DCs to bal-
set of FiWi network technologies, i.e., FiWi networks that ance the computing load and the data storage usage (for
consider optical access networks with a wider variety of wire- upload) [395]. These data movements between DCs can span
less access technologies, such as different forms of cellular large geographical areas and help ensure DC service QoS
access or combinations of cellular with other forms of wireless for the end users. Load balancing algorithms can exploit the
access. Also, virtualization of integrated access and metropoli- characteristics of the user requests. One such request char-
tan area networks [351]–[354] is an important future research acteristic is the high degree of time-correlation over various
direction. time scales ranging from several hours of a day (e.g., due
A set of studies has begun to explore optical networking to a sporting event) to several days in a year (e.g., due to a
support for SDN-enabled cloudnets that exploit virtualization political event). Zhao et al. [361] have proposed a time-aware
to dynamically pool resources across distributed data cen- SDN application using OpenFlow extensions to dynamically
ters. One important direction for future work on cloudnets balance the load across the DC resources so as to improve
is to examine moving data center resources closer to the the QoS. Specifically, a time correlated PCE algorithm based
users and the subsequent resource pooling across edge net- on flexi-grid optical transport (see Section IV-D2) has been
works [355]. Also, the exploration of the benefits of FiWi proposed. An SDN application monitors the DC resources
networks for decentralized cloudlets [356]–[359] that sup- and applies network rules to preserve the QoS. Evaluations
port mobile wireless network services is an important future of the algorithm indicate improvements in terms of network
research direction [360]. blocking probability, global blocking probability, and spec-
A fairly extensive set of studies has examined virtual net- trum consumption ratio. This study did not consider short
work embedding for metro/core networks. The virtual network time scale traffic bursts, which can significantly affect the load
embedding studies have considered the specific limitations and conditions.
constraints of optical networks and have begun to explore spe- We believe that in order to avoid pitfalls in the operation
cialized embedding strategies that strive to meet a specific of load balancing through PCE algorithms implemented with
optimization objective, such as survivability, dynamic adapt- SDN, a wide range of traffic conditions needs to be consid-
ability, or energy efficiency. Future research should seek to ered. The considered traffic range should include short and
develop a comprehensive framework of embedding algorithms long term traffic variations, which should be traded off with
that can be tuned with weights to achieve prescribed degrees various QoS aspects, such as type of application and delay con-
of the different optimization objectives. straints, as well as the resulting costs and control overheads.
A relatively smaller set of studies has developed and refined Khodakarami et al. [362] have taken steps in this direction by
hypervisors for creating VONs over metro/core optical net- forming a traffic forecasting model for both long-term and
works. Much of the SDON hypervisor research has centered short-term forecasts in a wide-area mesh network. Optical
on the OpenSlice hypervisor concept [329]. While OpenSlice lightpaths are then configured based on the overall traffic fore-
accounts for the specific characteristics of the optical transmis- cast, while electronic switching capacities are allocated based
sion medium, it is relatively complex as it requires a distributed on short-term forecasts.
implementation with specialized optical networking compo- 2) Short Term QoS: Users of a high-speed FTTH access
nents. Future research should seek to achieve the hypervisor network may request very large bandwidths due to simulta-
functionalities with a wider set of common optical components neously running applications that require high data rates. In
so as to reduce cost and complexity. Overall, SDON hypervi- such a scenario, applications requiring very high data rates
sor research should examine the performance-complexity/cost may affect each other. For instance, a video conference running
tradeoffs of distributed versus centralized approaches. Within simultaneously with the streaming of a sports video may result
this context of examining the spectrum of distributed to cen- in call drops in the video conference application and in stalls of
tralized hypervisors, future hypervisor research should further the sports video. Li et al. [363] proposed an SDN based band-
refine and optimize the virtualization mechanisms so as to width provisioning application in the broadband remote access
achieve strict isolation between virtual network slices, as server [396] network. They defined and assigned the minimum
THYAGATURU et al.: SDONs: A COMPREHENSIVE SURVEY 2761

Fig. 12. Classification of application layer SDON studies.

The algorithm considers the localized traffic information and

optical resource availability at the nodes. The algorithm does
not require synchronization, thus reducing the overhead while
simplifying the network design. In the proposed architecture,
optical switches are connected to ROADMs. The reconfigu-
ration application manages and controls the optical switches
through the SDN controller. A new WDM controller is intro-
duced to configure the lightpaths taking wavelength conversion
and lightpath switching at the ROADMs into consideration.
The SDN controller operates on the optical network which
Fig. 13. Optical SDN-based QoS-aware burst switching application [364]: appears as a static network, while the WDM controller con-
Based on short-term traffic burst estimates of an edge node, the SDN controller figures (and re-configures) the ROADMs to create multiple vir-
configures end-to-end light paths ensuring QoS. tual optical networks according to the traffic levels. Evaluation
results indicate improved utilization and throughput. The
results indicate that virtual topologies reconfigurations can
bandwidth, which they named “sweet point”, required for each significantly increase the flexibility of the network while
application to experience good QoE. Li et al. [363] showed achieving the desired QoS. However, the control overhead and
that maintaining the “sweet point” bandwidth for each applica- the delay aspects due to virtualization and separation of control
tion can significantly improve the QoE while other applications and lightwave paths needs to be carefully considered.
are being served according to their bandwidth requirements. 4) End-to-End QoS Routing: Interconnections between
In a similar study, Patel et al. [364] proposed a burst DCs involve typically multiple data paths. All the inter-
switching mechanism based on a software defined optical faces existing between DCs can be utilized by MultiPath
network. Bursts typically originate at the edge nodes and TCP (MPTCP). Ensuring QoS in such an MPTCP setting
the aggregation points due to statistical multiplexing of high while preserving throughput efficiency in a reconfigurable
speed optical transmissions. To ensure QoS for multiple traf- underlying burst switching optical network is a challenging
fic classes, bursts at the edge nodes have to be managed by task. Tariq and Bassiouni [366] have proposed QoS-aware
deciding their end-to-end path to meet their QoS require- bandwidth reservation for MPTCP in an SDON. The band-
ments, such as minimum delay and data rate. In non-SDN width reservation proceeds in two stages (i) path selection for
based mechanisms, complicated distributed protocols, such as MPTCP, and (ii) OBS wavelength reservation to assign the
GMPLS [278], [280], are used to route the burst traffic. In priorities for latency-sensitive flows. Larger portions of a
the proposed application, the centralized unified control plane wavelength reservation are assigned to high priority flows,
decides the routing path for the burst based on latency and resulting in reduced burst blocking probability while achieving
QoS requirements. A simplified procedure involves (i) burst the higher MPTCP throughput. The simulation results in [366]
evaluation at the edge node, (ii) reporting burst information validate the two-stage algorithm for QoS-aware MPTCP over
to the SDN controller, and (iii) sending of configurations an SDON, indicating decreased dropping probabilities, and
to the optical nodes by the controller to set up a lightpath increased throughputs.
as illustrated in Fig. 13. Simulations indicate an increase of Information To the Routing System (I2RS) [397] is a high-
performance in terms of throughput, network blocking proba- level architecture for communicating and interacting with
bility, and latency along with improved QoS when compared routing systems, such as BGP routers. A routing system may
to non-SDN GMPLS methods. consists of several complex functional entities, such as a
3) Virtual Topology Reconfigurations: The QoS experi- Routing Information Base (RIB), an RIB manager, topology
enced by traffic flows greatly depends on their route through a and policy databases, along with routing and signalling units.
network. Wette and Karl [365] have examined an application The I2RS provides a programmability platform that enables
algorithm that reconfigures WDM network virtual topolo- access and modifications of the configurations of the rout-
gies (see Section V-C1b) according to the traffic levels. ing system elements. The I2RS can be extended with SDN
2762 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 18, NO. 4, FOURTH QUARTER 2016

Fig. 15. SDN based video caching application in PON for mobile users [374]:
The SDN controller pushes frequently requested videos to base station (BS)
caches, whereby multicast can reach several BS caches.

wired technologies, such as DSL, and wireless technologies,

Fig. 14. Illustration of routing application with integrated control of such as LTE and WiFi. SDN controllers managed by the ISPs
access, metro, and core networks using SDN and the Information To the optimize the traffic flows to each user while accommodat-
Routing System (I2RS) [367]: The SDN controller interacts with the access
network, e.g., through the OpenFlow protocol, the metro network, e.g., ing large numbers of users and ensuring their minimum QoS.
through the I2RS, and the core network, e.g., through the Path Computation Additionally, Tego et al. [371] demonstrated an experimental
Elements (PCEs). SDN based QoS management setup to optimize the energy
utilization. GbE links are switched on and off based on the
traffic levels. The QoS management reroutes the traffic to
principles to achieve global network management and recon- avoid congestion and achieve efficient throughput. SDN appli-
figuration [398]. Sgambelluri et al. [367] presented an SDN cations conduct active QoS probing to monitor the network
based routing application within the I2RS framework to inte- QoS characteristics. Evaluations have indicated that the SDN
grate the control of the access, metro, and core networks as based techniques achieve significantly higher throughput than
illustrated in Fig. 14. The SDN controller communicates with non-SDN techniques [371].
the Path Computation Elements (PCEs) of the core network to 6) Video Applications: The application-aware SDN-
create Label Switched Paths (LSPs) based on the information enabled resource allocation application has been introduced
received by the OLTs. Experimental demonstrations validated by Chitimalla et al. [372] to improve the video QoE in a
the routing optimization based on the current traffic status and PON access network. The resource allocation application uses
previous load as well as the unified control interface for access, application level feedback to schedule the optical resources.
metro, and core networks. The video resolution is incrementally increased or decreased
Ilchmann et al. [368] developed an SDN application that based on the buffer utilization statistics that the client sends
communicates to an SDN controller via an HTTP-based REST to the controller. The scheduler at the OLT schedules the
API. Over time, lightpaths in an optical network can become packets based on weights calculated by the SDN controller,
inefficient for a number of reasons (e.g., optical spectrum frag- whereby the video applications at the clients communicate
mentation). For this reason, Ilchmann et al. [368] developed with the controller to determine the weights. If the network
an SDN application that evaluates existing lightpaths in an is congested, then the SDN controller communicates to the
optical network and offers an application user the option to re- clients to reduce the video resolution so as to reduce the
optimize the lightpath routing to improve various performance stalls and to improve the QoE.
metrics (e.g., path length). The application is user-interactive Caching of video data close the users is generally benefi-
in that the user can see the number of proposed lightpath cial for improving the QoE of video services [399], [400].
routing changes before they are made and can potentially Li et al. [374] have introduced caching mechanisms for
select a subset of the proposed changes to minimize network software-defined PONs. In particular, Li et al. [374] have
down-time. proposed joint provisioning of the bandwidth to service the
At the ingress and egress routers of optical networks (e.g., video and the cache management, as illustrated in Fig. 15.
the edge routers between access and metro networks), buffers Based on the request frequency for specific video content, the
are highly non-economical to implement, as they require large Base Station (BS) caches the content with the assistance of the
buffers sizes to accommodate the channel rates of 40 Mb/s or SDN controller. The proposed push-based mechanism deliv-
more. To reduce the buffer requirements at the edge routers, ers (pushes) the video to the BS caches when the PON is not
Chang et al. [369] have proposed a backpressure applica- congested. A specific PON transmission sub-band can be used
tion referred to as Refill and SDN-based Random Early to multicast video content that needs to be cached at multi-
Detection (RS-RED). RS-RED implements a refill queue at ple BSs. The simulation evaluation in [374] indicate that up
the ingress device and a droptail queue at the egress device, to 30% additional videos can be serviced while the service
whereby both queues are centrally managed by the RS-RED response delay is reduced to 50%.
algorithm running on the SDN controller. Simulation results
showed that at the expense of small delay increases, edge
router buffer sizes can be significantly reduced. B. Access Control and Security
5) QoS Management: Rückert et al. [370] proposed SDN 1) Flow-Based Access Control: Network Access Control
based controlled home-gateway supporting heterogeneous (NAC) is a networking application that regulates the access to
THYAGATURU et al.: SDONs: A COMPREHENSIVE SURVEY 2763

period, an SDN based high precision time synchronization has

been proposed. As a result, a fast hopping mechanism can be
implemented and executed in a coordinated manner. A hop
frame is defined and guard periods are added in between hop
frames. The experimental evaluations indicate that a maximum
hopping frequency of 1 MHz can be achieved with a BER of
1 × 10−3 . However, shortcomings of such mechanisms are
the secure exchange of hopping sequences between the trans-
mitter and the receiver. Although, centralized SDN control
provides authenticated provisioning of the hopping sequence,
additional mechanisms to secure the hopping sequence from
being obtained through man-in-the-middle attacks should be
Fig. 16. Overview of optical light path hopping mechanism to secure link
from eavesdropping and jamming [377]: The flow marked by the diagonal investigated.
shading hops from lightpath channel λ4 to λ2 , then to λ3 and on to λ1 . 3) Flow Timeout: SDN flow actions on the forwarding and
Transmissions by distinct flows on a given lightpath channel must be separated
by at least a guard period.
switching elements have generally a validity period. Upon
expiration of the validity period, i.e., the flow action time-
out, the forwarding or switching element drops the flow action
network services [255], [401]. A NAC based on traffic flows from the forwarding information base or the flow table. The
has been developed by Matias et al. [375]. FlowNAC exploits switching element CPU must be able to access the flow action
the forwarding rules of OpenFlow switches, which are set by information with very low latency so as to perform switch-
a central SDN controller, to control the access of traffic flows ing actions at the line rate. Therefore, the flow actions are
to network services. FlowNAC can implement the access con- commonly stored in Ternary Content Addressable Memories
trol based on various flow identifiers, such as MAC addresses (TCAMs) [403], which are limited to storing on the order
or IP source and destination addresses. Performance evalua- of thousands of distinct entries. In SDONs, the optical net-
tions measured the connection times for flows on a testbed work elements perform the actions set by the SDN controller.
and found average connection times on the order of 100 ms These actions have to be stored in a finite memory space.
for completing the flow access control. Therefore, it is important to utilize the finite memory space
In a related study, Nayak et al. [376] developed the as efficiently as possible [404]–[408]. In the dynamic time-
Resonance flow based access control system for an enterprise out approach [378], the SDN controller tracks the TCAM
network. In the Resonance system, the network elements, such occupancy levels in the switches and adjusts timeout dura-
as the routers themselves, dynamically enforce access control tions accordingly. However, a shortcoming of such techniques
policies. The access control policies are implemented through is that the bookkeeping processes at the SDN controllers
real-time alerts and flow based information that is exchanged can become cumbersome for a large network. Therefore,
with SDN principles. Nayak et al. [376] have demonstrated autonomous timeout management techniques that are imple-
the Resonance system on a production network at Georgia mented at the hypervisors can reduce the controller processing
Tech. The Resonance design can be readily implemented in load and are an important future research direction.
SDON networks and can be readily extended to wide area
networks. Consider for example multiple heterogeneous DCs
of multiple organizations that are connected by an optical C. Energy Efficiency
backbone network. The Resonance system can be extended The separation of the control plane from the data plane and
to provide access control mechanisms, such as authentication the global network perspective are unique advantages of SDN
and authorization, through such a wide area SDON. for improving the energy efficiency of networks, which is an
2) Lightpath Hopping Security: The broad network important goal [409], [410].
perspective of SDN controllers facilitates the implementa- 1) Power-Saving Application Controller: Ji et al. [379]
tion of security functions that require this broad perspec- have proposed an all optical energy-efficient network centered
tive [28], [29], [402]. However, SDN may also be vulnerable around an application controller [380], [381] that monitors
to a wide range of attacks and vulnerabilities, including power consumption characteristics and enforces power savings
unauthorized access, data leakage, data modification, and mis- policies. Ji et al. [379] first introduce energy-efficient varia-
configuration. Eavesdropping and jamming are security threats tions of Digital-to-Analog Converters (DACs) and wavelength
on the physical layer and are especially relevant for the opti- selective ROADMs as components for their energy-efficient
cal layer of SDONs. In order to prevent eavesdropping and network. Second, Jie et al. [379] introduce an energy-efficient
jamming in an optical lightpath, Li et al. [377] have pro- switch architecture that consists of multiple parallel switching
posed an SDN based fast lightpath hopping mechanism. As planes, whereby each plane consists of three stages with opti-
illustrated in Fig. 16, the hopping mechanism operates over cal burst switching employed in the second (central) switching
multiple lightpath channels. Conventional optical lightpath stage. Third, Jie et al. [379] detail a multilevel SDN based
setup times range from several hundreds of milliseconds to control architecture for the network built from the introduced
several seconds and would result in a very low hopping fre- components and switch. The control structure accommodates
quency. To avoid the optical setup times during each hopping multiple networks domains, whereby each network domain can
2764 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 18, NO. 4, FOURTH QUARTER 2016

involve multiple switching technologies, such as time-based

and frequency-based optical switching. All controllers for the
various domains and technologies are placed under the con-
trol of an application controller. Dedicated power monitors
that are distributed throughout the network update the SDN
based application controller about the energy consumption
characteristics of each network node. Based on the received
energy consumption updates, the application controller exe-
cutes power-saving strategies. The resulting control actions
are signalled by the application controller to the various con-
trollers for the different network domains and technologies.
Fig. 17. Illustration of application layer modules of SDN based network
An extension of this multi-level architecture to cloud-based reprovisioning framework for disaster aware networking [387].
radio access networks has been examined in [382].
2) Energy-Saving Routing: Tego et al. [383] have proposed
an energy-saving application that switches off under-utilized Network resource reprovisioning is a process to change the
GbE network links. Specifically, Tego et al. [383] proposed network configurations, e.g., the network topology and routes,
two methods: Fixed Upper Fixed Lower (FUFL) and Dynamic to recover from failures. A Backup Reprovisioning with Path
Upper and Fixed Lower (DLFU). In FUFL, the IP routing Protection (BRPP), based on SDN for optical networks has
and the connectivity of the logical topology are fixed. The been presented by Savas et al. [387]. An SDN application
utilization of physical GbE links (whereby multiple parallel framework as illustrated in Fig. 17 was designed to support
physical links form a logical link) is compared with a threshold the reprovisioning with services, such as provisioning the new
to determine whether to switch off or on individual physical connections, risk assessment, as well as service level and
links (that support a given logical link). The traffic on a phys- backup management. When new requests are received by the
ical link that is about to be switched off is rerouted on a BRPP application framework, the statistics module evaluates
parallel physical GbE link (within the same logical link). In the network state to find the primary path and a link-disjoint
contrast, in the DLFU approach, the energy saving applica- backup path. The computed backup paths are stored as logi-
tion monitors the load levels on the virtual links. If the load cal links without being provisioned on the physical network.
level on a given virtual link falls below a threshold value, The logical backup module manages and recalculates the log-
then the virtual link topology is reconfigured to eliminate the ical links when a new backup path cannot be accommodated
virtual link with the low load. A general pitfall of such link or to optimize the existing backup paths (e.g., minimize the
switch-off techniques is that energy savings may be achieved backup path distance). Savas et al. [387] introduce a degraded
at the expense of deteriorating QoS. The QoS should therefore backup path mechanism that reserves not the full, but a lower
be closely monitored when switching off links and re-routing (degraded) transmission capacity on the backup paths, so as
flows. to accommodate more requests. Emulations of the proposed
A similar SDN based routing strategy that strives to save mechanisms indicate improved network utilization while effec-
energy while preserving the QoS has been examined in the tively provisioning the backup paths for restoring the network
context of a GMPLS optical networks in [384]. Multipath after network failures.
routing optimizing applications that strive to save energy in As a part of DARPA’s core optical networks CORONET
an SDN based transport optical network have been presented project, a non-SDN based Robust Optical Layer End-to-end
in [385]. A similar SDN based optimization approach for X-connection (ROLEX) protocol has been demonstrated and
reducing the energy consumption in data centers has been presented along with the lessons learned [411]. ROLEX is
examined by Yoon and Kamal [386]. Yoon and Kamal [386] a distributed protocol for failure recovery which requires a
formulated a mixed integer linear program that models the considerable amount of signaling between nodes for the dis-
switches and hosts as queues. Essentially, the optimization tributed management. Therefore to avoid the pitfall of exces-
decides on the switches and hosts that could be turned off. sive signalling, it may be worthwhile to examine a ROLEX
As the problem is NP-hard, annealing algorithms are exam- version with centralized SDN control in future research to
ined. Simulations indicate that energy savings of more than reduce the recovery time and signaling overhead, as well as
80% are possible for low data center utilization rates, while the costs of restored paths while ensuring the user QoS.
the energy savings decrease to less than 40% for high data 2) Restoration Processing: During a restoration, the net-
center utilization rates. Traffic balancing in the metro optical work control plane simultaneously triggers backup provision-
access networks through the SDN based reconfiguration of ing of all disrupted paths. In GMPLS restoration, along with
optical subscriber units in a TWDM-PON systems for energy signal flooding, there can be contention of signal messages
savings has been additionally demonstrated in [353]. at the network nodes. Contentions may arise due to spectrum
conflicts of the lightpath, or node-configuration overrides, i.e.,
a new configuration request arrives while a preceding recon-
D. Failure Recovery and Restoration figuration is under way. Giorgetti et al. [388] have proposed
1) Network Reprovisioning: Network disruptions can dynamic restoration in the elastic optical network to avoid
occur due to various natural and/or man-made factors. signaling contention in SDN (i.e., of OpenFlow messages).
THYAGATURU et al.: SDONs: A COMPREHENSIVE SURVEY 2765

to physical breaks in the long feeder fibers. Along with the

high impact of connectivity break down or degraded service,
physical restoration time can be very long. Therefore, 1:1 pro-
tection for LR-PONs based on SDN has been proposed, where
primary and secondary (backup) OLTs are used without traffic
duplication. More specifically, Slyne et al. [391] have devised
and demonstrated an OpenFlow-Relay located at the switching
unit. The OpenFlow-Relay detects and reports a failure along
with fast updating of forwarding rules. Experimental demon-
stration show the backup OLT carrying protected traffic within
Fig. 18. Illustration of Application-Based Network Operation (ABNO)
architecture: The ABNO controller communicates with the Operation,
7.2 ms after a failure event.
Administration, and Maintenance (OAM) module, the Path Computation An experimental demonstration utilizing multiple paths in
Element (PCE) module as well as the topology modules and the provisioning optical transport networks for failure recovery has been dis-
manager to control the lower domain SDN controllers so as to recover from
network failures [389].
cussed by Kim et al. [392]. Kim et al. [392] have used
commercial grade IP WDM network equipment and imple-
mented multipath TCP in an SDN framework to emulate
Two SDN restoration mechanisms were presented: (i) the inde- inter-DC communication. They developed an SDN applica-
pendent restoration scheme (SDN-ind), and (ii) the bundle tion, consisting of an cross-layer service manager module and
restoration scheme (SDN-bund). In SDN-ind, the controller a cross-layer multipath transport module to reconfigure the
triggers simultaneous independent flow modification (Flow- optical paths for the recovery from connection impairments.
Mod) messages for each backup path to the switches involved Their evaluations show increased bandwidth utilization and
in the reconfigurations. During contention, switches enqueue reduced cost while being resilient to network impairments as
the multiple received Flow-Mod messages and process them the cross-layer multipath transport module does not reserve
sequentially. Although SDN-ind achieves reduced recovery the backup path on the transport network.
time as compared to non-SDN GMPLS, the waiting of mes- 4) Hierarchical Survivability: Networks can be made sur-
sages in the queue incurs a delay. In SDN-bund, the backup vivable by introducing resource redundancy. However, the
path reconfigurations are bundled into a single message, i.e., a cost of the network increases with increased redundancy.
Bundle Flow-Mod message, and sent to each involved switch. Zhang et al. [393] have demonstrated a highly survivable
Each switch then configures the flow modifications in one IP-Optical multilayered transport network. Hierarchal con-
reconfiguration, eliminating the delay incurred by the queuing trollers are placed for multilayer resource provisioning. Optical
of Flow-Mod messages. A similar OpenFlow enabled restora- nodes are controlled by Transport Controllers (TCs), while
tion in Elastic Optical Networks (EONs) has been studied higher layers (IP) are controlled by unified controllers (UCs).
in [412]. The UCs communicate with the TCs to optimize the routes
3) Reconfiguration: Aguado et al. [389] have demon- based on cross-layer information. If a fiber causes a ser-
strated a failure recovery mechanism as part of the EU FP7 vice disruption, TCs may directly set up alternate routes or
STRAUSS project with dynamic virtual reconfigurations using ask the UCs for optimized routes. A pitfall of such hier-
SDN. They considered multidomain hypervisors and domain- archical control techniques can be long restoration times.
specific controllers to virtualize the multidomain networks. However, the cross layer restorations can recover from
The Application-Based Network Operations (ABNO) frame- high degrees of failures, such as multipoint and concurrent
work illustrated in Fig. 18 enables network automation and failures.
programmability. ABNO can compute end-to-end optical paths 5) Robust Power Grid: The lack of a reliable communi-
and delegate the configurations to lower layer domain SDN cation infrastructure for power grid management was one the
controllers. Requirements for fast recovery from network fail- many reasons for the widespread blackout in the Northeastern
ures would be in the order of tens of milliseconds, which is U.S.A. in the year 2003, which affected the lives of 50 million
challenging to achieve in large scale networks. ABNO reduces people [414]. Since then building a reliable communication
the recovery times by pre-computing the backup connections infrastructure for the power grid has become an important
after the first failure, while the Operation, Administration and priority. Rastegarfar and Kilper [394] have proposed a com-
Maintenance (OAM) module [413] communicates with the munication infrastructure that is focused on monitoring and
ABNO controller to configure the new end-to-end connections can react to and recover from failures so as to reliably support
in response to a failure alarm. Failure alarms are triggered by power grid applications. More specifically, their architecture
the domain SDN controllers monitoring the traffic via the opti- was built on SDN based optical networking for implementing
cal power meters when power is below −20 dBm. In order robust power grid control applications. Control and infrastruc-
to ensure survivability, an adaptive survivability scheme that ture in the SDN based power grid management exhibits an
takes routing as well as spectrum assignment and modulation interdependency i.e., the physical fiber relies on the control
into consideration has been explored in [390]. plane for its operations and the logical control plane relies
A similar design for end-to-end protection and failure on the same physical fiber for its signalling communications.
recovery has been demonstrated by Slyne et al. [391] for a Therefore, they only focus on optical protection switching
long-reach (LR) PON. LR-PON failures are highly likely due instead of IP layer protection, for the resilience of the SDN
2766 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 18, NO. 4, FOURTH QUARTER 2016

control. Cascaded failure mechanisms were modeled and sim- (see [387], [389], [392], [393]) for improved end-to-end pro-
ulated for two geographical topologies (U.S. and E.U.). In tection and recovery.
addition, the impacts of cascaded failures were studied for two
scenarios (i) static optical layer (static OL), and (ii) dynamic
optical layer (dynamic OL). Results for a static OL illus- VII. O RCHESTRATION
trated that the failure cascades are persistent and are closely As introduced in Section II-A4, orchestration accomplishes
dependent on the network topology. However, for a dynamic higher layer abstract coordination of network services and
OL (i.e., with reconfiguration of the physical layer), failure operations. In the context of SDONs, orchestration has mainly
cascades were suppressed by an average of 73%. been studied in support of multilayer networking. Multilayer
networking in the context of SDN and network virtualiza-
E. Application Layer: Summary and Discussion tion generally refers to networking across multiple network
The SDON QoS application studies have mainly examined layers and their respective technologies, such as IP, MPLS,
traffic and network management mechanisms that are sup- and WDM, in combination with networking across multiple
ported through the OpenFlow protocol and the central SDN routing domains [74], [415]–[418]. The concept of multilayer
controller. The studied SDON QoS applications are struc- networking is generally an abstraction of providing network
turally very similar in that the traffic conditions or network services with multiple networking layers (technologies) and
states (e.g., congestion levels) are probed or monitored by multiple routing domains. The different network layers and
the central SDN controller. The centralized knowledge of their technologies are sometimes classified into Layer 0 (e.g.,
the traffic and network is then utilized to allocate or con- fiber-switch capable), Layer 1 (e.g., lambda switching capa-
figure resources, such as DC resources in [361], application ble), Layer 1.5 (e.g., TDM SONET/SDH), Layer 2 (e.g.,
bandwidths in [363], and topology configurations or routes Ethernet), Layer 2.5 (e.g., packet switching capable using
in [365]–[367], and [369]. Future research on SDON QoS MPLS), and Layer 3 (e.g., packet switching capable using IP
needs to further optimize the interactions of the controller with routing) [419]. Routing domains are also commonly referred
the network applications and data plane to quickly and cor- to as network domains, routing areas, or levels [415].
rectly react to changing user demands and network conditions, The recent multilayer networking review article [415] has
so as to assure consistent QoS. The specific characteristics and introduced a range of capability planes to represent the
requirements of video streaming applications have been con- grouping of related functionalities for a given networking
sidered in the few studies on video QoS [372]–[374]. Future technology. The capability planes include the data plane for
SDON QoS research should consider a wider range of specific transmitting and switching data. The control plane and the
prominent application traffic types with specific characteristics management plane directly interact with the data plane for
and requirements, e.g., Voice over IP (VoIP) traffic has rela- controlling and provisioning data plane services as well as for
tively low bit rate requirements, but requires low end-to-end trouble shooting and monitoring the data plane. Furthermore,
latency. an authentication and authorization plane, a service plane,
Very few studies have considered security and access con- and an application plane have been introduced for providing
trol for SDONs. The thorough study of the broad topic area network services to users.
of security and privacy is an important future research direc- Multilayer networking can involve vertical layering or hor-
tion in SDONs, as outlined in Section VIII-C Energy efficiency izontal layering [415], as illustrated in Fig. 19. In vertical
is similarly a highly important topic within the SDON research layering, a given layer, e.g., the routing layer, which may
area that has received relatively little attention so far and employ a particular technology, e.g., the Internet Protocol (IP),
presents overarching research challenges, see Section VIII-I. uses another (underlying) layer, e.g., the Wavelength Division
One common theme of the SDON application layer stud- Multiplexing (WDM) circuit switching layer, to provide ser-
ies focused on failure recovery and restoration has been to vices to higher layers. In horizontal layering, services are
exploit the global perspective of the SDN control. The global provided by “stitching” together a service path across multiple
perspective has been exploited for for improved planning of routing domains.
the recovery and restoration [387], [389], [393] as well as SDN provides a convenient control framework for these
for improved coordination of the execution of the restoration flexible multilayer networks [415]. Several research net-
processes [388], [412]. Generally, the existing failure recov- works, such as ESnet, Internet2, GEANT, Science DMZ
ery and restoration studies have focused on network (routing) (Demilitarized Zone) have experimented with these multilayer
domain that is owned by a particular organizational entity. networking concepts [420], [421]. In particular, SDN based
Future research should seek to examine the tradeoffs when multilayer network architectures (see [354], [422], [423]) are
exploiting the global perspective of orchestration of multiple formed by conjoining the layered technology regions (i) in
routing domains, i.e., the failure recovery and restoration tech- vertical fashion i.e., multiple technology layers internetwork
niques surveyed in this section could be combined with the within a single domain, or (ii) in horizontal layering fashion
multidomain orchestration techniques surveyed in Section VII. across multiple domains, i.e., technology layers internetwork
One concrete example of multidomain orchestration could be across distinct domains. Horizontal multilayer networking can
to coordinate the specific LR-PON access network protec- be viewed as a generalization of vertical multilayer network-
tion and failure recovery [391] with protection and recov- ing in that the horizontal networking may involve the same
ery techniques for metropolitan and core network domains or different (or even multiple) layers in the distinct domains.
THYAGATURU et al.: SDONs: A COMPREHENSIVE SURVEY 2767

Fig. 19. Illustration of SDN orchestration of multilayer networking: Vertical MultiLayer Networking (MLN) spans layers at different horizontal positions
within a given domain. Horizontal MLN spans multiple layers at the same horizontal position (or in different horizontal positions) across multiple domains.
The inter-domain SDN orchestrator coordinates the individual intra-domain SDN controllers.

Fig. 20. Classification of SDON orchestration studies: Multilayer orchestration studies focus on vertical multilayer networking within a single domain.
Multidomain orchestration studies focus on horizontal multilayer networking across multiple domains and may involve multiple vertical layers in the various
domains.

As illustrated in Fig. 19, the formed SDN based multilayer A. Multilayer Orchestration
network architecture is controlled by an SDN orchestrator. 1) Multilayer Orchestration Frameworks:
As illustrated in Fig. 20 we organize the SDON orchestra- a) Hierarchical multilayer control: Felix et al. [424] pre-
tion studies according to their focus into studies that primarily sented an hierarchical SDN control mechanism for packet
address the orchestration of vertical multilayer (multitechnol- optical networks. Multilayer optimization techniques are
ogy) networking, i.e., the vertical networking across multiple employed at the SDN orchestrator to integrate the optical
layers (that typically implement different technologies) within transport technology with packet services by provisioning
a given domain, and into studies that primarily address the end-to-end Ethernet services. Two aspects are investigated,
orchestration of horizontal multilayer (multidomain) network- namely (i) bandwidth optimization for the optical trans-
ing, i.e., the horizontal networking across multiple routing port services, and (ii) congestion control for packet net-
domains (which may possibly involve different or multiple work services in an integrated packet optical network. More
vertical layers in the different domains). We subclassify the specifically, the SDN controller initially allocates the min-
vertical multilayer studies into general (vertical) multilayer imum available bandwidth required for the services and
networking frameworks and studies focused on supporting then dynamically scales allocations based on the avail-
specific applications through vertical multilayer networking. ability. Optical-Virtual Private Networks (O-VPNs) are cre-
We subclassify the multidomain (horizontal multilayer) net- ated over the physical transport network. Services are then
working studies into studies on general network domains and mapped to O-VPNs based on class of service require-
studies focused on internetworking with Data Center (DC) ments. When congestion is detected for a service, the
network domains. SDN controller switches the service to another O-VPN,
2768 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 18, NO. 4, FOURTH QUARTER 2016

thus balancing the traffic to maintain the required class of the remainder capacities in the optical links, i.e., the capac-
service. ity reserved for failure recoveries. The multilayer optimization
Similar steps towards the orchestration of multi- technique was validated on an experimental testbed utilizing
layer networks have been taken within the OFELIA central path-computation (PCE) [67] within the SDN frame-
project [425]–[427]. Specifically, Shirazipour et al. [428] work. Experimental verification of failure recovery mechanism
have explored extensions to OpenFlow version 1.1 actions to resulted in recovery times on the order of sub-seconds for
enable multitechnology transport layers, including Ethernet MPLS restorations and several seconds for optical WSON
transport and optical transport. The explorations of the exten- restorations.
sions include justifications of the use of SDN in circuit-based b) Resource utilization: Liu et al. [431] proposed a
transport networks. method to improve resource utilization and to reduce trans-
b) Application centric orchestration: Gerstel et al. [429] mission latencies through the processes of virtualization and
proposed an application centric network service provisioning service abstraction. A centralized SDN control implements the
approach based on multilayer orchestration. This approach service abstraction layer (to enable SDN orchestrations) in
enables the network applications to directly interact with the order to integrate the network topology management (across
physical layer resource allocations to achieve the desired ser- both IP and WDM), and the spectrum resource allocation in
vice requirements. Application requirements for a network a single control platform. The SDN orchestrator also achieves
service may include maximum end-to-end latency, connec- dynamic and simultaneous connection establishment across
tion setup and hold times, failure protection, as well as both IP and OTN layers reducing the transmission laten-
security and encryption. In traditional IP networking, packets cies. The control plane design is split between local (child)
from multiple applications requiring heterogeneous services and root (parent) controllers. The local controller realizes
are simply aggregated and sent over a common transport link the label switched paths on the optical nodes while the root
(IP services). As a result, network applications are typically controller realizes the forwarding rules for realizing the IP
assigned to a single (common) transport service within an layer. Experimental evaluation of average transfer time mea-
optical link. Consider a failure recovery process with multi- surements showed IP layer latencies on the order of several
ple available paths. IP networking typically selects the single milliseconds, and several hundreds of milliseconds for the
path with the least end-to-end delay. However, some applica- OTN latencies, validating the feasibility of control plane
tions may tolerate higher latencies and therefore, the traffic can unification for IP over optical transport networks.
be split over multiple restoration paths achieving better traffic c) Virtual optical networks (VONs): Vilalta et al. [432]
management. The orchestrator needs to interact with multiple presented controller orchestration to integrate multiple trans-
network controllers operating across multiple (vertical) lay- port network technologies, such as IP and GMPLS. The
ers supported by north/south bound interfaces to achieve the proposed architectural framework devises VONs to enable the
application centric control. Dynamic additions of new IP links virtualization of the physical resources within each domain.
are demonstrated to accommodate the requirements of multi- VONs are managed by lower level physical controllers (PCs),
ple application services with multiple IP links when the load which are hierarchically managed by an SDN network orches-
on the existing IP link was increased. trator (NO). Network Virtualization Controllers (NVC) are
2) Application-Specific Orchestration: introduced (on top of the NO) to abstract the virtualized mul-
a) Failure recovery: Generally, network CapEx and tilayers across multiple domains. End-to-end provisioning of
OpEx increase as more protection against network failures VONs is facilitated through hierarchical control interaction
is added. Khaddam et al. [430] propose an SDN based over three levels, the customer controller, the NO&NVCs, and
integration of multiple layers, such as WDM and IP, in a fail- the PCs. An experimental evaluation demonstrated average
ure recovery mechanism to improve the utilization (i.e., to VON provisioning delays on the order of several seconds (5 s
eventually reduce CapEx and OpEx while maintaining high and 10 s), validating the flexibility of dynamic VON deploy-
protection levels). An observation study was conducted over ments over the optical transport networks. Longer provisioning
a five year period to understand the impact of network fail- delays may impact the network application requirements, such
ures on the real deployment of backbone networks. Results as failure recovery processes, congestion control, and traffic
showed 75 distinct failures following a Pareto distribution, in engineering. General pitfalls of such hierarchical structures are
which, 48% of the total deployed capacity was affected by the increased control plane complexity, risk of controller failures,
top (i.e., the highest impact) 20% of the failures. And, 10% and maintenance of reliable communication links between
of the total deployed capacity was impacted by the top two control plane entities.
failure instances. These results emphasize the significance of
backup capacities in the optical links for restoration processes.
However, attaining the optimal protection capacities while B. Multidomain Orchestration
achieving a high utilization of the optical links is challenging. Large scale network deployments typically involve multiple
A failure recovery mechanism is proposed based on a “hybrid” domains, which have often heterogeneous layer technolo-
(i.e., combination of optical transport and IP) multilayer opti- gies. Achieve high utilization of the networking resources
mization. The hybrid mechanism improved the optical link while provisioning end-to-end network paths and services
utilization up to 50%. Specifically, 30% increase of the trans- across multiple domains and their respective layers and
port capacity utilization is achieved by dynamically reusing respective technologies is highly challenging [442]–[444].
THYAGATURU et al.: SDONs: A COMPREHENSIVE SURVEY 2769

2
Multidomain SDN orchestration studies have sought to exploit SDN Controller SDN Controller
the unified SDN control plane to aid the resource-efficient Domain-A Domain-B

provisioning across the multiple domains. 4

1 5 3
1) General Multidomain Networks:
a) Optical multitechnologies across multiple domains:
Optical nodes are becoming increasingly reconfigurable
(e.g., through variable BVTs and OFDM transceivers, see Domain-A
Domain-B
Section III), adding flexibility to the switching elements.
When a single end-to-end service establishment is consid-
ered, it is more likely that a service is supported by different
Fig. 21. Inter-domain lightpath provisioning mechanism facilitated by
optical technologies that operate across multiple domains. an Inter-Domain Protocol (IDP, which provides inter-domain request and
Yoshida et al. [419] have demonstrated SDN based orches- inter-domain reply messages) by employing the Routing and Spectrum
tration with emphasis on the physical interconnects between Allocation (RSA) algorithm proposed in [434]. Steps 1 through 5 provision
an end-to-end path across multiple domains.
multiple domains and multiple technology specific controllers
so as to realize end-to-end services. OpenFlow capabilities
have been extended for fixed-length variable capacity opti-
cal packet switching [445]. That is, when an optical switch
matches the label on an incoming optical packet, if a rule exists
in the switch (flow entry in the table) for a specific label, a
defined action is performed on the optical packet by the switch.
Otherwise, the optical packet is dropped and the controller is
notified. Interconnects between optical packet switching net-
works and elastic optical networks are enabled through a novel
OPS-EON interface card. The OPS-EON interface is designed
as an extension to a reconfigurable, programmable and flexi-
grid EON supporting the OpenFlow protocol. The testbed
implementation of OPS-EON interface cards demonstrated the
orchestration of multiple domain controllers and the recon-
figurability of FL-VC OPS across multidomain, multilayer,
multitechnology scenarios. Fig. 22. Illustration of multilevel virtualization enabled by the Multidomain
b) Hierarchical multidomain control: Jing et al. [433] Network Hypervisor (MNH) [435] operating over a network orchestrator
controller and domain specific SDN controllers to provide the multidomain
have also examined the integration of multiple optical trans- end-to-end virtualization.
port technologies from multiple vendors across multiple
domains, focusing on the control mechanisms across mul-
tiple domains. Jing et al. [433] proposed hierarchical SDN for the end-to-end provisioning of services in the SD-EONs.
orchestration with parent and domain controllers. Domain The distributed RSA algorithm operates on the domain specific
controllers abstract the physical layer by virtualizing the controllers using the IDP protocol. The RSA considers both
network resources. A Parent Controller (PC) encompasses a transparent lightpath connections, i.e., all-optical lightpath,
Connection Controller (CC) and a Routing Controller (RC) and translucent lightpath connections, i.e., optical-electrical-
to process the abstracted virtual network. When a new con- optical connections. The benefit of such techniques is privacy,
nection setup request is received by the PC, the RC (within since the domain specific policies and topology information
the PC) evaluates the end-to-end routing mechanisms and are not shared among other network entities. Neighbor dis-
forwards the information to the CC. The CC breaks the covery is independently conducted by the domain specific
end-to-end routing information into shorter link segments controller or can initially be configured. A domain appears
belonging to a domain. Segmented routes are then sent to the as an abstracted virtual node to all other domain specific con-
respective domain controllers for link provisioning over the trollers. Each controller then assigns the shortest path routing
physical infrastructures. The proposed mechanism was exper- within a domain between its border nodes. An experimental
imentally verified on a testbed built with the commercial OTN setup validating the proposed mechanism was demonstrated
equipment. across geographically-distributed domains in the USA and
c) Inter-domain protocol: Zhu et al. [434] followed a China.
different approach for the SDN multidomain control mech- d) Multidomain network hypervisors: Vilalta et al. [435]
anisms by considering the flat arrangement of controllers as presented a mechanism for virtualizing multitechnology
shown in Fig. 21. Each domain is autonomously managed by optical, multitenant networks. The Multidomain Network
an SDN controller specific to the domain. An Inter-Domain Hypervisor (MNH) creates customer specific virtual network
Protocol (IDP) was devised to establish the communication slices managed by the customer specific SDN controllers
between domain specific controllers to coordinate the light- (residing at the customers’ locations) as illustrated in Fig. 22.
path setup across multiple domains. Zhu et al. [434] also Physical resources are managed by their domain specific phys-
proposed a Routing and Spectrum Allocation (RSA) algorithm ical SDN controllers. The MNH operates over the network
2770 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 18, NO. 4, FOURTH QUARTER 2016

Fig. 24. Illustration of SDN orchestration based on Hierarchical Path

Fig. 23. The application-based network operations (ABNO) based SDN mul- Computation Element (H-PCE) [439]: The H-PCE internetworks GMPLS
tilayer orchestrator [436] receives the physical topology information from the inter-DC communication and OpenFlow intra-DC communication. The parent-
OpenFlow/GMPLS controllers. The orchestrator centrally computes paths and PCE (pPCE) aggregates the active PCE states from the child-PCEs (cPCEs)
sends the path information to the lower level controllers for path provisioning. of both GMPLS and OpenFlow.

orchestrator and physical SDN controllers for provisioning based ROADMs, or may be heterogeneous, i.e., have dif-
VONs on the physical infrastructures. The MNH abstracts ferent types of network technologies, e.g., OpenFlow based
both (i) multiple optical transport technologies, such as opti- ROADMs and GMPLS based WSON. The SDN control struc-
cal packet switching and Elastic Optical Networks (EONs), tures for a multidomain network can be broadly classified
and (ii) multiple control domains, such as GMPLS and into the categories of (i) single SDN orchestrator/controller,
OpenFlow. Experimental assessments on a testbed achieved (ii) multiple mesh SDN controllers, and (iii) multiple hierar-
VON provisioning within a few seconds (5 s), and control chical SDN controllers [437], [438]. The single SDN orches-
overhead delay on the order of several tens of milliseconds. trator/controller has to support heterogeneous SBIs in order
Related virtualization mechanisms for multidomain optical to operate with multiple heterogeneous domains, e.g., the
SDN networks with end-to-end provisioning have been inves- Path Computation Element Protocol (PCEP) for GMPLS net-
tigated in [446] and [447]. work domains and the OpenFlow protocol for OpenFlow
e) Application-based network operations: supported ROADMs. Also, domain specific details, such as
Muñoz et al. [436], have presented an SDN orchestration topology, as well as network statistics and configurations, have
mechanism based on the application-based network operations to be exposed to an external entity, namely the single SDN
(ABNO) framework, which is being defined by the IETF [448]. orchestrator/controller, raising privacy concerns. Furthermore,
The ABNO based SDN orchestrator integrates OpenFlow and a single controller may result in scalability issues. Mesh SDN
GMPLS in transport networks. Two SDN orchestration designs control connects the domain-specific controllers side-by-side
have been presented: (i) with centralized physical network by extending the east/west bound interfaces. Although mesh
topology aware path computation (illustrated in Fig. 23), SDN control addresses the scalability and privacy issues, the
and (ii) with topology abstraction and distributed path com- distributed nature of the control mechanisms may lead to
putation. In the centralized design, OpenFlow and GMPLS sub-optimal solutions. With hierarchical SDN control, a logi-
controllers (lower level control) expose the physical topology cally centralized controller (parent SDN controller) is placed
information to the ABNO-orchestrator (higher level control). above the domain-specific controllers (child SDN controllers),
The PCE in the ABNO-orchestrator has the global view of extending the north/south bound interfaces. Domain-specific
the network and can compute end-to-end paths with com- controllers virtualize the underlying networks inside their
plete knowledge of the network. Computed paths are then domains, exposing only the abstracted view of the domains
provisioned through the lower level controllers. The pitfalls of to the parent controller, which addresses the privacy con-
such centralized designs are (i) computationally intensive path cerns. Centralized path computation at the parent controller
computations, (ii) continuous updates of topology and traffic can achieve optimal solutions. Multiple hierarchical levels can
information, and (iii) sharing of confidential network infor- address the scalability issues. These advantages of hierarchal
mation and policies with other network elements. To reduce SDN control are achieved at the expense of an increased
the computational load at the orchestrator, the second design number of network entities, resulting in the operational com-
implements distributed path computation at the lower level plexities.
controllers (instead of path computation at the centralized b) Hierarchical PCE: Casellas et al. [439] consid-
orchestrator). However, such distributed mechanisms may lead ered DC connectivities involving both intra-DC and inter-DC
to suboptimal solutions due to the limited network knowledge. communications. Intra-DC communications enabled through
2) Multidomain Data Center Orchestration: OpenFlow networks are supported by an OpenFlow controller.
a) Control architectures: Geographically distributed The inter-DC communications are enabled by optical transport
DCs are typically interconnected by links traversing multiple networks involving more complex control, such as GMPLS,
domains. The traversed domains may be homogeneous i.e., as illustrated in Fig. 24. To achieve the desired SDN benefits
have the same type of network technology, e.g., OpenFlow of flexibility and scalability, a common centralized control
THYAGATURU et al.: SDONs: A COMPREHENSIVE SURVEY 2771

platform spanning across heterogeneous control domains is Relatively more SDN orchestration studies to date have
proposed. More specifically, an Hierarchical PCE (H-PCE) examined multidomain networking than multilayer network-
aggregates PCE states from multiple domains. The end-to- ing (within a single domain). As the completed multidomain
end path setup between DCs is orchestrated by a parent-PCE orchestration studies have demonstrated, the SDN orchestra-
(pPCE) element, while the paths are provisioned by the child- tion can help greatly in coordinating complex network man-
PCEs (cPCEs) on the physical resources, i.e., the OpenFlow agement decisions across multiple distributed routing domains.
and GMPLS domains. The proposed mechanism utilizes exist- The completed studies have illustrated the fundamental trade-
ing protocol interfaces, such as BGP-LS and PCEP, which are off between centralized decision making in a hierarchical
extended with OpenFlow to support the H-PCE. orchestration structure and distributed decision making in a
c) Virtual-SDN control: Muñoz et al. [440] and flat orchestration structure. In particular, most studies have
Vilalta et al. [441] proposed a mechanism to virtualize the focused on hierarchical structures [433], [439], [440], while
SDN control functions in a DC/cloud by integrating SDN with only one study has mainly focused on a flat orchestration
Network Function Virtualization (NFV). In the considered structure [434]. In the context of DC internetworking, the
context, NFV refers to realizing network functions by soft- studies [437], [438] have sought to bring out the trade-
ware modules running on generic computing hardware inside offs between these two structures by examining a range of
a DC; these network functions were conventionally imple- structures from centralized to distributed. While centralized
mented on specialized hardware modules. The orchestration of orchestration can make decisions with a wide knowledge hori-
Virtual Network Functions (VNFs) is enabled by an integrated zon across the states in multiple domains, distributed decision
SDN and NFV management which dynamically instantiates making preserves the privacy of network status information,
virtual SDN controllers. The virtual SDN controllers control reduces control traffic, and can make fast localized decisions.
the Virtual Tenant Networks (VTNs), i.e., virtual multidomain Future research needs to shed further light on these complex
and multitechnology networks. Multiple VNFs running on a tradeoffs for a wide range of combinations of optical technolo-
Virtual Machine (VM) in a DC are managed by a VNF manger. gies employed in the various domains. Throughout, it will be
A virtual SDN controller is responsible for creating, manag- critical to abstract and convey the key characteristics of optical
ing, and tearing down the VNF achieving the flexibility in the physical layer components and switching nodes to the over-
control plane management of the multilayer and the multido- all orchestration protocols. Optimizing each abstraction step
main networks. Additionally, as an extension to the proposed as well as the overall orchestration and examining the various
mechanism, the virtualization of the control functions of the performance tradeoffs are important future research directions.
LTE Evolved Packet Core (EPC) has been discussed in [449].
VIII. O PEN C HALLENGES AND F UTURE SDON
R ESEARCH D IRECTIONS
C. Orchestration: Summary and Discussion We have outlined open challenges and future Software
Relatively few SDN orchestration studies to date have Defined Optical Network (SDON) research directions for each
focused on vertical multilayer networking within a given sub-category of surveyed SDON studies in the Summary and
domain. The few studies have developed two general orches- Discussion subsections in the preceding survey sections. In
tration frameworks and have examined a few orchestration this section, we focus on the overall cross-cutting open chal-
strategies for some specific applications. More specifically, lenges that span across the preceding considered categories
one orchestration framework has focused on optimal band- of SDON studies. That is, we focus on open challenges and
width allocation based mainly on congestion [424], while the research directions that span the vertical (inter-layer) and hor-
other framework has focused on exploiting application traf- izontal (inter-domain) SDON aspects. The vertical SDON
fic tolerances for delays for efficiently routing traffic [429]. aspects encompass the seamless integration of the various
SDN orchestration of vertical multilayer optical network- (vertical) layers of the SDON architecture; especially the
ing is thus still a relatively little explored area. Future optical layer, which is not considered in general SDN tech-
research can develop orchestration frameworks that accom- nology. The horizontal SDON aspects include the integration
modate the specific optical communication technologies in of SDONs with existing non-SDN optical networking ele-
the various layers and rigorously examine their performance- ments, and the internetworking with other domains, which may
complexity tradeoffs. Similarly, relatively few applications have similar or different SDN architectures. A key challenge
have been examined to date in the application-specific orches- for SDON research is to enable the use of SDON concepts
tration studies for vertical multilayer networking [430]–[432]. in operational real-time network infrastructures. Importantly,
The examination of the wide range of existing applica- the SDON concepts need to demonstrate performance gains
tions and any newly emerging network application in the and cost reductions to be considered by network and service
context of SDN orchestrated vertical multilayer networking providers. Therefore, we cater some of the open challenges
presents rich research opportunities. The cross-layer perspec- and future directions towards enabling and demonstrating the
tive of the SDN orchestrator over a given domain could, successful use of SDON in operational networks.
for instance, be exploited for strengthening security and pri- The SDON research and development effort to date have
vacy mechanisms or for accommodating demanding real-time resulted in insights for making the use of SDN in optical
multimedia. transport networks feasible and have demonstrated advantages
2772 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 18, NO. 4, FOURTH QUARTER 2016

of SDN based optical network management. However, most optimize their operation across combinations of network archi-
network and service providers depend on optical transport tecture structures and across various network protocol layers.
to integrate with multiple industries to complete the network Simplicity is an essential part of this challenge, since overly
infrastructure. Often, network and service providers struggle complex solutions are generally not deployed due to the risk
to integrate hardware components and to provide accessible of high expenditures.
software management to customers. For example, companies
that develop hardware optical components do not always have
a complete associated software stack for the hardware compo- B. North Bound Interface
nents. Thus, network and service providers using the hardware The NorthBound Interface (NBI) comprises the commu-
optical components often have to maintain a software devel- nication from the controller to the applications. This is an
opment team to integrate the various hardware components important area of future research as applications and their
through software based management into their network, which needs are generally the driving force for deploying SDON
is often a costly endeavor. Thus, improving SDN technology so infrastructures. Any application, such as video on demand,
that it seamlessly integrates with components of various indus- VoIP, file transfer, or peer-to-peer networking, is applied from
tries and helps the integration of components from various the NBI to the SDN controller which consequently con-
industries is an essential underlying theme for future SDON ducts the necessary actions to implement the service behaviors
research. on the physical network infrastructure. Applications often
require specific service behaviors that need to be implemented
on the overall network infrastructure. For example, applica-
A. Simplicity and Efficiency tions requiring high data rates and reliability, such as Netflix,
Optical network structures typically span heterogeneous depend on data centers and the availability of data from servers
devices ranging from the end user nodes and local area net- with highly resilient failure protection mechanisms. The asso-
works via ONUs and OLTs in the access networks to edge ciated management network needs to stack redundant devices
routers and metro network nodes and on to backbone (core) as to safeguard against outages. Services are provided as poli-
network infrastructures. These different devices often come cies through the NBI to the SDN controller, which in turn
from different vendors. The heterogeneity of devices and generates flow rules for the switching devices. These flow
their vendors often requires manual configuration and mainte- rules can be prioritized based on the customer use cases. An
nance of optical networks. Moreover, different communication important challenge for future NBI research is to provide a
technologies typically require the implementation of native simple interface for a wide variety of service deployments
functions that are specific to the communication technology without vendor lock-in, as vendor lock-in generally drives
characteristics, e.g., the transmission and propagation proper- costs up. Also, new forms of communication to the controller,
ties. By centralizing the optical network control in an SDN in addition to current techniques, such as REpresentational
controller, the SDN networking paradigm creates a unified State Transfer (REST) [64] and HTTP, should be researched.
view of the entire optical network. The specific native func- Moreover, future research should develop an NBI framework
tions for specific communication devices can be migrated to that spans horizontally across multiple controllers, so that
the software layer and be implemented by a central node, service customers are not restricted to using only a single
rather than through manual node-by-node configurations. The controller.
central node would typically be readily accessible and could Future research should examine control mechanisms that
reduce the required physical accesses to distributed devices optimally exploit the central SDN control to provide simple
at their on-site locations. This centralization can simplify the and efficient mechanisms for automatic network management
network management and reduce operational expenditures. An and dynamic service deployment [450]. The NBI of SDONs
important challenge in this central management is the efficient is a challenging facet of research and development because
SDN control of components from multiple vendors. Detailed of the multitude of interfaces that need to be managed on the
vendor contract specifications of open-source middleware may physical layer and transport layer. Optical physical layer com-
be needed to efficiently control components from different ponents and infrastructures require high capital and operational
vendors. expenditures and their management is generally not associated
The heterogeneity of devices may reduce the efficiency of with network or service providers but rather with optical com-
network infrastructures due to the required multiple software ponent/infrastructure vendors. Future research should develop
and hardware modules for a complete networking solution. novel Application Program Interfaces (APIs) for optical layer
Future research should investigate efficient mechanisms for components and infrastructures that facilitate SDN control
making complete networking solutions available for spe- and are amenable to efficient NBI communication. Essentially,
cific use cases. For example, the use of SDON for an the challenge of efficient NBI communication with the SDN
access network provider may require multiple SDN controllers controller should be considered when designing the APIs
co-located within the OLT to enable the control of the access that interface with the physical optical layer components and
network infrastructure from one central location. While the infrastructures.
SDON studies reviewed in this survey have led initial inves- One specific strategy for simplifying network management
tigations of simple and dynamic network management, future and operation could be to explore the grouping of con-
research needs to refine these management strategies and trol policies of similar service applications, e.g., applications
THYAGATURU et al.: SDONs: A COMPREHENSIVE SURVEY 2773

with similar QoS requirements. The grouping can reduce devices to the controller has received relatively little attention.
the number of control policies at the expense of slightly For example, there are three types of OpenFlow messages,
coarser granularity of the service offerings. The emerging namely Packet-In, Packet-Out, and Flow-Mod. The Packet-In
Intent-Based Networking (IBN) paradigm, which drafts intents messages are sent from the OpenFlow switches to the con-
for services and policies, can provide a specific avenue for troller, the Packet-Out message is sent from the controller
simplifying dynamic automatic configuration and virtualiza- to the device, and the Flow-Mod message is used to modify
tion [451], [452]. Currently network applications are deployed and monitor the flow rules in the flow table. Future research
based on how the network should behave for a specific action. should examine extensions of the Packet-In message to send
For example, for inter domain routing, the Border Gateway specific status updates in support of network and device fail-
Protocol (BGP) is used, and the network gateways are config- ure monitoring to the controller. These status messages could
ured to communicate with the BGP protocol. This complicates be monitored by a dedicated failure monitoring service. The
the provisioning of services that typically require multiple pro- status update messages could be broadly defined to cover a
tocols and limits the flexibility of service provisioning. With wide range of network management aspects, including system
IBN, the application gives an intent, for example, transferring health monitoring and network failure protection.
video across multiple domains. This intent is then associated A related future research direction is to secure configuration
with automated dynamic configurations of the network ele- and operation of SDONs through trusted encryption and key
ments to communicate data over the domains using appropriate management systems [28]. Moreover, mechanisms to ensure
protocols. The grouping of service policies, such as intents, the privacy of the communication should be explored. The
can facilitate easy and dynamic service provisioning. Intent security and privacy mechanisms should strive to exploit the
groups can be described in a graph to simplify the compilation natural immunity of optical transmission segments to electro-
of service policies and to resolve conflicts [453]. magnetic interferences.
In summary, security and privacy of SDON communication
are largely open research areas. The optical physical layer
C. Reliability, Security, and Privacy infrastructure has traditionally not been controlled remotely,
The SDN paradigm is based on a centrally managed which in general reduces the occurrences of security breaches.
network. Faulty behaviors, security infringements, or failures However, centralized SDN management and control increase
of the control would likely result in extensive disruptions and the risk of security breaches, requiring extensive research on
performance losses that are exacerbated by the centralized SDON security, so as to reap the benefits of centralized SDN
nature of the SDN control. Instances of extensive disruptions management and control in a secure manner.
and losses due to SDN control failures or infringements would
likely reduce the trust in SDN deployments. Therefore, it is
very important to ensure reliable network operation [454] and D. Scalability
to provision for security and privacy of the communication. Optical networks are expensive and used for high-bandwidth
Hence, reliability, security, and privacy are prominent SDON services, such as long-distance network access and data center
research challenges. Security in SDON techniques is a fairly interconnections. Optical network infrastructures either span
open research area, with only few published findings. As a few long distances between multiple geographically distributed
reviewed studies (see Section VI-D) have explored, the central locations, or could be short-distance incremental additions
SDN control can facilitate reliable network service through (interconnects) of computing devices. Scalability in multiple
speeding up failure recovery. The central SDN control can dimensions is therefore an important aspect for future SDON
continuously scan the network and the status messages from research. For example, a myriad of tiny end devices need to
the network devices. Or, the SDN control can redirect the sta- be provided with network access in the emerging Internet of
tus messages to a monitoring service that analyzes the data Things (IoT) paradigm [268]. The IoT requires access network
network. Security breaches can be controlled by broadcast- architectures and protocols to scale vertically (across proto-
ing messages from the controller to all affected devices to col layers and technologies) and horizontally (across network
block traffic in a specific direction. Future research should domains). At the same time, the ongoing growth of multimedia
refine these reliability functions to optimize automated fault services requires data centers to scale up optical network band-
and performance diagnostics and reconfigurations for quick widths to maintain the quality of experience of the multimedia
failure recovery. services. Broadly speaking, scalability includes in the vertical
Network failures can either occur within the physical layer dimension the support for multiple network devices and tech-
infrastructure, or as errors within the higher protocol layers, nologies. Scalability in the horizontal direction includes the
e.g., in the classical data link (L2), network (L3), of transport communication between a large number of different domains
(L4) layers. In the context of SDONs, physical layer failures as well as support for existing non-SDON infrastructures.
present important future research opportunities. Physical layer A specific scalability challenge arising with SDN infras-
devices need to be carefully monitored by sending feedback tructure is that the scalability of the control plane (OpenFlow
from the devices to the controller. The research and develop- protocol signalling) communication and the scalability of
ment on communication between the SDN controller and the the data plane communication which transports the data
network devices has mainly focused on sending flow rules to plane flows need to be jointly considered. For example,
the network devices while feedback communicated from the the Openflow protocol 1.4 currently supports 34 Flow-Mod
2774 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 18, NO. 4, FOURTH QUARTER 2016

messages [455], which can communicate between the network well as horizontal multilayer (multidomain) networking across
devices and the controller. This number limits the functional- multiple domains. We proceed to outline open challenges and
ity of the SBI communication. Recent studies have explored a future research directions for vertical multilayer networking
protocol-agnostic approach [59], [456], which is a data plane in the context of SDON, which includes an optical physical
protocol that extends the use of multiple protocols for com- layer, in this subsection. Horizontal multilayer (multidomain)
munication between the control plane and data plane. The networking is considered in Section VIII-G.
protocol-agnostic approach resolves the challenges faced by For the vertical multilayer networking in a single domain,
OpenFlow and, in general, any particular protocol. Exploring the optical physical layer is the key distinguishing feature
this novel protocol-agnostic approach presents many new of SDONs compared to conventional SDN architectures for
SDON research directions. general IP networks. Most of the higher layers in SDONs
Scalability would also require SDN technology to overlay have similar multilayer networking challenges as general IP
and scale over existing non-SDN infrastructures. Vendors pro- networks. However, the optical physical layer requires the
vide support for known non-SDN devices, but this area is provisioning of specific optical transmission parameters, such
still a challenge. There are no known protocols that could as wavelengths and signal strengths. These parameters are
modify the flow tables of existing popularly described “non- managed by optical devices, such as the OLT in PON net-
OpenFlow” switches. In the case of optical networks, as SDN works. For SDON networks, so-called optical orchestrators,
is still being incrementally deployed, the overlaying with non- which are commercially available, e.g., from ADVA Optical
SDN infrastructure still requires significant attention. Ideally, Networking, provide a single interface to provision the optical
the overlay mechanisms should ensure seamless integration layer parameters. We illustrate this optical orchestrator layer
and should scale with the growing deployment of SDN tech- in the context of an SDON multilayer network in the right-
nologies while incurring only low costs. Overall, scalability most branch of Fig. 2. The optical orchestrator resides above
poses highly important future SDON research directions that the optical devices and below the SDN controller. The opti-
require economical solutions. cal orchestrator uses common SDN SBI interface protocols,
such as OpenFlow, to communicate with the optical devices
E. Standardization
in the south-bound direction and with the controller in the
Networking protocols have traditionally followed a uniform north-bound direction.
standard system for all the communication across multi- The SDN controller in the control plane is responsible
ple domains. Standardization has helped vendors to provide for the management of the SDN-enabled switches, poten-
products that work in and across different network infras- tially via an optical orchestrator. Communicating over the
tructures. In order to ensure the compatible inter-operation SBI using different protocols can be challenging for the con-
of SDON components (both hardware and software) from a troller. This challenge can be addressed by using south-bound
various vendors, key aspects of the inter-operation protocols renderers. South-bound renderers are APIs that reside within
need to be standardized. Towards the standardization goal, the controller and provide a communication channel to any
communities, such as Open Networking Foundation (ONF), desired SBI protocol. Most SDN controllers currently have an
have created boards and committees to standardize protocols, OpenFlow renderer to be able to communicate to OpenFlow-
such as OpenFlow. Standardization should ensure that SDON enabled network switches. But there are also SNMP and
infrastructures can be flexibly configured and operated with NETCONF-based renderers, which communicate with tradi-
components from various vendors. The use of open-source tional non-OpenFlow switches. This enables the existence of
software can further facilitate the inter-operation. Proprietary hybrid networks with already existing switches. The effective
hardware and software components generally create vendor support of such hybrid networks, in conjunction with appro-
lock-in, which restricts the flexibility of network operation and priate south-bound renderers and optical orchestrators, is an
reduces the innovation of network and service providers. important direction for future research.
As groundwork for standardization, it may be necessary to
develop and optimize a common (or a small set) of SDON
architectures and network protocol configurations that can G. Multidomain Networks
serve as a basis for standardization efforts. The standardization A network domain usually belongs to a single organiza-
process may involve a common platform that is built thorough tion that owns (i.e., financially supports and uses) the network
the cooperation of multiple manufacturers. Another thrust domain. The management of multidomain networking involves
of standardization groundwork could be the development of the important aspects of configuring the access control as
open-source software that supports SDON architectures. For well as the authentication, authorization, and accounting.
example, Openstack is a cloud based management framework Efficient SDN control mechanisms for configuring these mul-
that has been adopted and supported by multiple networking tidomain networking aspects is an important direction for
vendors. Such efforts should be extended to SDONs in future future research and development.
work. Multidomain SDONs may also need novel routing algo-
rithm that enhance the capabilities of the currently used BGP
F. Multilayer Networking protocol. Multidomain research [457] has now taken interest in
As discussed in Section VII, multilayer networking involves the Intent-Based Networking (NBI) paradigm for SDN control,
vertical multilayer networking across the vertical layers as where Intent-APIs can solve the problems of spanning across
THYAGATURU et al.: SDONs: A COMPREHENSIVE SURVEY 2775

multiple domains. For instance, the intent of an application to of SDN, NFV, and optical technologies to achieve the grow-
transfer information across multiple domains is translated into ing demands of broadcasters and media. Moreover, automated
service instances that access configurations between domains provisioning strategies of QoS and the incorporation of quality
that have been pre-configured based on contracts. Currently, of protection and security with traditional QoS are important
costly manual configurations between domains are required for direction for future QoS research in SDONs.
such applications. Future research needs to develop concrete
models for NBI based multidomain networking in SDONs. J. Performance Evaluation
Comprehensive performance evaluation methodologies and
H. Fiber-Wireless (FiWi) Networking metrics need to be developed to assess the SDON designs
The optical (fiber) and wireless network domains have many addressing the preceding future research directions ranging
differences. At the physical layer, wireless networks are char- from simplicity and efficiency (Section VIII-A) to optical-
acterized by varying channel qualities, potentially high losses, wireless networks (Section VIII-H). The performance evalua-
and generally lower transmission bit rates than optical fiber. tions need to encompass the data plane, the control plane, as
Wireless end nodes are typically mobile and may connect well as the overall data and control plane interactions with the
dynamically to wireless network domains. The mobile wire- SDN interfaces and need to take virtualization and orchestra-
less nodes are generally the end-nodes in a FiWi network tion mechanisms into consideration. In the case of the SDON
and connect via intermediate optical nodes to the Internet. infrastructure, the performance evaluations will need to include
Due to these different characteristics, the management of the optical physical layer [470]. While there have been some
wireless networks with mobile end nodes is very different efforts to develop evaluation frameworks for general SDN
from the management of optical network nodes. For exam- switches [471], [472], such evaluation frameworks need to
ple, wireless access points should maintain their own routing be adapted to the specific characteristics of SDON architec-
table to accommodate access to dynamically connected mobile tures. Similarly, some evaluation frameworks for general SDN
devices. Combining the control of both wireless and optical controllers have been explored [473], [474]; these need to be
networks in a single SDN controller requires concrete APIs extended to the specific SDON control mechanisms.
that handle the respective control functions of wireless and Generally, performance metrics obtained with SDN and vir-
optical networks. Currently, service providers maintain sep- tualization mechanisms should be benchmarked against the
arate physical management services without a unified logical corresponding conventional network without any SDN or vir-
control and management plane for FiWi networks. Developing tualization components. Thus, the performance tradeoffs and
integrated controls for FiWi networks can be viewed as a costs of the flexibility gained through SDN and virtualization
special case of multilayer networking and integration. mechanism can be quantified. This quantified data would then
Developing specialized multilayer networking strategies for need to be assessed and compared in the context of business
FiWi networks is an important future research directions needs. To identify some of the important aspects of perfor-
as many aspects of wireless networks have dramatically mance we analyze the sample architecture in Fig. 14. The SDN
advanced in recent years. For instance, the cell structure of controller in the SDON architecture in Fig. 14 spans across
wireless cellular networks [458] has advanced to femtocell multiple elements, such as ONUs, OLTs, routers/switches in
networks [459] as well as heterogeneous and multitier cellular the metro-section, as well as PCEs in the core section. A
structures [460], [461]. At the same time, machine-to-machine meaningful performance evaluation of such a network requires
communication [462], [463] and energy savings [464], [465] comprehensive analysis of data plane performance aspects and
have drawn research attention. related metrics, including noise spectral analysis, bandwidth
and link rate monitoring, as well as evaluation of failure
I. QoS and Energy Efficiency resilience. Performance evaluation mechanisms need to be
developed to enable the SDON controller to obtain and analyze
Different types of applications have vastly different traffic
these performance data. In addition, mechanisms for con-
bit rate characteristics and QoS requirements. For instance,
trol layer performance analysis are needed. The control plane
streaming high-definition video requires high bit rates, but
performance evaluation should, for instance assess the con-
can tolerate some delays with appropriate playout buffering.
troller efficiency and performance characteristics, such as the
On the other hand, VoIP (packet voice) or video conference
OpenFlow message rates and the rates and delays of flow table
applications have typically low to moderate bit rates, but
management actions.
require low latencies. Achieving these application-dependent
QoS levels in an energy-efficient manner [465]–[467] is an
important future research direction. A related future research IX. C ONCLUSION
direction is to exploit SDN control for QoS adaptations of real- We have presented a comprehensive survey of software
time media and broadcasting services. Broadcasting services defined optical networking (SDON) studies to date. We have
involve typically data rates ranging from 3–48 Gb/s to deliver mainly organized our survey according to the SDN infrastruc-
video at various resolutions to the users within a reasonable ture, control, and application layer structure. In addition, we
time limit. In addition to managing the QoS, the network has have dedicated sections to SDON virtualization and orchestra-
to manage the multicast groups for efficient routing of traffic tion studies. Our survey has found that SDON infrastructure
to the users. Recent studies [468], [469] discuss the potential studies have examined optical (photonic) transmission and
2776 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 18, NO. 4, FOURTH QUARTER 2016

switching components that are suitable for flexible SDN con- [12] L. Cui, F. R. Yu, and Q. Yan, “When big data meets software-defined
trolled operation. Moreover, flexible SDN controlled switching networking: SDN for big data and big data for SDN,” IEEE Netw.,
vol. 30, no. 1, pp. 58–65, Jan./Feb. 2016.
paradigms and optical performance monitoring frameworks [13] H. Farhady, H. Lee, and A. Nakao, “Software-defined networking:
have been investigated. A survey,” Comput. Netw., vol. 81, pp. 79–95, Apr. 2015.
SDON control studies have developed and evaluated SDN [14] N. Feamster, J. Rexford, and E. Zegura, “The road to SDN: An intel-
lectual history of programmable networks,” SIGCOMM Comput.
control frameworks for the wide range of optical network Commun. Rev., vol. 44, no. 2, pp. 87–98, Apr. 2014.
transmission approaches and network structures. Virtualization [15] R. Jain and S. Paul, “Network virtualization and software defined net-
allows for flexible operation of multiple Virtual Optical working for cloud computing: A survey,” IEEE Commun. Mag., vol. 51,
no. 11, pp. 24–31, Nov. 2013.
Networks (VONs) over a given installed physical optical net- [16] Y. Jarraya, T. Madi, and M. Debbabi, “A survey and a layered taxonomy
work infrastructure. The surveyed SDON virtualization studies of software-defined networking,” IEEE Commun. Surveys Tuts., vol. 16,
have examined the provisioning of VONs for access net- no. 4, pp. 1955–1980, 4th Quart. 2014.
[17] M. Jarschel, T. Zinner, T. Hossfeld, P. Tran-Gia, and W. Kellerer,
works, exploiting the specific physical and Medium Access “Interfaces, attributes, and use cases: A compass for SDN,” IEEE
Control (MAC) layer characteristics of access networks. The Commun. Mag., vol. 52, no. 6, pp. 210–217, Jun. 2014.
virtualization studies have also examined the provisioning of [18] B. Khasnabish, B.-Y. Choi, and N. Feamster, “JONS: Special issue
VONs in metro and backbone networks, examining algorithms on management of software-defined networks,” J. Netw. Syst. Manag.,
vol. 23, no. 2, pp. 249–251, 2015.
for embedding the VON topologies on the physical net- [19] C.-S. Li et al., “Software defined environments: An introduction,” IBM
work topology under consideration of the optical transmission J. Res. Develop., vol. 58, nos. 2–3, pp. 1–11, Mar./Apr. 2014.
characteristics. [20] F. A. Lopes, M. Santos, R. Fidalgo, and S. Fernandes, “A software engi-
neering perspective on SDN programmability,” IEEE Commun. Surveys
SDON application layer studies have developed mecha- Tuts., vol. 18, no. 2, pp. 1255–1272, 2nd Quart. 2016.
nisms for achieving Quality of Service (QoS), access control [21] O. Michel, M. Coughlin, and E. Keller, “Extending the software-
and security, as well as energy efficiency and failure recov- defined network boundary,” ACM SIGCOMM Comput. Commun. Rev.,
vol. 44, no. 4, pp. 381–382, Oct. 2014.
ery. SDON orchestration studies have examined coordination [22] B. A. A. Nunes, M. Mendonca, X.-N. Nguyen, K. Obraczka, and
mechanisms across multiple layers (in the vertical dimension T. Turletti, “A survey of software-defined networking: Past, present,
of the network protocol layer stack) as well as across multiple and future of programmable networks,” IEEE Commun. Surveys Tuts.,
vol. 16, no. 3, pp. 1617–1634, 3rd Quart. 2014.
network domains (that may belong to different organizations). [23] S. Racherla, D. Cain, S. Irwin, P. Patil, and A. M. Tarenzio,
While the SDON studies to date have established basic Implementing IBM Software Defined Network for Virtual Environments.
principles for incorporating and exploiting SDN control in Poughkeepsie, NY, USA: IBM Redbooks, 2014.
[24] C. Trois, M. D. D. D. Fabro, L. C. E. de Bona, and M. Martinello,
optical networks, there remain many open research challenges. “A survey on SDN programming languages: Towards a taxonomy,”
We have outlined open research challenges for each indi- IEEE Commun. Surveys Tuts., to be published.
vidual category of studies as well as cross-cutting research [25] B. J. van Asten, N. L. van Adrichem, and F. A. Kuipers,
“Scalability and resilience of software-defined networking: An
challenges. overview,” arXiv:1408.6760, 2014.
[26] J. A. Wickboldt et al., “Software-defined networking: Management
requirements and challenges,” IEEE Commun. Mag., vol. 53, no. 1,
R EFERENCES pp. 278–285, Jan. 2015.
[27] W. Xia, Y. Wen, C. H. Foh, D. Niyato, and H. Xie, “A survey on
[1] D. Comer, Automated Network Management Systems. software-defined networking,” IEEE Commun. Surveys Tuts., vol. 17,
Upper Saddle River, NJ, USA: Prentice Hall, 2007. no. 1, pp. 27–51, 1st Quart. 2015.
[2] F. Hu, Q. Hao, and K. Bao, “A survey on software-defined network and [28] I. Ahmad, S. Namal, M. Ylianttila, and A. Gurtov, “Security in software
OpenFlow: From concept to implementation,” IEEE Commun. Surveys defined networks: A survey,” IEEE Commun. Surveys Tuts., vol. 17,
Tuts., vol. 16, no. 4, pp. 2181–2206, 4th Quart. 2014. no. 4, pp. 2317–2346, 4th Quart. 2015.
[3] S. Jain et al., “B4: Experience with a globally-deployed software [29] S. Scott-Hayward, S. Natarajan, and S. Sezer, “A survey of security
defined WAN,” ACM SIGCOMM Comput. Commun. Rev., vol. 43, in software defined networks,” IEEE Commun. Surveys Tuts., vol. 18,
no. 4, pp. 3–14, 2013. no. 1, pp. 623–654, 1st Quart. 2016.
[4] S. J. Vaughan-Nichols, “OpenFlow: The next generation of the net- [30] L. Bertaux et al., “Software defined networking and virtualization for
work?” Computer, vol. 44, no. 8, pp. 13–15, Aug. 2011. broadband satellite networks,” IEEE Commun. Mag., vol. 53, no. 3,
[5] Open Networking Foundation, “SDN architecture overview, ver- pp. 54–60, Mar. 2015.
sion 1.1, ONF TR-504,” Palo Alto, CA, USA, Nov. 2014. [31] G. Zhang, M. De Leenheer, A. Morea, and B. Mukherjee, “A survey on
[6] D. Kreutz et al., “Software-defined networking: A comprehensive OFDM-based elastic core optical networking,” IEEE Commun. Surveys
survey,” Proc. IEEE, vol. 103, no. 1, pp. 14–76, Jan. 2015. Tuts., vol. 15, no. 1, pp. 65–87, 1st Quart. 2013.
[7] A. Lara, A. Kolasani, and B. Ramamurthy, “Network innovation using [32] P. Bhaumik et al., “Software-defined optical networks (SDONs):
OpenFlow: A survey,” IEEE Commun. Surveys Tuts., vol. 16, no. 1, A survey,” Photon. Netw. Commun., vol. 28, no. 1, pp. 4–18, 2014.
pp. 493–512, 1st Quart. 2014. [33] M. Channegowda, R. Nejabati, and D. Simeonidou, “Software-defined
[8] R. Enns, M. Bjorklund, J. Schoenwaelder, and A. Bierman, “Network optical networks technology and infrastructure: Enabling software-
configuration protocol (NETCONF),” Internet Requests for Comments, defined optical network operations [invited],” IEEE/OSA J. Opt.
RFC 6241, Fremont, CA, USA, Jun. 2011. [Online]. Available: Commun. Netw., vol. 5, no. 10, pp. A274–A282, Oct. 2013.
http://www.rfc-editor.org/rfc/rfc6241.txt [34] L. Liu et al., “Field trial of an OpenFlow-based unified control plane
[9] ONOS—A New Carrier-Grade SDN Network Operating System for multilayer multigranularity optical switching networks,” J. Lightw.
Designed for High Availability, Performance, Scale-Out. (2016). Technol., vol. 31, no. 4, pp. 506–514, Feb. 15, 2013.
[Online]. Available: http://onosproject.org/ [35] N. Cvijetic, “SDN for optical access networks,” in Proc. Photon.
[10] I. F. Akyildiz, A. Lee, P. Wang, M. Luo, and W. Chou, “A roadmap Switching, San Diego, CA, USA, 2014, p. PM3C.4.
for traffic engineering in SDN-OpenFlow networks,” Comput. Netw., [36] N. Cvijetic, “Optical network evolution for 5G mobile applications and
vol. 71, pp. 1–30, Oct. 2014. SDN-based control,” in Proc. Int. Telecommun. Netw. Strategy Plan.
[11] T. Chen, M. Matinmikko, X. Chen, X. Zhou, and P. Ahokangas, Symp. (Networks), Funchal, Portugal, Sep. 2014, pp. 1–5.
“Software defined mobile networks: Concept, survey, and research [37] T. Starr, J. M. Cioffi, and P. J. Silverman, Understanding Digital
directions,” IEEE Commun. Mag., vol. 53, no. 11, pp. 126–133, Subscriber Line Technology. Upper Saddle River, NJ, USA: Prentice
Nov. 2015. Hall, 1999.
THYAGATURU et al.: SDONs: A COMPREHENSIVE SURVEY 2777

[38] K. J. Kerpez et al., “Software-defined access networks,” IEEE [63] D. Fellows and D. Jones, “DocsisTM cable modem technology,” IEEE
Commun. Mag., vol. 52, no. 9, pp. 152–159, Sep. 2014. Commun. Mag., vol. 39, no. 3, pp. 202–209, Mar. 2001.
[39] N. Bitar, “Software defined networking and applicability to access [64] C. Severance, “Roy T. Fielding: Understanding the REST style,”
networks,” in Proc. OFC, San Francisco, CA, USA, Mar. 2014, Computer, vol. 48, no. 6, pp. 7–9, Jun. 2015.
pp. 1–3. [65] L. Li, W. Chou, W. Zhou, and M. Luo, “Design patterns and extensi-
[40] J. R. de Almeida Amazonas, G. Santos-Boada, and J. Sole-Pareta, bility of REST API for networking applications,” IEEE Trans. Netw.
“A critical review of OpenFlow/SDN-based networks,” in Proc. Int. Service Manag., vol. 13, no. 1, pp. 154–167, Mar. 2016.
Conf. Transp. Opt. Netw. (ICTON), Graz, Austria, Jul. 2014, pp. 1–5. [66] A. Farrel, J.-P. Vasseur, and J. Ash, “A path computation ele-
[41] M. Y. Arslan, K. Sundaresan, and S. Rangarajan, “Software-defined ment (PCE)-based architecture,” Internet Requests for Comments,
networking in cellular radio access networks: Potential and challenges,” RFC 4655, Fremont, CA, USA, Aug. 2006. [Online]. Available:
IEEE Commun. Mag., vol. 53, no. 1, pp. 150–156, Jan. 2015. http://www.rfc-editor.org/rfc/rfc4655.txt
[42] C. J. Bernardos et al., “An architecture for software defined wire- [67] J. Vasseur and J. L. Roux, “Path computation element (PCE)
less networking,” IEEE Wireless Commun., vol. 21, no. 3, pp. 52–61, communication protocol (PCEP),” Internet Requests for Comments,
Jun. 2014. RFC 5440, Fremont, CA, USA, Mar. 2009. [Online]. Available:
[43] I. T. Haque and N. Abu-Ghazaleh, “Wireless software defined net- http://www.rfc-editor.org/rfc/rfc5440.txt
working: A survey and taxonomy,” IEEE Commun. Surveys Tuts., to [68] R. P. Goldberg, “Survey of virtual machine research,” Computer, vol. 7,
be published. no. 6, pp. 34–45, Sep. 1974.
[44] N. A. Jagadeesan and B. Krishnamachari, “Software-defined network- [69] F. Douglis and O. Krieger, “Virtualization,” IEEE Internet Comput.,
ing paradigms in wireless networks: A survey,” ACM Comput. Surveys, vol. 17, no. 2, pp. 6–9, Mar./Apr. 2013.
vol. 47, no. 2, pp. 1–11, 2014. [70] A. Belbekkouche, M. M. Hasan, and A. Karmouch, “Resource discov-
[45] M. R. Sama et al., “Software-defined control of the virtualized mobile ery and allocation in network virtualization,” IEEE Commun. Surveys
packet core,” IEEE Commun. Mag., vol. 53, no. 2, pp. 107–115, Tuts., vol. 14, no. 4, pp. 1114–1128, 4th Quart. 2013.
Feb. 2015. [71] Q. Duan, Y. Yan, and A. V. Vasilakos, “A survey on service-
[46] K. Sood, S. Yu, and Y. Xiang, “Software defined wireless networking oriented network virtualization toward convergence of networking and
opportunities and challenges for Internet of Things: A review,” IEEE cloud computing,” IEEE Trans. Netw. Service Manag., vol. 9, no. 4,
Internet Things J., to be published. pp. 373–392, Dec. 2012.
[47] M. Yang et al., “Software-defined and virtualized future mobile and [72] A. Fischer, J. F. Botero, M. T. Beck, H. De Meer, and X. Hesselbach,
wireless networks: A survey,” Mobile Netw. Appl., vol. 20, no. 1, “Virtual network embedding: A survey,” IEEE Commun. Surveys Tuts.,
pp. 4–18, 2015. vol. 15, no. 4, pp. 1888–1906, 4th Quart. 2013.
[48] C. Liang and F. R. Yu, “Wireless network virtualization: A survey, [73] B. Han, V. Gopalakrishnan, L. Ji, and S. Lee, “Network function
some research issues and challenges,” IEEE Commun. Surveys Tuts., virtualization: Challenges and opportunities for innovations,” IEEE
vol. 17, no. 1, pp. 358–380, 1st Quart. 2015. Commun. Mag., vol. 53, no. 2, pp. 90–97, Feb. 2015.
[74] A. Leon-Garcia and L. G. Mason, “Virtual network resource manage-
[49] I. Khan et al., “Wireless sensor network virtualization: Early archi-
ment for next-generation networks,” IEEE Commun. Mag., vol. 41,
tecture and research perspectives,” IEEE Netw., vol. 29, no. 3,
no. 7, pp. 102–109, Jul. 2003.
pp. 104–112, May/Jun. 2015.
[75] R. Mijumbi et al., “Network function virtualization: State-of-the-art
[50] V.-G. Nguyen, T.-X. Do, and Y. Kim, “SDN and virtualization-based
and research challenges,” IEEE Commun. Surveys Tuts., vol. 18, no. 1,
LTE mobile network architectures: A comprehensive survey,” Wireless
pp. 236–262, 1st Quart. 2016.
Pers. Commun., vol. 86, no. 3, pp. 1401–1438, 2016.
[76] K. Pentikousis et al., “Guest editorial: Network and service virtualiza-
[51] Huawei Technologies. (2014). SoftCOM—The Way to
tion,” IEEE Commun. Mag., vol. 53, no. 2, pp. 88–89, Feb. 2015.
Reconstruct the Future Telecom Industry. [Online]. Available:
[77] A. Wang, M. Iyer, R. Dutta, G. N. Rouskas, and I. Baldine, “Network
http://www.huawei.com/mwc2014/en/articles/hw-u_319945.htm
virtualization: Technologies, perspectives, and frontiers,” J. Lightw.
[52] M. Peng, Y. Li, Z. Zhao, and C. Wang, “System architecture and Technol., vol. 31, no. 4, pp. 523–537, Feb. 15, 2013.
key technologies for 5G heterogeneous cloud radio access networks,”
[78] A. Khan, A. Zugenmaier, D. Jurca, and W. Kellerer, “Network virtual-
arXiv:1412.6677, 2014.
ization: A hypervisor for the Internet?” IEEE Commun. Mag., vol. 50,
[53] R. Trivisonno, R. Guerzoni, I. Vaishnavi, and D. Soldani, “SDN-based no. 1, pp. 136–143, Jan. 2012.
5G mobile networks: Architecture, functions, procedures and backward [79] R. Sherwood et al., “Flowvisor: A network virtualization
compatibility,” Trans. Emerg. Telecommun. Technol., vol. 26, no. 1, layer,” Open Netw. Found., OpenFlow Switch Consortium,
pp. 82–92, 2015. Palo Alto, CA, USA, Tech. Rep. OPENFLOW-TR-2009-1,
[54] V. Yazici, U. C. Kozat, and M. O. Sunay, “A new control plane for 5G 2009. [Online]. Available: http://archive.openflow.org/downloads/
network architecture with a case study on unified handoff, mobility, and technicalreports/openflow-tr-2009-1-flowvisor.pdf
routing management,” IEEE Commun. Mag., vol. 52, no. 11, pp. 76–85, [80] A. Blenk, A. Basta, M. Reisslein, and W. Kellerer, “Survey on net-
Nov. 2014. work virtualization hypervisors for software defined networking,” IEEE
[55] E. Haleplidis et al., “Software-defined networking (SDN): Layers Commun. Surveys Tuts., vol. 18, no. 1, pp. 655–685, 1st Quart. 2016.
and architecture terminology,” Internet Requests for Comments, [81] D. Drutskoy, E. Keller, and J. Rexford, “Scalable network virtualization
RFC 7426, Fremont, CA, USA, Jan. 2015. [Online]. Available: in software-defined networks,” IEEE Internet Comput., vol. 17, no. 2,
http://www.rfc-editor.org/rfc/rfc7426.txt pp. 20–27, Mar./Apr. 2013.
[56] N. McKeown et al., “OpenFlow: Enabling innovation in campus net- [82] H. Hawilo, A. Shami, M. Mirahmadi, and R. Asal, “NFV: State of the
works,” ACM SIGCOMM Comput. Commun. Rev., vol. 38, no. 2, art, challenges, and implementation in next generation mobile networks
pp. 69–74, 2008. (vEPC),” IEEE Netw., vol. 28, no. 6, pp. 18–26, Nov./Dec. 2014.
[57] Open Networking Foundation, “SDN architecture, issue 1.1, [83] Y. Li and M. Chen, “Software-defined network function virtualization:
ONF TR-521,” Palo Alto, CA, USA, 2016. A survey,” IEEE Access, vol. 3, pp. 2542–2553, 2015.
[58] L. Yang, R. Dantu, T. Anderson, and R. Gopal, “Forwarding and [84] T. Lin, Z. Zhou, M. Tornatore, and B. Mukherjee, “Demand-aware
control element separation (ForCES) framework,” Internet Requests network function placement,” J. Lightw. Technol., vol. 34, no. 11,
for Comments, RFC 3746, Fremont, CA, USA, Apr. 2004. [Online]. pp. 2590–2600, Jun. 1, 2016.
Available: http://www.rfc-editor.org/rfc/rfc3746.txt [85] J. Matias, J. Garay, N. Toledo, J. Unzilla, and E. Jacob, “Toward an
[59] P. Bosshart et al., “P4: Programming protocol-independent packet pro- SDN-enabled NFV architecture,” IEEE Commun. Mag., vol. 53, no. 4,
cessors,” ACM SIGCOMM Comput. Commun. Rev., vol. 44, no. 3, pp. 187–193, Apr. 2015.
pp. 87–95, 2014. [86] Z. Ye, X. Cao, J. Wang, H. Yu, and C. Qiao, “Joint topology design
[60] H. Song, “Protocol-oblivious forwarding: Unleash the power of SDN and mapping of service function chains for efficient, scalable, and
through a future-proof forwarding plane,” in Proc. ACM SIGCOMM reliable network functions virtualization,” IEEE Netw., vol. 30, no. 3,
Workshop Hot Topics SDN, Hong Kong, 2013, pp. 127–132. pp. 81–87, May/Jun. 2016.
[61] Cisco DevNet: Onepk. (2016). [Online]. Available: [87] X. Gong, L. Guo, Y. Liu, Y. Zhou, and H. Li, “Optimization mech-
https://developer.cisco.com/site/onepk/ anisms in multi-dimensional and flexible PONs: Challenging issues
[62] Cord (Central Office Re-Architected as a Datacenter): The Killer App and possible solutions,” Opt. Switching Netw., vol. 18, pp. 120–134,
for SDN & NFV. (2016). [Online]. Available: http://opencord.org/ Nov. 2015.
2778 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 18, NO. 4, FOURTH QUARTER 2016

[88] J. Jue, V. Eramo, V. López, and Z. Zhu, “Software-defined elastic [112] M. Chen, J. Wan, S. González, X. Liao, and V. C. M. Leung, “A sur-
optical networks,” Photon. Netw. Commun., vol. 28, no. 1, pp. 19–33, vey of recent developments in home M2M networks,” IEEE Commun.
2014. Surveys Tuts., vol. 16, no. 1, pp. 98–114, 1st Quart. 2014.
[89] I. Tomkos, S. Azodolmolky, J. Solé-Pareta, D. Careglio, and [113] F. Capozzi, G. Piro, L. A. Grieco, G. Boggia, and P. Camarda,
E. Palkopoulou, “A tutorial on the flexible optical networking “Downlink packet scheduling in LTE cellular networks: Key design
paradigm: State of the art, trends, and research challenges,” Proc. IEEE, issues and a survey,” IEEE Commun. Surveys Tuts., vol. 15, no. 2,
vol. 102, no. 9, pp. 1317–1337, Sep. 2014. pp. 678–700, 2nd Quart. 2013.
[90] B. C. Chatterjee, N. Sarma, and E. Oki, “Routing and spectrum allo- [114] A. Damnjanovic et al., “A survey on 3GPP heterogeneous networks,”
cation in elastic optical networks: A tutorial,” IEEE Commun. Surveys IEEE Wireless Commun., vol. 18, no. 3, pp. 10–21, Jun. 2011.
Tuts., vol. 17, no. 3, pp. 1776–1800, 3rd Quart. 2015. [115] S. Schwarz et al., “Pushing the limits of LTE: A survey on research
[91] S. Talebi et al., “Spectrum management techniques for elastic opti- enhancing the standard,” IEEE Access, vol. 1, pp. 51–62, 2013.
cal networks: A survey,” Opt. Switching Netw., vol. 13, pp. 34–48, [116] J. Li, M. Peng, A. Cheng, Y. Yu, and C. Wang, “Resource allocation
Jul. 2014. optimization for delay-sensitive traffic in fronthaul constrained cloud
[92] X. Yu et al., “Spectrum engineering in flexible grid data center optical radio access networks,” IEEE Syst. J., to be published.
networks,” Opt. Switching Netw., vol. 14, pp. 282–288, Aug. 2014. [117] C. Liu et al., “A novel multi-service small-cell cloud radio access net-
[93] O. Gerstel, M. Jinno, A. Lord, and S. J. B. Yoo, “Elastic optical net- work for mobile backhaul and computing based on radio-over-fiber
working: A new dawn for the optical layer?” IEEE Commun. Mag., technologies,” J. Lightw. Technol., vol. 31, no. 17, pp. 2869–2875,
vol. 50, no. 2, pp. S12–S20, Feb. 2012. Sep. 1, 2013.
[94] R. Ramaswami, K. Sivarajan, and G. Sasaki, Optical Networks:
[118] S.-H. Park, O. Simeone, O. Sahin, and S. S. Shitz, “Fronthaul com-
A Practical Perspective. Burlington, MA, USA: Morgan Kaufmann,
pression for cloud radio access networks: Signal processing advances
2009.
inspired by network information theory,” IEEE Signal Process. Mag.,
[95] J. M. Simmons, Optical Network Design and Planning. Cham,
vol. 31, no. 6, pp. 69–79, Nov. 2014.
Switzerland: Springer, 2014.
[96] M. Forzati et al., “Next-generation optical access seamless evolution: [119] M. Peng, C. Wang, V. Lau, and H. V. Poor, “Fronthaul-constrained
Concluding results of the European FP7 project OASE,” IEEE/OSA J. cloud radio access networks: Insights and challenges,” IEEE Wireless
Opt. Commun. Netw., vol. 7, no. 2, pp. 109–123, Feb. 2015. Commun., vol. 22, no. 2, pp. 152–160, Apr. 2015.
[97] M. Hajduczenia, H. J. da Silva, and P. P. Monteiro, “EPON versus [120] H. Raza, “A brief survey of radio access network backhaul evolution:
APON and GPON: A detailed performance comparison,” J. Opt. Netw., Part II,” IEEE Commun. Mag., vol. 51, no. 5, pp. 170–177, May 2013.
vol. 5, no. 4, pp. 298–319, 2006. [121] O. Tipmongkolsilp, S. Zaghloul, and A. Jukan, “The evolution of cel-
[98] B. Skubic, J. Chen, J. Ahmed, L. Wosinska, and B. Mukherjee, lular backhaul technologies: Current issues and future trends,” IEEE
“A comparison of dynamic bandwidth allocation for EPON, GPON, Commun. Surveys Tuts., vol. 13, no. 1, pp. 97–113, 1st Quart. 2011.
and next-generation TDM PON,” IEEE Commun. Mag., vol. 47, no. 3, [122] M. Yang et al., “OpenRAN: A software-defined RAN architecture via
pp. S40–S48, Mar. 2009. virtualization,” ACM SIGCOMM Comput. Commun. Rev., vol. 43, no. 4,
[99] B. Kantarci and H. T. Mouftah, “Bandwidth distribution solutions pp. 549–550, 2013.
for performance enhancement in long-reach passive optical net- [123] I. F. Akyildiz, X. Wang, and W. Wang, “Wireless mesh networks:
works,” IEEE Commun. Surveys Tuts., vol. 14, no. 3, pp. 714–733, A survey,” Comput. Netw., vol. 47, no. 4, pp. 445–487, 2005.
3rd Quart. 2012. [124] E. Alotaibi and B. Mukherjee, “A survey on routing algorithms for
[100] M. P. McGarry, M. Reisslein, F. Aurzada, and M. Scheutzow, “Shortest wireless ad-hoc and mesh networks,” Comput. Netw., vol. 56, no. 2,
propagation delay (SPD) first scheduling for EPONs with heteroge- pp. 940–965, 2012.
neous propagation delays,” IEEE J. Sel. Areas Commun., vol. 28, no. 6, [125] D. Benyamina, A. Hafid, and M. Gendreau, “Wireless mesh net-
pp. 849–862, Aug. 2010. works design: A survey,” IEEE Commun. Surveys Tuts., vol. 14, no. 2,
[101] M. P. McGarry and M. Reisslein, “Investigation of the DBA algo- pp. 299–310, 2nd Quart. 2012.
rithm design space for EPONs,” J. Lightw. Technol., vol. 30, no. 14, [126] M. S. Kuran and T. Tugcu, “A survey on emerging broadband wireless
pp. 2271–2280, Jul. 15, 2012. access technologies,” Comput. Netw., vol. 51, no. 11, pp. 3013–3046,
[102] J. Zheng and H. T. Mouftah, “A survey of dynamic bandwidth alloca- 2007.
tion algorithms for Ethernet passive optical networks,” Opt. Switching [127] P. H. Pathak and R. Dutta, “A survey of network design problems and
Netw., vol. 6, no. 3, pp. 151–162, 2009. joint design approaches in wireless mesh networks,” IEEE Commun.
[103] A. Mercian, M. P. McGarry, and M. Reisslein, “Offline and online Surveys Tuts., vol. 13, no. 3, pp. 396–428, 3rd Quart. 2011.
multi-thread polling in long-reach PONs: A critical evaluation,” [128] K. S. Vijayalayan, A. Harwood, and S. Karunasekera, “Distributed
J. Lightw. Technol., vol. 31, no. 12, pp. 2018–2028, Jun. 15, 2013. scheduling schemes for wireless mesh networks: A survey,” ACM
[104] A. Nag, D. B. Payne, and M. Ruffini, “N:1 protection design for Comput. Surveys, vol. 46, no. 1, p. 14, 2013.
minimizing OLTs in resilient dual-homed long-reach passive optical [129] Y. Cai, Y. Yan, Z. Zhang, and Y. Yang, “Survey on converged data
network,” IEEE/OSA J. Opt. Commun. Netw., vol. 8, no. 2, pp. 93–99, center networks with DCB and FCoE: Standards and protocols,” IEEE
Feb. 2016. Netw., vol. 27, no. 4, pp. 27–32, Jul./Aug. 2013.
[105] H. Song, B.-W. Kim, and B. Mukherjee, “Long-reach optical access [130] C. Kachris and I. Tomkos, “A survey on optical interconnects for data
networks: A survey of research challenges, demonstrations, and band- centers,” IEEE Commun. Surveys Tuts., vol. 14, no. 4, pp. 1021–1036,
width assignment mechanisms,” IEEE Commun. Surveys Tuts., vol. 12, 4th Quart. 2012.
no. 1, pp. 112–123, 1st Quart. 2010.
[131] P. Samadi, K. Wen, J. Xu, and K. Bergman, “Software-defined optical
[106] F. Aurzada, M. Lévesque, M. Maier, and M. Reisslein, “FiWi access
network for metro-scale geographically distributed data centers,” Opt.
networks based on next-generation PON and gigabit-class WLAN tech-
Exp., vol. 24, no. 11, pp. 12310–12320, May 2016.
nologies: A capacity and delay analysis,” IEEE/ACM Trans. Netw.,
vol. 22, no. 4, pp. 1176–1189, Aug. 2014. [132] H. Yang et al., “SUDOI: Software defined networking for ubiquitous
[107] N. Ghazisaidi and M. Maier, “Fiber-wireless (FiWi) access networks: data center optical interconnection,” IEEE Commun. Mag., vol. 54,
Challenges and opportunities,” IEEE Netw., vol. 25, no. 1, pp. 36–42, no. 2, pp. 86–95, Feb. 2016.
Jan./Feb. 2011. [133] J. Zhang, F. Ren, and C. Lin, “Survey on transport control in data center
[108] J. Liu et al., “New perspectives on future smart FiWi networks: networks,” IEEE Netw., vol. 27, no. 4, pp. 22–26, Jul./Aug. 2013.
Scalability, reliability and energy efficiency,” IEEE Commun. Surveys [134] A. Bianco, T. Bonald, D. Cuda, and R.-M. Indre, “Cost, power con-
Tuts., vol. 18, no. 2, pp. 1045–1072, 2nd Quart. 2016. sumption and performance evaluation of metro networks,” IEEE/OSA
[109] A. G. Sarigiannidis et al., “Architectures and bandwidth allocation J. Opt. Commun. Netw., vol. 5, no. 1, pp. 81–91, Jan. 2013.
schemes for hybrid wireless-optical networks,” IEEE Commun. Surveys [135] A. Bianco, D. Cuda, and J. M. Finochietto, “Short-term fairness in
Tuts., vol. 17, no. 1, pp. 427–468, 1st Quart. 2015. slotted WDM rings,” Comput. Netw., vol. 83, pp. 235–248, Jun. 2015.
[110] T. Tsagklas and F. Pavlidou, “A survey on radio-and-fiber FiWi net- [136] I.-F. Chao and M. C. Yuang, “Toward wireless backhaul using circuit
work architectures,” J. Select. Areas Telecommun., vol. 1, pp. 18–24, emulation over optical packet-switched metro WDM ring network,”
Mar. 2011. J. Lightw. Technol., vol. 31, no. 18, pp. 3032–3042, Sep. 15, 2013.
[111] E. I. Gurrola, M. P. McGarry, Y. Luo, and F. Effenberger, “PON/xDSL [137] M. Maier, M. Reisslein, and A. Wolisz, “A hybrid MAC protocol for a
hybrid access networks,” Opt. Switching Netw., vol. 14, pp. 32–42, metro WDM network using multiple free spectral ranges of an arrayed-
Aug. 2014. waveguide grating,” Comput. Netw., vol. 41, no. 4, pp. 407–433, 2003.
THYAGATURU et al.: SDONs: A COMPREHENSIVE SURVEY 2779

[138] C. Rottondi, M. Tornatore, A. Pattavina, and G. Gavioli, “Routing, [160] Y. Ou, A. Hammad, S. Peng, R. Nejabati, and D. Simeonidou, “Online
modulation level, and spectrum assignment in optical metro ring net- and offline virtualization of optical transceiver,” IEEE/OSA J. Opt.
works using elastic transceivers,” IEEE/OSA J. Opt. Commun. Netw., Commun. Netw., vol. 7, no. 8, pp. 748–760, Aug. 2015.
vol. 5, no. 4, pp. 305–315, Apr. 2013. [161] C. Matrakidis, T. G. Orphanoudakis, A. Stavdas, J. P. F.-P. Giménez,
[139] M. Scheutzow, M. Maier, M. Reisslein, and A. Wolisz, “Wavelength and A. Manzalini, “HYDRA: A scalable ultra long reach/high capacity
reuse for efficient packet-switched transport in an AWG-based metro access network architecture featuring lower cost and power con-
WDM network,” J. Lightw. Technol., vol. 21, no. 6, pp. 1435–1455, sumption,” J. Lightw. Technol., vol. 33, no. 2, pp. 339–348, Jan. 15,
Jun. 2003. 2015.
[140] A. Autenrieth et al., “Evaluation of technology options for software- [162] N. Amaya et al., “Fully-elastic multi-granular network with
defined transceivers in fixed WDM grid versus flexible WDM grid space/frequency/time switching using multi-core fibres and pro-
optical transport networks,” in Proc. VDE ITG Symp. Photonic Netw., grammable optical nodes,” Opt. Exp., vol. 21, no. 7, pp. 8865–8872,
Leipzig, Germany, 2013, pp. 1–5. 2013.
[141] J.-P. Elbers and A. Autenrieth, “From static to software-defined optical [163] N. Amaya et al., “Software defined networking (SDN) over space
networks,” in Proc. IEEE Int. Conf. Opt. Netw. Design Model. (ONDM), division multiplexing (SDM) optical networks: Features, benefits and
Colchester, U.K., 2012, pp. 1–4. experimental demonstration,” Opt. Exp., vol. 22, no. 3, pp. 3638–3647,
[142] S. Gringeri, B. Basch, V. Shukla, R. Egorov, and T. J. Xia, “Flexible 2014.
architectures for optical transport nodes and networks,” IEEE Commun. [164] J. M. Galve, I. Gasulla, S. Sales, and J. Capmany, “Reconfigurable radio
Mag., vol. 48, no. 7, pp. 40–50, Jul. 2010. access networks using multicore fibers,” IEEE J. Quantum Electron.,
[143] H. Y. Choi, L. Liu, T. Tsuritani, and I. Morita, “Demonstration of vol. 52, no. 1, pp. 1–7, Jan. 2016.
BER-adaptive WSON employing flexible transmitter/receiver with an [165] B. Collings, “New devices enabling software-defined optical networks,”
extended OpenFlow-based control plane,” IEEE Photon. Technol. Lett., IEEE Commun. Mag., vol. 51, no. 3, pp. 66–71, Mar. 2013.
vol. 25, no. 2, pp. 119–121, Jan. 15, 2013. [166] N. Amaya, G. Zervas, and D. Simeonidou, “Introducing node archi-
[144] L. Liu et al., “Demonstration of a dynamic transparent optical network tecture flexibility for elastic optical networks,” IEEE/OSA J. Opt.
employing flexible transmitters/receivers controlled by an OpenFlow- Commun. Netw., vol. 5, no. 6, pp. 593–608, Jun. 2013.
stateless PCE integrated control plane [invited],” IEEE/OSA J. Opt. [167] B. R. Rofoee, G. Zervas, Y. Yan, N. Amaya, and D. Simeonidou,
Commun. Netw., vol. 5, no. 10, pp. A66–A75, Oct. 2013. “All programmable and synthetic optical network: Architecture and
[145] J. A. Lazaro et al., “Flexible PON key technologies: Digital advanced implementation,” IEEE/OSA J. Opt. Commun. Netw., vol. 5, no. 9,
modulation formats and devices,” in Proc. Int. Conf. Transp. Opt. pp. 1096–1110, Sep. 2013.
Netw. (ICTON), Graz, Austria, Jul. 2014, pp. 1–5. [168] R. Younce, J. Larikova, and Y. Wang, “Engineering 400G for colorless-
[146] N. Iiyama, S.-Y. Kim, T. Shimada, S. Kimura, and N. Yoshimoto, directionless-contentionless architecture in metro/regional networks
“Co-existent downstream scheme between OOK and QAM signals in [invited],” IEEE/OSA J. Opt. Commun. Netw., vol. 5, no. 10,
an optical access network using software-defined technology,” in Proc. pp. A267–A273, Oct. 2013.
OFC/NFOEC, Los Angeles, CA, USA, Mar. 2012, pp. 1–3. [169] W. I. Way, P. N. Ji, and A. N. Patel, “Wavelength contention-free via
[147] C.-H. Yeh et al., “Using OOK modulation for symmetric 40-Gb/s long- optical bypass within a colorless and directionless ROADM [invited],”
reach time-sharing passive optical networks,” IEEE Photon. Technol. IEEE/OSA J. Opt. Commun. Netw., vol. 5, no. 10, pp. A220–A229,
Lett., vol. 22, no. 9, pp. 619–621, May 1, 2010. Oct. 2013.
[148] J. Yu, M.-F. Huang, D. Qian, L. Chen, and G.-K. Chang, “Centralized
[170] M. Garrich et al., “Experimental demonstration of function pro-
lightwave WDM-PON employing 16-QAM intensity modulated OFDM
grammable add/drop architecture for ROADMs [invited],” IEEE/OSA
downstream and OOK modulated upstream signals,” IEEE Photon.
J. Opt. Commun. Netw., vol. 7, no. 2, pp. A335–A343, Feb. 2015.
Technol. Lett., vol. 20, no. 18, pp. 1545–1547, Sep. 15, 2008.
[171] A. Sadasivarao et al., “Open transport switch: A software defined net-
[149] F. Vacondio et al., “Flexible TDMA access optical networks enabled
working architecture for transport networks,” in Proc. ACM SIGCOMM
by burst-mode software defined coherent transponders,” in Proc. 39th
Workshop Hot Topics Softw. Defined Netw., Hong Kong, 2013,
Eur. Conf. Exhibit. Opt. Commun. (ECOC), London, U.K., Sep. 2013,
pp. 115–120.
pp. 1–3.
[150] M. Bolea, R. P. Giddings, and J. M. Tang, “Digital orthogo- [172] A. Panda, S. Shenker, M. McCauley, T. Koponen, and M. Casado,
nal filter-enabled optical OFDM channel multiplexing for software- “Extending SDN to large-scale networks,” in Proc. ONS Res. Track,
reconfigurable elastic PONs,” J. Lightw. Technol., vol. 32, no. 6, Apr. 2013, pp. 1–2.
pp. 1200–1206, Mar. 15, 2014. [173] R. Nejabati, S. Peng, B. Gou, M. Channegowda, and D. Simeonidou,
[151] M. Bolea, R. P. Giddings, M. Bouich, C. Aupetit-Berthelemot, and “Toward a completely softwareized optical network [invited],”
J. M. Tang, “Digital filter multiple access PONs with DSP-enabled IEEE/OSA J. Opt. Commun. Netw., vol. 7, no. 12, pp. B222–B231,
software reconfigurability,” IEEE/OSA J. Opt. Commun. Netw., vol. 7, Dec. 2015.
no. 4, pp. 215–222, Apr. 2015. [174] S. S. W. Lee, K.-Y. Li, and M.-S. Wu, “Design and implementation
[152] A. Malacarne et al., “Multiplexing of asynchronous and independent of a GPON-based virtual OpenFlow-enabled SDN switch,” J. Lightw.
ASK and PSK transmissions in SDN-controlled intra-data center net- Technol., vol. 34, no. 10, pp. 2552–2561, May 15, 2016.
work,” J. Lightw. Technol., vol. 32, no. 9, pp. 1794–1800, May 1, [175] R. Gu, Y. Ji, P. Wei, and S. Zhang, “Software defined flexible and
2014. efficient passive optical networks for intra-datacenter communications,”
[153] M. Jinno, H. Takara, Y. Sone, K. Yonenaga, and A. Hirano, “Multiflow Opt. Switching Netw., vol. 14, pp. 289–302, Aug. 2014.
optical transponder for efficient multilayer optical networking,” IEEE [176] A. Amokrane, J. Hwang, J. Xiao, and N. Anerousis, “Software defined
Commun. Mag., vol. 50, no. 5, pp. 56–65, May 2012. enterprise passive optical network,” in Proc. IEEE Int. Conf. Netw.
[154] N. Sambo et al., “Programmable transponder, code and differentiated Service Manag., Rio de Janeiro, Brazil, Nov. 2014, pp. 406–411.
filter configuration in elastic optical networks,” J. Lightw. Technol., [177] A. Amokrane, J. Xiao, J. Hwang, and N. Anerousis, “Dynamic capacity
vol. 32, no. 11, pp. 2079–2086, Jun. 1, 2014. management and traffic steering in enterprise passive optical networks,”
[155] N. Sambo et al., “Next generation sliceable bandwidth variable in Proc. IFIP/IEEE Int. Symp. Integr. Netw. Manag. (IM), Ottawa, ON,
transponders,” IEEE Commun. Mag., vol. 53, no. 2, pp. 163–171, Canada, 2015, pp. 406–413.
Feb. 2015. [178] C.-H. Yeh, C.-W. Chow, M.-H. Yang, and D.-Z. Hsu, “A flexible and
[156] N. Sambo et al., “Sliceable transponder architecture including mul- reliable 40-Gb/s OFDM downstream TWDM-PON architecture,” IEEE
tiwavelength source,” IEEE J. Opt. Commun. Netw., vol. 6, no. 7, Photon. J., vol. 7, no. 6, pp. 1–9, Dec. 2015.
pp. 590–600, Jul. 2014. [179] R. Gu, S. Zhang, Y. Ji, T. Guo, and X. Wang, “Efficient software-
[157] F. Cugini et al., “Toward plug-and-play software-defined elastic optical defined passive optical network with network coding,” Photon. Netw.
networks,” J. Lightw. Technol., vol. 34, no. 6, pp. 1494–1500, Mar. 15, Commun., vol. 31, no. 2, pp. 239–250, 2016.
2016. [180] M. Forzati and A. Gavler, “Flexible next-generation optical access,”
[158] M. S. Moreolo et al., “SDN-enabled sliceable BVT based on mul- in Proc. Int. Conf. Transp. Opt. Netw. (ICTON), Cartagena, Spain,
ticarrier technology for multiflow rate/distance and grid adaptation,” Jun. 2013, pp. 1–8.
J. Lightw. Technol., vol. 34, no. 6, pp. 1516–1522, Mar. 15, 2016. [181] K. Kondepu, A. Sgambelluri, L. Valcarenghi, F. Cugini, and P. Castoldi,
[159] Y. Ou et al., “Demonstration of virtualizeable and software-defined “An SDN-based integration of green TWDM-PONs and metro net-
optical transceiver,” J. Lightw. Technol., vol. 34, no. 8, pp. 1916–1924, works preserving end-to-end delay,” in Proc. OSA OFC, Los Angeles,
Apr. 15, 2016. CA, USA, 2015, pp. 1–3.
2780 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 18, NO. 4, FOURTH QUARTER 2016

[182] S. Das, G. Parulkar, and N. McKeown, “Unifying packet and circuit [205] S. Gringeri, N. Bitar, and T. J. Xia, “Extending software defined net-
switched networks,” in Proc. IEEE GLOBECOM Workshops, Honolulu, work principles to include optical transport,” IEEE Commun. Mag.,
HI, USA, Nov. 2009, pp. 1–6. vol. 51, no. 3, pp. 32–40, Mar. 2013.
[183] S. Das et al., “Packet and circuit network convergence with OpenFlow,” [206] M. Jinno, H. Takara, K. Yonenaga, and A. Hirano, “Virtualization
in Proc. OFC/NFOEC, San Diego, CA, USA, Mar. 2010, pp. 1–3. in optical networks from network level to hardware level [invited],”
[184] R. Veisllari, S. Bjornstad, and K. Bozorgebrahimi, “Integrated IEEE/OSA J. Opt. Commun. Netw., vol. 5, no. 10, pp. A46–A56,
packet/circuit hybrid network field trial with production traffic Oct. 2013.
[invited],” IEEE/OSA J. Opt. Commun. Netw., vol. 5, no. 10, [207] G. N. Rouskas, R. Dutta, and I. Baldine, “A new Internet architecture to
pp. A257–A266, Oct. 2013. enable software defined optics and evolving optical switching models,”
[185] S. Azodolmolky et al., “Integrated OpenFlow–GMPLS control plane: in Proc. Int. Conf. Broadband Commun. Netw. Syst. (BROADNETS),
An overlay model for software defined packet over optical networks,” London, U.K., Sep. 2008, pp. 71–76.
Opt. Exp., vol. 19, no. 26, pp. B421–B428, Dec. 2011. [208] D. Hillerkuss and J. Leuthold, “Software-defined transceivers in
[186] M. Shirazipour, W. John, J. Kempf, H. Green, and M. Tatipamula, dynamic access networks,” J. Lightw. Technol., vol. 34, no. 2,
“Realizing packet-optical integration with SDN and OpenFlow 1.1 pp. 792–797, Jan. 15, 2016.
extensions,” in Proc. IEEE ICC, Ottawa, ON, Canada, Jun. 2012, [209] P. J. Winzer and R.-J. Essiambre, “Advanced optical modulation
pp. 6633–6637. formats,” Proc. IEEE, vol. 94, no. 5, pp. 952–985, May 2006.
[210] A. P. T. Lau et al., “Advanced DSP techniques enabling high spectral
[187] W. Cerroni, G. Leli, and C. Raffaelli, “Design and test of a software
efficiency and flexible transmissions: Toward elastic optical networks,”
defined hybrid network architecture,” in Proc. Workshop High Perform.
IEEE Signal Process. Mag., vol. 31, no. 2, pp. 82–92, Mar. 2014.
Program. Netw. (HPPN), New York, NY, USA, 2013, pp. 1–8.
[211] R. Schmogrow et al., “Real-time software-defined multiformat trans-
[188] S. Yin et al., “UltraFlow access testbed: Experimental exploration of
mitter generating 64QAM at 28 GBd,” IEEE Photon. Technol. Lett.,
dual-mode access networks,” IEEE/OSA J. Opt. Commun. Netw., vol. 5,
vol. 22, no. 21, pp. 1601–1603, Nov. 1, 2010.
no. 12, pp. 1361–1372, Dec. 2013.
[212] N. Yoshimoto, J.-I. Kani, S.-Y. Kim, N. Iiyama, and J. Terada, “DSP-
[189] T. S. R. Shen, S. Yin, A. R. Dhaini, and L. G. Kazovsky, based optical access approaches for enhancing NG-PON2 systems,”
“Reconfigurable long-reach UltraFlow access network: A flexible, cost- IEEE Commun. Mag., vol. 51, no. 3, pp. 58–64, Mar. 2013.
effective, and energy-efficient solution,” J. Lightw. Technol., vol. 32, [213] J. Barton, E. Skogen, M. Mašanović, S. Denbaars, and L. Coldren,
no. 13, pp. 2353–2363, Jul. 1, 2014. “A widely tunable high-speed transmitter using an integrated SGDBR
[190] S. Yin et al., “A novel quasi-passive, software-defined, and energy effi- laser-semiconductor optical amplifier and Mach-Zehnder modulator,”
cient optical access network for adaptive intra-PON flow transmission,” IEEE J. Sel. Topics Quantum Electron., vol. 9, no. 5, pp. 1113–1117,
J. Lightw. Technol., vol. 33, no. 22, pp. 4536–4546, Nov. 15, 2015. Sep./Oct. 2003.
[191] N. Cvijetic, “Software-defined optical access networks for multiple [214] H. Y. Choi, T. Tsuritani, and I. Morita, “BER-adaptive flexible-format
broadband access solutions,” in Proc. Int. Conf. Photon. Switching transmitter for elastic optical networks,” Opt. Exp., vol. 20, no. 17,
(OECC/PS) OptoElectron. Commun. Conf. Held Jointly, Kyoto, Japan, pp. 18652–18658, 2012.
Jun. 2013, pp. 1–2. [215] R. Rodes et al., “Carrierless amplitude phase modulation of VCSEL
[192] N. Cvijetic et al., “SDN and OpenFlow for dynamic flex-grid optical with 4 bit/s/Hz spectral efficiency for use in WDM-PON,” Opt. Exp.,
access and aggregation networks,” J. Lightw. Technol., vol. 32, no. 4, vol. 19, no. 27, pp. 26551–26556, 2011.
pp. 864–870, Feb. 15, 2014. [216] N. Bonello, S. Chen, and L. Hanzo, “Low-density parity-check codes
[193] J. Oliveira et al., “Experimental testbed of reconfigurable flexgrid opti- and their rateless relatives,” IEEE Commun. Surveys Tuts., vol. 13,
cal network with virtualized GMPLS control plane and autonomic no. 1, pp. 3–26, 1st Quart. 2011.
controls towards SDN,” in Proc. SBMO/IEEE MTT-S Int. Microwave [217] N. Sambo et al., “Control of frequency conversion and defragmentation
Optoelectron. Conf. (IMOC), Rio de Janeiro, Brazil, Aug. 2013, for super-channels [invited],” IEEE/OSA J. Opt. Commun. Netw., vol. 7,
pp. 1–5. no. 1, pp. A126–A134, Jan. 2015.
[194] Y. Zhao, J. Zhang, H. Yang, and X. Yu, “Data center optical networks [218] M. S. Rasras et al., “Demonstration of a tunable microwave-photonic
(DCON) with OpenFlow based software defined networking (SDN),” notch filter using low-loss silicon ring resonators,” J. Lightw. Technol.,
in Proc. CHINACOM, Guilin, China, Aug. 2013, pp. 771–775. vol. 27, no. 12, pp. 2105–2110, Jun. 15, 2009.
[195] I. de Miguel et al., “Cognitive dynamic optical networks [invited],” [219] K. Wang et al., “Migration strategies for FTTx solutions based on active
IEEE/OSA J. Opt. Commun. Netw., vol. 5, no. 10, pp. A107–A118, optical networks,” IEEE Commun. Mag., vol. 54, no. 2, pp. 78–85,
Oct. 2013. Feb. 2016.
[196] A. Caballero et al., “Cognitive, heterogeneous and reconfigurable opti- [220] D. J. Richardson, J. M. Fini, and L. E. Nelson, “Space-division mul-
cal networks: The CHRON project,” J. Lightw. Technol., vol. 32, no. 13, tiplexing in optical fibres,” Nat. Photon., vol. 7, no. 5, pp. 354–362,
pp. 2308–2323, Jul. 1, 2014. 2013.
[197] R. J. Durán et al., “Experimental assessment of a cognitive mechanism [221] J. He et al., “A survey on recent advances in optical communications,”
to reduce the impact of outdated TEDs in optical networks,” Photon. Comput. Elect. Eng., vol. 40, no. 1, pp. 216–240, 2014.
Netw. Commun., vol. 31, no. 2, pp. 259–271, 2016. [222] Y. Wang and X. Cao, “Multi-granular optical switching: A classified
overview for the past and future,” IEEE Commun. Surveys Tuts., vol. 14,
[198] J. Oliveira et al., “Toward terabit autonomic optical networks based on
no. 3, pp. 698–713, 3rd Quart. 2012.
a software defined adaptive/cognitive approach [invited],” IEEE/OSA
[223] Y. Fu et al., “Orion: A hybrid hierarchical control plane of software-
J. Opt. Commun. Netw., vol. 7, no. 3, pp. A421–A431, Mar. 2015.
defined networking for large-scale networks,” in Proc. IEEE Int. Conf.
[199] A. D. Giglio et al., “Cross-layer, dynamic network orchestration, lever-
Netw. Protocols (ICNP), Raleigh, NC, USA, 2014, pp. 569–576.
aging software-defined optical performance monitors,” in Proc. IEEE
[224] R. Bassoli, H. Marques, J. Rodriguez, K. W. Shum, and R. Tafazolli,
Fotonica AEIT Ital. Conf. Photon. Technol., Turin, Italy, May 2015,
“Network coding theory: A survey,” IEEE Commun. Surveys Tuts.,
pp. 1–4.
vol. 15, no. 4, pp. 1950–1978, 4th Quart. 2013.
[200] U. Moura et al., “SDN-enabled EDFA gain adjustment cognitive [225] V. W. S. Chan, “Optical flow switching networks,” Proc. IEEE,
methodology for dynamic optical networks,” in Proc. IEEE ECOC, vol. 100, no. 5, pp. 1079–1091, May 2012.
Valencia, Spain, Sep. 2015, pp. 1–3. [226] G. Shen, H. Guo, and S. K. Bose, “Survivable elastic optical net-
[201] U. Moura et al., “Cognitive methodology for optical amplifier gain works: Survey and perspective,” Photon. Netw. Commun., vol. 31, no. 1,
adjustment in dynamic DWDM networks,” J. Lightw. Technol., vol. 34, pp. 71–87, 2016.
no. 8, pp. 1971–1979, Apr. 15, 2016. [227] J. Schonwalder, M. Bjorklund, and P. Shafer, “Network configuration
[202] F. Paolucci et al., “Superfilter technique in SDN-controlled elastic opti- management using NETCONF and YANG,” IEEE Commun. Mag.,
cal networks [invited],” IEEE/OSA J. Opt. Commun. Netw., vol. 7, no. 2, vol. 48, no. 9, pp. 166–173, Sep. 2010.
pp. A285–A292, Feb. 2015. [228] N. Gude et al., “NOX: Towards an operating system for networks,”
[203] H. Carvalho et al., “WSS/EDFA-based optimization strategies for soft- ACM SIGCOMM Comput. Commun. Rev., vol. 38, no. 3, pp. 105–110,
ware defined optical networks,” in Proc. IEEE Microw. Optoelectron. 2008.
Conf., Nov. 2015, pp. 1–5. [229] Y. Zhao, J. Zhang, L. Gao, and H. Yang, “Unified control system for
[204] X. Wang, Y. Fei, M. Razo, A. Fumagalli, and M. Garrich, “Network- heterogeneous networks with software defined networking (SDN),” in
wide signal power control strategies in WDM networks,” in Proc. IEEE Proc. Int. ICST Conf. Commun. Netw. China (CHINACOM), Guilin,
Int. Conf. ONDM, Pisa, Italy, May 2015, pp. 218–221. China, 2013, pp. 781–784.
THYAGATURU et al.: SDONs: A COMPREHENSIVE SURVEY 2781

[230] D. Dahan, U. Mahlab, A. Teixeira, I. Zacharopoulos, and I. Tomkos, [253] Y. Lee and Y. Kim, “A design of 10 gigabit capable passive optical
“Optical performance monitoring for translucent/transparent optical network (XG-PON1) architecture based on software defined net-
networks,” IET Optoelectron., vol. 5, no. 1, pp. 1–18, Feb. 2011. work (SDN),” in Proc. IEEE Int. Conf. Inf. Netw. (ICOIN), Siem Reap,
[231] Q. Sui, A. P. T. Lau, and C. Lu, “OSNR monitoring in the presence Cambodia, 2015, pp. 402–404.
of first-order PMD using polarization diversity and DSP,” J. Lightw. [254] P. Parol and M. Pawlowski, “Towards networks of the future: SDN
Technol., vol. 28, no. 15, pp. 2105–2114, Aug. 1, 2010. paradigm introduction to PON networking for business applications,”
[232] J. Schröder et al., “OSNR monitoring of a 1.28 Tbaud signal by inter- in Proc. Federated Conf. Comput. Sci. Inf. Syst. (FedCSIS), Kraków,
ferometry inside a wavelength-selective switch,” J. Lightw. Technol., Poland, Sep. 2013, pp. 829–836.
vol. 29, no. 10, pp. 1542–1546, May 15, 2011. [255] P. Parol and M. Pawlowski, “Future proof access networks for B2B
[233] T. Saida et al., “In-band OSNR monitor with high-speed integrated applications,” Informatica, vol. 38, no. 3, pp. 193–204, 2014.
Stokes polarimeter for polarization division multiplexed signal,” OSA [256] B. Yan, J. Zhou, J. Wu, and Y. Zhao, “Poster: SDN based energy
Opt. Exp., vol. 20, no. 26, pp. B165–B170, 2012. management system for optical access network,” in Proc. CHINACOM,
[234] D. R. Zimmerman and L. H. Spiekman, “Amplifiers for the masses: Maoming, China, Aug. 2014, pp. 658–659.
EDFA, EDWA, and SOA amplets for metro and access applications,” [257] K. Kanonakis, N. Cvijetic, I. Tomkos, and T. Wang, “Dynamic
J. Lightw. Technol., vol. 22, no. 1, pp. 63–70, Jan. 2004. software-defined resource optimization in next-generation optical
[235] V. Fuentes et al., “Integrating complex legacy systems under OpenFlow access enabled by OFDMA-based meta-MAC provisioning,” J. Lightw.
control: The DOCSIS use case,” in Proc. IEEE Eur. Workshop Technol., vol. 31, no. 14, pp. 2296–2306, Jul. 15, 2013.
SDN (EWSDN), Budapest, Hungary, Sep. 2014, pp. 37–42. [258] R. Muñoz, R. Casellas, R. Vilalta, and R. Martínez, “Dynamic and
[236] Y. Luo, “Activities, drivers and benefits of extending PON over other adaptive control plane solutions for flexi-grid optical networks based
media,” in Proc. OSA NFOEC, Anaheim, CA, USA, 2013, pp. 1–3. on stateful PCE,” J. Lightw. Technol., vol. 32, no. 16, pp. 2703–2715,
[237] X. Yu et al., “Spectrum defragmentation implementation based on soft- Aug. 15, 2014.
ware defined networking (SDN) in flexi-grid optical networks,” in Proc. [259] Z. Zhu et al., “OpenFlow-assisted online defragmentation in single-
ICNC, Honolulu, HI, USA, Feb. 2014, pp. 502–505. /multi-domain software-defined elastic optical networks [invited],”
[238] C. Chen et al., “Demonstrations of efficient online spectrum defrag- IEEE/OSA J. Opt. Commun. Netw., vol. 7, no. 1, pp. A7–A15,
mentation in software-defined elastic optical networks,” J. Lightw. Jan. 2015.
Technol., vol. 32, no. 24, pp. 4701–4711, Dec. 15, 2014. [260] G. Meloni et al., “Software-defined defragmentation in space-division
[239] P. N. Ji et al., “Demonstration of OpenFlow-enabled traffic and network multiplexing with quasi-hitless fast core switching,” J. Lightw. Technol.,
adaptive transport SDN,” in Proc. IEEE Int. Conf. OFC, San Francisco, vol. 34, no. 8, pp. 1956–1962, Apr. 15, 2016.
CA, USA, Mar. 2014, pp. 1–3. [261] J. Wu, Y. Zhao, J. Zhang, J. Zhou, and J. Sun, “Global dynamic band-
[240] S. Das, G. Parulkar, and N. McKeown, “Simple unified control for width optimization for software defined optical access and aggregation
packet and circuit networks,” in Proc. IEEE Photon. Soc. Summer networks,” in Proc. IEEE Int. Conf. Opt. Commun. Netw. (ICOCN),
Topical Future Glob. Netw., Newport Beach, CA, USA, Jul. 2009, Suzhou, China, 2014, pp. 1–4.
pp. 1–2. [262] Y. Zhao, B. Yan, J. Wu, and J. Zhang, “Software-defined dynamic band-
[241] Open Networking Foundation, “Addendum to OpenFlow protocol spec- width optimization (SD-DBO) algorithm for optical access and aggre-
ification (v1.0)—Circuit switch addendum v0.3,” Palo Alto, CA, USA, gation networks,” Photon. Netw. Commun., vol. 31, no. 2, pp. 251–258,
Jun. 2010. Apr. 2016.
[242] M. Channegowda et al., “Experimental demonstration of an OpenFlow [263] D. Bojic et al., “Advanced wireless and optical technologies for small-
based software-defined optical network employing packet, fixed and cell mobile backhaul with dynamic software-defined management,”
flexible DWDM grid technologies on an international multi-domain IEEE Commun. Mag., vol. 51, no. 9, pp. 86–93, Sep. 2013.
testbed,” Opt. Exp., vol. 21, no. 5, pp. 5487–5498, Mar. 2013. [264] A. Tanaka and N. Cvijetic, “Software defined flexible optical access
[243] S. Baik, C. Hwang, and Y. Lee, “SDN-based architecture for end-to-end networks enabling throughput optimization and OFDM-based dynamic
path provisioning in the mixed circuit and packet network environ- service provisioning for future mobile backhaul,” IEICE Trans.
ment,” in Proc. IEEE Asia Pac. Netw. Oper. Manag. Symp., Hsinchu, Commun., vol. E97.B, no. 7, pp. 1244–1251, 2014.
Taiwan, Sep. 2014, pp. 1–4. [265] J. Costa-Requena, J. L. Santos, and V. F. Guasch, “Mobile backhaul
[244] X. Cao et al., “Dynamic OpenFlow-controlled optical packet switching transport streamlined through SDN,” in Proc. IEEE ICTON, Budapest,
network,” J. Lightw. Technol., vol. 33, no. 8, pp. 1500–1507, Apr. 15, Hungary, Jul. 2015, pp. 1–4.
2015. [266] F. Slyne and M. Ruffini, “An SDN-driven approach to a flat
[245] H. Harai, H. Furukawa, K. Fujikawa, T. Miyazawa, and N. Wada, layer-2 telecommunications network,” in Proc. Int. Conf. Transp. Opt.
“Optical packet and circuit integrated networks and software Netw. (ICTON), Graz, Austria, Jul. 2014, pp. 1–4.
defined networking extension,” J. Lightw. Technol., vol. 32, no. 16, [267] U. Mandal et al., “Heterogeneous bandwidth provisioning for virtual
pp. 2751–2759, Aug. 15, 2014. machine migration over SDN-enabled optical networks,” in Proc. OFC,
[246] R. Alvizu and G. Maier, “Can open flow make transport networks San Francisco, CA, USA, Mar. 2014, pp. 1–3.
smarter and dynamic? An overview on transport SDN,” in Proc. IEEE [268] J. Wang, N. Cvijetic, K. Kanonakis, T. Wang, and G.-K. Chang, “Novel
Int. Conf. SaCoNeT, Vilanova i la Geltrú, Spain, Jun. 2014, pp. 1–6. optical access network virtualization and dynamic resource allocation
[247] L. Liu, T. Tsuritani, I. Morita, H. Guo, and J. Wu, “OpenFlow- algorithms for the Internet of Things,” in Proc. OSA OFC, Los Angeles,
based wavelength path control in transparent optical networks: CA, USA, 2015, pp. 1–3.
A proof-of-concept demonstration,” in Proc. Eur. Conf. Exhibit. Opt. [269] L. Liu et al., “Interworking between OpenFlow and PCE for
Commun. (ECOC), Geneva, Switzerland, Sep. 2011, pp. 1–3. dynamic wavelength path control in multi-domain WSON,” in Proc.
[248] L. Liu, T. Tsuritani, I. Morita, H. Guo, and J. Wu, “Experimental OFC/NFOEC, Los Angeles, CA, USA, Mar. 2012, pp. 1–3.
validation and performance evaluation of OpenFlow-based wavelength [270] R. Casellas et al., “Control and management of flexi-grid optical
path control in transparent optical networks,” Opt. Exp., vol. 19, no. 27, networks with an integrated stateful path computation element and
pp. 26578–26593, Dec. 2011. OpenFlow controller [invited],” IEEE/OSA J. Opt. Commun. Netw.,
[249] L. Liu et al., “Experimental demonstration of an OpenFlow/PCE inte- vol. 5, no. 10, pp. A57–A65, Oct. 2013.
grated control plane for IP over translucent WSON with the assistance [271] L. Liu, T. Tsuritani, and I. Morita, “From GMPLS to PCE/GMPLS to
of a per-request-based dynamic topology server,” in Proc. Eur. Conf. OpenFlow: How much benefit can we get from the technical evolution
Exhibit. Opt. Commun. (ECOC), Amsterdam, The Netherlands, 2012, of control plane in optical networks?” in Proc. Int. Conf. Transp. Opt.
pp. 1–3. Netw. (ICTON), Coventry, U.K., Jul. 2012, pp. 1–4.
[250] L. Liu et al., “First field trial of an OpenFlow-based unified control [272] Y. Zhao, J. Zhang, H. Yang, and Y. Yu, “Which is more suitable for
plane for multi-layer multi-granularity optical networks,” in Proc. OFC, the control over large scale optical networks, GMPLS or OpenFlow?”
Los Angeles, CA, USA, 2012, pp. 1–3. in Proc. OFC/NFOEC, Anaheim, CA, USA, Mar. 2013, pp. 1–3.
[251] R. G. Clegg et al., “Pushing software defined networking to the [273] N. Cvijetic, M. Angelou, A. Patel, P. N. Ji, and T. Wang, “Defining
access,” in Proc. Eur. Workshop Softw. Def. Netw. (EWSDN), Budapest, optical software-defined networks (SDN): From a compilation of demos
Hungary, Sep. 2014, pp. 31–36. to network model synthesis,” in Proc. OFC/NFOEC, Anaheim, CA,
[252] H. Khalili, D. Rincón, and S. Sallent, “Towards an integrated SDN- USA, Mar. 2013, pp. 1–3.
NFV architecture for EPON networks,” in Advances in Communication [274] R. Veisllari, N. Stol, S. Bjornstad, and C. Raffaelli, “Scalability analysis
Networking (LNCS 8846), Y. Kermarrec, Ed. Cham, Switzerland: of SDN-controlled optical ring MAN with hybrid traffic,” in Proc. IEEE
Springer, 2014, pp. 74–84. ICC, Sydney, NSW, Australia, Jun. 2014, pp. 3283–3288.
2782 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 18, NO. 4, FOURTH QUARTER 2016

[275] R. Sanchez, J. A. Hernández, and D. Larrabeiti, “Using transpar- [298] A. Pages et al., “Optimal virtual slice composition toward multi-
ent WDM metro rings to provide an out-of-band control network tenancy over hybrid OCS/OPS data center networks,” IEEE/OSA J.
for OpenFlow in MAN,” in Proc. IEEE ICTON, Cartagena, Spain, Opt. Commun. Netw., vol. 7, no. 10, pp. 974–986, Oct. 2015.
Jun. 2013, pp. 1–4. [299] G. M. Saridis et al., “LIGHTNESS: A function-virtualizable software
[276] M. C. Penna, E. Jamhour, and M. L. F. Miguel, “A clustered SDN defined data center network with all-optical circuit/packet switching,”
architecture for large scale WSON,” in Proc. IEEE Adv. Inf. Netw. J. Lightw. Technol., vol. 34, no. 7, pp. 1618–1627, Apr. 1, 2016.
Appl., Victoria, BC, Canada, May 2014, pp. 374–381. [300] B. Kantarci and H. T. Mouftah, “Resilient design of a cloud system
[277] J. Shin et al., “Transport SDN: Trends, standardization and architec- over an optical backbone,” IEEE Netw., vol. 29, no. 4, pp. 80–87,
ture,” in Proc. IEEE Int. Conf. Inf. Commun. Technol. Convergence, Jul./Aug. 2015.
Busan, South Korea, Oct. 2014, pp. 138–139. [301] J. Ahmed, P. Monti, L. Wosinska, and S. Spadaro, “Enhancing restora-
[278] R. Muñoz, R. Casellas, R. Martínez, and R. Vilalta, “PCE: What is it, tion performance using service relocation in PCE-based resilient optical
how does it work and what are its limitations?” J. Lightw. Technol., clouds,” in Proc. OSA OFC, San Francisco, CA, USA, 2014, pp. 1–3.
vol. 32, no. 4, pp. 528–543, Feb. 15, 2014. [302] R. Doverspike et al., “Using SDN technology to enable cost-effective
[279] E. Oki et al., “Dynamic multilayer routing schemes in GMPLS- bandwidth-on-demand for cloud services [invited],” IEEE/OSA J. Opt.
based IP+ optical networks,” IEEE Commun. Mag., vol. 43, no. 1, Commun. Netw., vol. 7, no. 2, pp. A326–A334, Feb. 2015.
pp. 108–114, Jan. 2005. [303] W. Xie, J. Zhu, C. Huang, M. Luo, and W. Chou, “Dynamic resource
pooling and trading mechanism in flexible-grid optical network vir-
[280] F. Paolucci, F. Cugini, A. Giorgetti, N. Sambo, and P. Castoldi,
tualization,” in Proc. IEEE Int. Conf. Cloud Netw. (CloudNet), 2014,
“A survey on the path computation element (PCE) architecture,” IEEE
pp. 167–172.
Commun. Surveys Tuts., vol. 15, no. 4, pp. 1819–1841, 4th Quart. 2013.
[304] L. Velasco, A. Asensio, J. L. Berral, A. Castro, and V. López, “Towards
[281] R. Casellas, R. Muñoz, R. Martínez, and R. Vilalta, “Applications
a carrier SDN: An example for elastic inter-datacenter connectivity,”
and status of path computation elements [invited],” IEEE/OSA J. Opt.
Opt. Exp., vol. 22, no. 1, pp. 55–61, 2014.
Commun. Netw., vol. 5, no. 10, pp. A192–A203, Oct. 2013.
[305] L. Zhang, T. Han, and N. Ansari, “Renewable energy-aware inter-
[282] W. Wei et al., “PONIARD: A programmable optical networking infras- datacenter virtual machine migration over elastic optical networks,”
tructure for advanced research and development of future Internet,” arXiv:1508.05400, 2015.
J. Lightw. Technol., vol. 27, no. 3, pp. 233–242, Feb. 1, 2009. [306] A. Tzanakaki et al., “A converged network architecture for energy effi-
[283] W. Wei, J. Hu, C. Wang, T. Wang, and C. Qiao, “A programmable cient mobile cloud computing,” in Proc. IEEE Int. Conf. Opt. Netw.
router interface supporting link virtualization with adaptive optical Design Model., Stockholm, Sweden, 2014, pp. 120–125.
OFDMA transmission,” in Proc. OSA OFC, San Diego, CA, USA, [307] S. Peng, R. Nejabati, S. Azodolmolky, E. Escalona, and D. Simeonidou,
2009, pp. 1–3. “An impairment-aware virtual optical network composition mechanism
[284] W. Wei, C. Wang, and X. Liu, “Adaptive IP/optical OFDM networking for future Internet,” Opt. Exp., vol. 19, no. 26, pp. B251–B259, 2011.
design,” in Proc. OSA OFC, San Diego, CA, USA, 2010, pp. 1–3. [308] S. Peng, R. Nejabati, and D. Simeonidou, “Impairment-aware opti-
[285] C. Li, W. Guo, W. Wang, W. Hu, and M. Xia, “Programmable cal network virtualization in single-line-rate and mixed-line-rate
bandwidth management in software-defined EPON architecture,” Opt. WDM networks,” IEEE/OSA J. Opt. Commun. Netw., vol. 5, no. 4,
Commun., vol. 370, pp. 43–48, Jul. 2016. pp. 283–293, Apr. 2013.
[286] M. Jinno and Y. Tsukishima, “Virtualized optical network (VON) for [309] S. Zhang, L. Shi, C. S. K. Vadrevu, and B. Mukherjee, “Network virtu-
agile cloud computing environment,” in Proc. OSA OFC, San Diego, alization over WDM and flexible-grid optical networks,” Opt. Switching
CA, USA, 2009, pp. 1–3. Netw., vol. 10, no. 4, pp. 291–300, 2013.
[287] L. Zhou, G. Peng, and N. Chand, “Demonstration of a novel [310] J. Zhao, S. Subramaniam, and M. Brandt-Pearce, “Virtual topology
software-defined Flex PON,” Photon. Netw. Commun., vol. 29, no. 3, mapping in elastic optical networks,” in Proc. IEEE ICC, Budapest,
pp. 282–290, Jun. 2015. Hungary, 2013, pp. 3904–3908.
[288] Q. Dai, J. Zou, G. Shou, Y. Hu, and Z. Guo, “Network virtualization [311] L. Gong and Z. Zhu, “Virtual optical network embedding (VONE)
based seamless networking scheme for fiber-wireless (FiWi) networks,” over elastic optical networks,” J. Lightw. Technol., vol. 32, no. 3,
China Commun., vol. 11, no. 5, pp. 1–16, May 2014. pp. 450–460, Feb. 1, 2014.
[289] Q. Dai, G. Shou, Y. Hu, and Z. Guo, “A general model for hybrid [312] X. Wang, Q. Zhang, I. Kim, P. Palacharla, and M. Sekiya, “Virtual
fiber-wireless (FiWi) access network virtualization,” in Proc. IEEE network provisioning over distance-adaptive flexible-grid optical net-
Int. Conf. Commun. Workshops (ICC), Budapest, Hungary, Jun. 2013, works [invited],” IEEE/OSA J. Opt. Commun. Netw., vol. 7, no. 2,
pp. 858–862. pp. A318–A325, Feb. 2015.
[290] S. He, G. Shou, Y. Hu, and Z. Guo, “Performance of multipath in [313] Q. Hu, Y. Wang, and X. Cao, “Survivable network virtualization for
fiber-wireless (FiWi) access network with network virtualization,” in single facility node failure: A network flow perspective,” Opt. Switching
Proc. IEEE Mil. Commun. Conf. (MILCOM), San Diego, CA, USA, Netw., vol. 10, no. 4, pp. 406–415, 2013.
Nov. 2013, pp. 928–932. [314] Z. Ye, A. N. Patel, P. N. Ji, and C. Qiao, “Survivable virtual infrastruc-
ture mapping with dedicated protection in transport software-defined
[291] S. He, G. Shou, Y. Hu, and Z. Guo, “Intelligent multipath access in
networks [invited],” IEEE/OSA J. Opt. Commun. Netw., vol. 7, no. 2,
fiber-wireless (FiWi) network with network virtualization,” in Proc.
pp. A183–A189, Feb. 2015.
OSA Asia Commun. Photon. Conf., Beijing, China, 2013, p. AF2G.38.
[315] W. Xie et al., “Survivable impairment-constrained virtual optical net-
[292] X. Meng, G. Shou, Y. Hu, and Z. Guo, “Efficient load balancing
work mapping in flexible-grid optical networks,” IEEE/OSA J. Opt.
multipath algorithm for fiber-wireless network virtualization,” in Proc.
Commun. Netw., vol. 6, no. 11, pp. 1008–1017, Nov. 2014.
IET Int. Conf. Inf. Commun. Technol. (ICT), Nanjing, China, 2014,
[316] B. Chen et al., “Cost-effective survivable virtual optical network map-
pp. 1–6.
ping in flexible bandwidth optical networks,” J. Lightw. Technol.,
[293] Q.-L. Dai, G.-C. Shou, Y.-H. Hu, and Z.-G. Guo, “Performance vol. 34, no. 10, pp. 2398–2412, May 15, 2016.
improvement for applying network virtualization in fiber- [317] H. Jiang, Y. Wang, L. Gong, and Z. Zhu, “Availability-aware survivable
wireless (FiWi) access networks,” J. Zhejiang Univ. SCIENCE virtual network embedding in optical datacenter networks,” IEEE/OSA
C, vol. 15, no. 11, pp. 1058–1070, 2014. J. Opt. Commun. Netw., vol. 7, no. 12, pp. 1160–1171, Dec. 2015.
[294] A. R. Dhaini, P.-H. Ho, and X. Jiang, “WiMAX-VPON: A framework [318] Y. Sone et al., “Bandwidth squeezed restoration in spectrum-sliced
of layer-2 VPNs for next-generation access networks,” J. Opt. Commun. elastic optical path networks (SLICE),” IEEE/OSA J. Opt. Commun.
Netw., vol. 2, no. 7, pp. 400–414, 2010. Netw., vol. 3, no. 3, pp. 223–233, Mar. 2011.
[295] A. R. Dhaini, P.-H. Ho, and X. Jiang, “Performance analysis of QoS- [319] F. Paolucci, A. Castro, F. Cugini, L. Velasco, and P. Castoldi,
aware layer-2 VPNs over fiber-wireless (FiWi) networks,” in Proc. “Multipath restoration and bitrate squeezing in SDN-based elastic
IEEE Glob. Telecommun. Conf. (GLOBECOM), Miami, FL, USA, optical networks [invited],” Photon. Netw. Commun., vol. 28, no. 1,
Dec. 2010, pp. 1–6. pp. 45–57, 2014.
[296] W. Miao et al., “SDN-enabled OPS with QoS guarantee for recon- [320] K. D. R. Assis et al., “Network virtualization over elastic optical net-
figurable virtual data center networks,” IEEE/OSA J. Opt. Commun. works with different protection schemes,” IEEE/OSA J. Opt. Commun.
Netw., vol. 7, no. 7, pp. 634–643, Jul. 2015. Netw., vol. 8, no. 4, pp. 272–281, Apr. 2016.
[297] S. Peng et al., “Multi-tenant software-defined hybrid optical switched [321] S. Hong et al., “Survivable virtual topology design in IP over WDM
data centre,” J. Lightw. Technol., vol. 33, no. 15, pp. 3224–3233, multi-domain networks,” in Proc. IEEE ICC, London, U.K., Jun. 2015,
Aug. 1, 2015. pp. 5150–5155.
THYAGATURU et al.: SDONs: A COMPREHENSIVE SURVEY 2783

[322] Z. Ye et al., “Upgrade-aware virtual infrastructure mapping in software- [347] H. Zhu, H. Zang, K. Zhu, and B. Mukherjee, “A novel generic graph
defined elastic optical networks,” Photon. Netw. Commun., vol. 28, model for traffic grooming in heterogeneous WDM mesh networks,”
no. 1, pp. 34–44, 2014. IEEE/ACM Trans. Netw., vol. 11, no. 2, pp. 285–299, Apr. 2003.
[323] J. Zhang et al., “Dynamic virtual network embedding over multilayer [348] A. P. Bianzino, C. Chaudet, D. Rossi, and J.-L. Rougier, “A survey
optical networks,” IEEE/OSA J. Opt. Commun. Netw., vol. 7, no. 9, of green networking research,” IEEE Commun. Surveys Tuts., vol. 14,
pp. 918–927, Sep. 2015. no. 1, pp. 3–20, 1st Quart. 2012.
[324] L. Nonde, T. E. H. El-Gorashi, and J. M. H. Elmirghani, “Energy [349] J. Zhang et al., “Energy-efficient traffic grooming in sliceable-
efficient virtual network embedding for cloud networks,” J. Lightw. transponder-equipped IP-over-elastic optical networks [invited],”
Technol., vol. 33, no. 9, pp. 1828–1849, May 1, 2015. IEEE/OSA J. Opt. Commun. Netw., vol. 7, no. 1, pp. A142–A152,
[325] B. Chen, “Power-aware virtual optical network provisioning in flexible Jan. 2015.
bandwidth optical networks [invited],” Photon. Netw. Commun., to be [350] M. Jinno et al., “Spectrum-efficient and scalable elastic optical path
published. network: Architecture, benefits, and enabling technologies,” IEEE
[326] G. Shen, Y. Lui, and S. K. Bose, “’Follow the Sun, Follow the Wind’ Commun. Mag., vol. 47, no. 11, pp. 66–73, Nov. 2009.
lightpath virtual topology reconfiguration in IP over WDM network,”
[351] A. Ahmed and A. Shami, “RPR–EPON–WiMAX hybrid network:
J. Lightw. Technol., vol. 32, no. 11, pp. 2094–2105, Jun. 1, 2014.
A solution for access and metro networks,” IEEE/OSA J. Opt. Commun.
[327] H. Wang and G. N. Rouskas, “Hierarchical traffic grooming: A tuto-
Netw., vol. 4, no. 3, pp. 173–188, Mar. 2012.
rial,” Comput. Netw., vol. 69, pp. 147–156, Aug. 2014.
[328] M. A. de Siqueira, F. N. C. van ’t Hooft, J. R. F. de Oliveira, [352] J. Segarra, V. Sales, and J. Prat, “An all-optical access–metro inter-
E. R. M. Madeira, and C. E. Rothenberg, “Providing optical network face for hybrid WDM/TDM PON based on OBS,” J. Lightw. Technol.,
as a service with policy-based transport SDN,” J. Netw. Syst. Manag., vol. 25, no. 4, pp. 1002–1016, Apr. 2007.
vol. 23, no. 2, pp. 360–373, 2015. [353] L. Valcarenghi et al., “Experimenting the integration of green optical
[329] L. Liu et al., “OpenSlice: An OpenFlow-based control plane for spec- access and metro networks based on SDN,” in Proc. 17th Int. Conf.
trum sliced elastic optical path networks,” Opt. Exp., vol. 21, no. 4, Transp. Opt. Netw. (ICTON), Budapest, Hungary, 2015, pp. 1–4.
pp. 4194–4204, Feb. 2013. [354] H. Woesner and D. Fritzsche, “SDN and OpenFlow for converged
[330] S. Azodolmolky et al., “Optical FlowVisor: An OpenFlow-based opti- access/aggregation networks,” in Proc. OFC/NFOEC, Anaheim, CA,
cal network virtualization approach,” in Proc. NFOEC, Los Angeles, USA, 2013, pp. 1–3.
CA, USA, 2012, pp. 1–3. [355] A. Manzalini, R. Minerva, F. Callegati, W. Cerroni, and A. Campi,
[331] C. Bock et al., “Techno-economics and performance of convergent “Clouds of virtual machines in edge networks,” IEEE Commun. Mag.,
radio and fibre architectures,” in Proc. 16th Int. Conf. Transp. Opt. vol. 51, no. 7, pp. 63–70, Jul. 2013.
Netw. (ICTON), Graz, Austria, Jul. 2014, pp. 1–4. [356] H. T. Dinh, C. Lee, D. Niyato, and P. Wang, “A survey of mobile
[332] X. Liu and F. Effenberger, “Trends in PON - fiber/wireless conver- cloud computing: Architecture, applications, and approaches,” Wireless
gence and software-defined transmission and networking,” in Proc. Commun. Mobile Comput., vol. 13, no. 18, pp. 1587–1611, 2013.
IEEE OECC, Shanghai, China, Jun. 2015, pp. 1–3. [357] M. Satyanarayanan, P. Bahl, R. Caceres, and N. Davies, “The case for
[333] Y. Chen, K. Wu, and Q. Zhang, “From QoS to QoE: A tutorial on VM-based cloudlets in mobile computing,” IEEE Pervasive Comput.,
video quality assessment,” IEEE Commun. Surveys Tuts., vol. 17, no. 2, vol. 8, no. 4, pp. 14–23, Oct./Dec. 2009.
pp. 1126–1165, 2nd Quart. 2015.
[358] G. Schaffrath, S. Schmid, and A. Feldmann, “Optimizing long-lived
[334] M. Seufert et al., “A survey on quality of experience of HTTP adaptive
cloudnets with migrations,” in Proc. IEEE 5th Int. Conf. Utility Cloud
streaming,” IEEE Commun. Surveys Tuts., vol. 17, no. 1, pp. 469–492,
Comput. (UCC), Chicago, IL, USA, 2012, pp. 99–106.
1st Quart. 2015.
[335] S. Azodolmolky, P. Wieder, and R. Yahyapour, “SDN-based cloud [359] T. Verbelen, P. Simoens, F. De Turck, and B. Dhoedt, “Cloudlets:
computing networking,” in Proc. IEEE Int. Conf. Transp. Opt. Bringing the cloud to the mobile user,” in Proc. 3rd ACM Workshop
Netw. (ICTON), 2013, pp. 1–4. Mobile Cloud Comput. Services, 2012, pp. 29–36.
[336] M. Banikazemi, D. Olshefski, A. Shaikh, J. Tracey, and G. Wang, [360] M. Maier and B. P. Rimal, “Invited paper: The audacity of fiber-
“Meridian: An SDN platform for cloud network services,” IEEE wireless (FiWi) networks: Revisited for clouds and cloudlets,” IEEE
Commun. Mag., vol. 51, no. 2, pp. 120–127, Feb. 2013. China Commun., vol. 12, no. 8, pp. 33–45, Aug. 2015.
[337] M. F. Bari et al., “Data center network virtualization: A survey,” IEEE [361] Y. Zhao et al., “Time-aware software defined networking (Ta-SDN)
Commun. Surveys Tuts., vol. 15, no. 2, pp. 909–928, 2nd Quart. 2013. for flexi-grid optical networks supporting data center application,” in
[338] N. Fernando, S. W. Loke, and W. Rahayu, “Mobile cloud computing: Proc. IEEE Globecom Workshops (GC Wkshps), Atlanta, GA, USA,
A survey,” Future Gener. Comput. Syst., vol. 29, no. 1, pp. 84–106, Dec. 2013, pp. 1221–1226.
2013. [362] H. Khodakarami, B. S. G. Pillai, and W. Shieh, “Quality of service pro-
[339] T. Wood, K. K. Ramakrishnan, P. Shenoy, and J. Van der Merwe, visioning and energy minimized scheduling in software defined flexible
“CloudNet: Dynamic pooling of cloud resources by live WAN migra- optical networks,” IEEE/OSA J. Opt. Commun. Netw., vol. 8, no. 2,
tion of virtual machines,” ACM SIGPLAN Notices, vol. 46, no. 7, pp. 118–128, Feb. 2016.
pp. 121–132, Jul. 2011. [363] K. Li et al., “QoE-based bandwidth allocation with SDN in FTTH
[340] T. Wood, K. Ramakrishnan, J. Van Der Merwe, and P. Shenoy, networks,” in Proc. IEEE Int. Conf. NOMS, Kraków, Poland, May 2014,
“CloudNet: A platform for optimized WAN migration of virtual pp. 1–8.
machines,” Dept. Comput. Sci., Univ. Massachusetts, Boston, MA, [364] A. N. Patel, P. N. Ji, and T. Wang, “QoS-aware optical burst switching
USA, Tech. Rep. TR-2010-002, 2010. in OpenFlow based software-defined optical networks,” in Proc. IEEE
[341] S. Peng, R. Nejabati, and D. Simeonidou, “Role of optical network vir- Conf. ONDM, Brest, France, Apr. 2013, pp. 275–280.
tualization in cloud computing [invited],” IEEE/OSA J. Opt. Commun.
[365] P. Wette and H. Karl, “On the quality of selfish virtual topology
Netw., vol. 5, no. 10, pp. A162–A170, Oct. 2013.
reconfiguration in IP-over-WDM networks,” in Proc. IEEE LANMAN,
[342] R. Dutta and G. N. Rouskas, “A survey of virtual topology design
Brussels, Belgium, Apr. 2013, pp. 1–6.
algorithms for wavelength routed optical networks,” Opt. Netw. Mag.,
vol. 1, no. 1, pp. 73–89, 2000. [366] S. Tariq and M. Bassiouni, “QAMO-SDN: QoS aware multipath
[343] M. R. Rahman and R. Boutaba, “SVNE: Survivable virtual network TCP for software defined optical networks,” in Proc. IEEE CCNC,
embedding algorithms for network virtualization,” IEEE Trans. Netw. Las Vegas, NV, USA, Jan. 2015, pp. 485–491.
Service Manag., vol. 10, no. 2, pp. 105–118, Jun. 2013. [367] A. Sgambelluri, F. Paolucci, F. Cugini, L. Valcarenghi, and P. Castoldi,
[344] A. Nag, M. Tornatore, and B. Mukherjee, “Optical network design with “Generalized SDN control for access/metro/core integration in the
mixed line rates and multiple modulation formats,” J. Lightw. Technol., framework of the interface to the routing system (I2RS),” in Proc. IEEE
vol. 28, no. 4, pp. 466–475, Feb. 15, 2010. Globecom Workshops (GC Wkshps), Atlanta, GA, USA, Dec. 2013,
[345] S. Azodolmolky et al., “A survey on physical layer impairments aware pp. 1216–1220.
routing and wavelength assignment algorithms in optical networks,” [368] I. Ilchmann, L. Dembeck, and J. Milbrandt, “A Transport-SDN appli-
Comput. Netw., vol. 53, no. 7, pp. 926–944, 2009. cation for incremental on-line network optimization,” in Proc. IEEE
[346] C. V. Saradhi and S. Subramaniam, “Physical layer impairment aware Int. Conf. Photon. Netw., Leipzig, Germany, May 2015, pp. 1–4.
routing (PLIAR) in WDM optical networks: Issues and challenges,” [369] X. Chang et al., “Software defined backpressure mechanism for edge
IEEE Commun. Surveys Tuts., vol. 11, no. 4, pp. 109–130, 4th Quart. router,” in Proc. IEEE Int. Symp. Qual. Service, Portland, OR, USA,
2009. Jun. 2015, pp. 171–176.
2784 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 18, NO. 4, FOURTH QUARTER 2016

[370] J. Rückert, R. Bifulco, M. Rizwan-Ul-Haq, H.-J. Kolbe, and [392] Y.-J. Kim, J. E. Simsarian, and M. Thottan, “Cross-layer orchestration
D. Hausheer, “Flexible traffic management in broadband access net- for elastic and resilient packet service in a reconfigurable optical trans-
works using software defined networking,” in Proc. IEEE Netw. Oper. port network,” in Proc. IEEE Conf. (OFC), Los Angeles, CA, USA,
Manag. Symp. (NOMS), Kraków, Poland, May 2014, pp. 1–8. Mar. 2015, pp. 1–3.
[371] E. Tego, F. Matera, V. Attanasio, and D. Del Buono, “Quality of service [393] D. Zhang et al., “Highly survivable software defined synergis-
management based on software defined networking approach in wide tic IP+optical transport networks,” in Proc. IEEE Int. Conf. OFC,
GbE networks,” in Proc. Euro Med Telco Conf. (EMTC), Naples, Italy, San Francisco, CA, USA, Mar. 2014, pp. 1–3.
Nov. 2014, pp. 1–5. [394] H. Rastegarfar and D. C. Kilper, “Robust software-defined optical
[372] D. Chitimalla et al., “Application-aware software-defined EPON access networking for the power grid,” in Proc. IEEE ICNC, Feb. 2016,
network,” Photon. Netw. Commun., vol. 30, no. 3, pp. 324–336, 2015. pp. 1–5.
[373] D. Chitimalla et al., “Application-aware software-defined EPON [395] C. DeCusatis, “Optical interconnect networks for data communica-
upstream resource allocation,” in Proc. IEEE Int. Conf. OFC, Los tions,” J. Lightw. Technol., vol. 32, no. 4, pp. 544–552, Feb. 15,
Angeles, CA, USA, Mar. 2015, pp. 1–3. 2014.
[374] X. Li et al., “Joint bandwidth provisioning and cache management [396] T. Dietz et al., “Enhancing the BRAS through virtualization,” in Proc.
for video distribution in software-defined passive optical networks,” in IEEE Conf. NetSoft, London, U.K., Apr. 2015, pp. 1–5.
Proc. IEEE OFC, San Francisco, CA, USA, Mar. 2014, pp. 1–3. [397] I2RS Status IETF. (2016). [Online]. Available:
http://tools.ietf.org/wg/i2rs/
[375] J. Matias, J. Garay, A. Mendiola, N. Toledo, and E. Jacob, “FlowNAC:
[398] S. Hares and R. White, “Software-defined networks and the interface
Flow-based network access control,” in Proc. 3rd Eur. Workshop Softw.
to the routing system (I2RS),” IEEE Internet Comput., vol. 17, no. 4,
Defined Netw. (EWSDN), Budapest, Hungary, Sep. 2014, pp. 79–84.
pp. 84–88, Jul./Aug. 2013.
[376] A. K. Nayak, A. Reimers, N. Feamster, and R. Clark, “Resonance: [399] B. Ahlgren, C. Dannewitz, C. Imbrenda, D. Kutscher, and B. Ohlman,
Dynamic access control for enterprise networks,” in Proc. ACM “A survey of information-centric networking,” IEEE Commun. Mag.,
Workshop Res. Enterprise Netw., Barcelona, Spain, 2009, pp. 11–18. vol. 50, no. 7, pp. 26–36, Jul. 2012.
[377] Y. Li, N. Hua, Y. Song, S. Li, and X. Zheng, “Fast lightpath hopping [400] J. Choi, A. S. Reaz, and B. Mukherjee, “A survey of user behavior in
enabled by time synchronization for optical network security,” IEEE VoD service and bandwidth-saving multicast streaming schemes,” IEEE
Commun. Lett., vol. 20, no. 1, pp. 101–104, Jan. 2016. Commun. Surveys Tuts., vol. 14, no. 1, pp. 156–169, 1st Quart. 2012.
[378] H. Zhu, H. Fan, X. Luo, and Y. Jin, “Intelligent timeout mas- [401] M. Casado et al., “Ethane: Taking control of the enterprise,”
ter: Dynamic timeout for SDN-based data centers,” in Proc. IEEE ACM SIGCOMM Comput. Commun. Rev., vol. 37, no. 4, pp. 1–12,
Int. Symp. Integr. Netw. Manag., Ottawa, ON, Canada, May 2015, Oct. 2007.
pp. 734–737. [402] H. Shuangyu, L. Jianwei, M. Jian, and C. Jie, “Hierarchical solution
[379] Y. Ji et al., “All optical switching networks with energy-efficient tech- for access control and authentication in software defined networks,”
nologies from components level to network level,” IEEE J. Sel. Areas in Network and System Security (LNCS 8792). Cham, Switzerland:
Commun., vol. 32, no. 8, pp. 1600–1614, Aug. 2014. Springer, 2014, pp. 70–81.
[380] H. Yang et al., “Multi-stratum resource integration for OpenFlow-based [403] K. Pagiamtzis and A. Sheikholeslami, “Content-addressable memory
data center interconnect [invited],” IEEE/OSA J. Opt. Commun. Netw., (CAM) circuits and architectures: A tutorial and survey,” IEEE J. Solid-
vol. 5, no. 10, pp. A240–A248, Oct. 2013. State Circuit, vol. 41, no. 3, pp. 712–727, Mar. 2006.
[381] H. Yang et al., “Performance evaluation of multi-stratum resources [404] P. Bull, R. Austin, and M. Sharma, “Pre-emptive flow installation for
integration based on network function virtualization in software defined Internet of things devices within software defined networks,” in Proc.
elastic data center optical interconnect,” Opt. Exp., vol. 23, no. 24, IEEE Int. Conf. FiCloud, Rome, Italy, Aug. 2015, pp. 124–130.
pp. 31192–31205, Nov. 2015. [405] H. Liang, P. Hong, J. Li, and D. Ni, “Effective idle_timeout value for
[382] H. Yang et al., “Performance evaluation of multi-stratum resources instant messaging in software defined networks,” in Proc. IEEE ICCW
optimization with network functions virtualization for cloud-based Workshop, Jun. 2015, pp. 352–356.
radio over optical fiber networks,” Opt. Exp., vol. 24, no. 8, [406] X.-N. Nguyen, D. Saucez, C. Barakat, and T. Turletti, “Rules placement
pp. 8666–8678, Apr. 2016. problem in OpenFlow networks: A survey,” IEEE Commun. Surveys
[383] E. Tego, F. Matera, and D. D. Buono, “Software defined network- Tuts., vol. 18, no. 2, pp. 1273–1286, 2nd Quart. 2016.
ing experimental approach for energy saving in GbE networks,” [407] L. Xie et al., “An adaptive scheme for data forwarding in software
in Proc. Fotonica AEIT Ital. Conf. Photon. Technol., Naples, Italy, defined network,” in Proc. 6th Int. Conf. Wireless Commun. Signal
May 2014, pp. 1–4. Process. (WCSP), Hefei, China, Oct. 2014, pp. 1–5.
[384] J. Wang, Y. Yan, and L. Dittmann, “Design of energy efficient optical [408] L. Zhang, S. Wang, S. Xu, R. Lin, and H. Yu, “TimeoutX: An adaptive
networks with software enabled integrated control plane,” IET Netw., flow table management method in software defined networks,” in Proc.
vol. 4, no. 1, pp. 30–36, Jan. 2015. IEEE GLOBECOM, San Diego, CA, USA, Dec. 2015, pp. 1–6.
[385] O. Yevsieieva and Y. Ilyashenko, “Multipath routing as a tool for energy [409] R. S. Tucker, “Green optical communications—Part II: Energy limi-
saving in transport software defined network,” in Proc. IEEE Int. Sci. tations in networks,” IEEE J. Sel. Topics Quantum Electron., vol. 17,
Pract. Conf. PICST, Kharkiv, Ukraine, Oct. 2015, pp. 25–28. no. 2, pp. 261–274, Mar./Apr. 2011.
[410] Y. Zhang, P. Chowdhury, M. Tornatore, and B. Mukherjee, “Energy
[386] M. S. Yoon and A. E. Kamal, “Power minimization in fat-tree SDN
efficiency in telecom optical networks,” IEEE Commun. Surveys Tuts.,
datacenter operation,” in Proc. IEEE GLOBECOM, San Diego, CA,
vol. 12, no. 4, pp. 441–458, 4th Quart. 2010.
USA, Dec. 2015, pp. 1–7.
[411] A. V. Lehmen et al., “CORONET: Testbeds, demonstration, and lessons
[387] S. S. Savas, C. Ma, M. Tornatore, and B. Mukherjee, “Backup reprovi- learned [invited],” IEEE/OSA J. Opt. Commun. Netw., vol. 7, no. 3,
sioning with partial protection for disaster-survivable software-defined pp. A447–A458, Mar. 2015.
optical networks,” Photon. Netw. Commun., vol. 31, no. 2, pp. 186–195, [412] L. Liu et al., “Dynamic OpenFlow-based lightpath restoration in elastic
2016. optical networks on the GENI testbed,” J. Lightw. Technol., vol. 33,
[388] A. Giorgetti, F. Paolucci, F. Cugini, and P. Castoldi, “Dynamic restora- no. 8, pp. 1531–1539, Apr. 15, 2015.
tion with GMPLS and SDN control plane in elastic optical networks,” [413] F. Paolucci, A. Sgambelluri, N. Sambo, F. Cugini, and P. Castoldi,
IEEE/OSA J. Opt. Commun. Netw., vol. 7, no. 2, pp. A174–A182, “Hierarchical OAM infrastructure for proactive control of SDN-based
Feb. 2015. elastic optical networks,” in Proc. IEEE GLOBECOM, San Diego, CA,
[389] A. Aguado et al., “Dynamic virtual network reconfiguration over SDN USA, Dec. 2015, pp. 1–6.
orchestrated multitechnology optical transport domains,” J. Lightw. [414] M. Parandehgheibi, E. Modiano, and D. Hay, “Mitigating cascading
Technol., vol. 34, no. 8, pp. 1933–1938, Apr. 15, 2016. failures in interdependent power grids and communication networks,”
[390] M. Aibin, K. Walkowiak, and A. Sen, “Software-defined adaptive sur- in Proc. IEEE Int. Conf. SmartGridComm, Venice, Italy, Nov. 2014,
vivability for elastic optical networks,” Opt. Switching Netw., to be pp. 242–247.
published. [415] T. Lehman et al., “Multilayer networks: An architecture framework,”
[391] F. Slyne, N. Kituswan, S. McGettrick, D. B. Payne, and M. Ruffini, IEEE Commun. Mag., vol. 49, no. 5, pp. 122–130, May 2011.
“Design and experimental test of 1:1 end-to-end protection for LR-PON [416] M. Ruiz et al., “Survivable IP/MPLS-over-WSON multilayer net-
using an SDN multi-tier control plane,” in Proc. Eur. Conf. Opt. work optimization,” IEEE/OSA J. Opt. Commun. Netw., vol. 3, no. 8,
Commun. (ECOC), Cannes, France, Sep. 2014, pp. 1–3. pp. 629–640, Aug. 2011.
THYAGATURU et al.: SDONs: A COMPREHENSIVE SURVEY 2785

[417] F. Touvet and D. Harle, “Network resilience in multilayer net- [439] R. Casellas et al., “SDN orchestration of OpenFlow and GMPLS flexi-
works: A critical review and open issues,” in Networking—ICN 2001. grid networks with a stateful hierarchical PCE [invited],” IEEE/OSA J.
Heidelberg, Germany: Springer, 2001, pp. 829–837. Opt. Commun. Netw., vol. 7, no. 1, pp. A106–A117, Jan. 2015.
[418] M. Vigoureux et al., “Multilayer traffic engineering for GMPLS- [440] R. Muñoz et al., “Integrated SDN/NFV management and orchestration
enabled networks,” IEEE Commun. Mag., vol. 43, no. 7, pp. 44–50, architecture for dynamic deployment of virtual SDN control instances
Jul. 2005. for virtual tenant networks [invited],” IEEE/OSA J. Opt. Commun.
[419] Y. Yoshida et al., “SDN-based network orchestration of variable- Netw., vol. 7, no. 11, pp. B62–B70, Nov. 2015.
capacity optical packet switching network over programmable flexi-grid [441] R. Vilalta, A. Mayoral, R. Muñoz, R. Casellas, and R. Martínez,
elastic optical path network,” J. Lightw. Technol., vol. 33, no. 3, “Multitenant transport networks with SDN/NFV,” J. Lightw. Technol.,
pp. 609–617, Feb. 1, 2015. vol. 34, no. 6, pp. 1509–1515, Mar. 15, 2016.
[420] E. Kissel, M. Swany, B. Tierney, and E. Pouyoul, “Efficient wide area [442] A. Mayoral, R. Vilalta, R. Casellas, R. Muñoz, and R. Martínez,
data transfer protocols for 100 Gbps networks and beyond,” in Proc. “Traffic engineering enforcement in multi-domain SDN orchestra-
Int. Workshop Netw. Aware Data Manag. (NDM), Denver, CO, USA, tion of multi-layer (packet/optical) networks,” in Proc. IEEE ECOC,
2013, pp. 1–10. Valencia, Spain, Sep. 2015, pp. 1–3.
[421] H. Rodrigues et al., “Traffic optimization in multi-layered WANs using [443] R. Muñoz, R. Vilalta, R. Casellas, and R. Martínez, “SDN orchestration
SDN,” in Proc. IEEE Symp. High Perform. Interconnects (HOTI), and virtualization of heterogeneous multi-domain and multi-layer trans-
Mountain View, CA, USA, Aug. 2014, pp. 71–78. port networks: The STRAUSS approach,” in Proc. IEEE BlackSeaCom,
[422] Y. Iizawa and K. Suzuki, “Multi-layer and multi-domain network Constanţa, Romania, May 2015, pp. 142–146.
orchestration by ODENOS,” in Proc. OSA OFC, Anaheim, CA, USA, [444] Y. Yu et al., “Field demonstration of multi-domain software-defined
2016, p. Th1A.3. transport networking with multi-controller collaboration for data center
[423] R. Muñoz et al., “The need for a transport API in 5G networks: The interconnection [invited],” IEEE/OSA J. Opt. Commun. Netw., vol. 7,
control orchestration protocol,” in Proc. OSA OFC, Anaheim, CA, no. 2, pp. A301–A308, Feb. 2015.
USA, 2016, p. Th3K.4. [445] P. J. Argibay-Losada, Y. Yoshida, A. Maruta, M. Schlosser, and
[424] A. Felix et al., “Multi-layer SDN on a commercial network control plat- K.-I. Kitayama, “Performance of fixed-length, variable-capacity pack-
form for packet optical networks,” in Proc. IEEE OFC, San Francisco, ets in optical packet-switching networks,” IEEE/OSA J. Opt. Commun.
CA, USA, Mar. 2014, pp. 1–3. Netw., vol. 7, no. 7, pp. 609–617, Jul. 2015.
[425] M. Gerola et al., “Demonstrating inter-testbed network virtualiza- [446] T. Szyrkowiec et al., “Demonstration of SDN based optical net-
tion in OFELIA SDN experimental facility,” in Proc. IEEE Conf. work virtualization and multidomain service orchestration,” in Proc.
Comput. Commun. Workshops (INFOCOM WKSHPS), Turin, Italy, Eur. Workshop Softw. Defined Netw. (EWSDN), Budapest, Hungary,
2013, pp. 39–40. Sep. 2014, pp. 137–138.
[426] S. Salsano, N. Blefari-Melazzi, A. Detti, G. Morabito, and [447] R. Vilalta et al., “Hierarchical SDN orchestration of wireless and
L. Veltri, “Information centric networking over SDN and OpenFlow: optical networks with E2E provisioning and recovery for future 5G
Architectural aspects and experiments on the OFELIA testbed,” networks,” in Proc. OSA OFC, Anaheim, CA, USA, 2016, p. W2A.43.
Comput. Netw., vol. 57, no. 16, pp. 3207–3221, 2013. [448] D. King and A. Farrel, “A PCE-based architecture for application-
[427] M. Suñé et al., “Design and implementation of the OFELIA FP7 based network operations,” Internet Requests for Comments,
facility: The European OpenFlow testbed,” Comput. Netw., vol. 61, RFC 7491, Fremont, CA, USA, Mar. 2015. [Online]. Available:
pp. 132–150, Mar. 2014. http://www.rfc-editor.org/rfc/rfc7491.txt
[428] M. Shirazipour, Y. Zhang, N. Beheshti, G. Lefebvre, and [449] R. Martínez et al., “Integrated SDN/NFV orchestration for the
M. Tatipamula, “OpenFlow and multi-layer extensions: Overview and dynamic deployment of mobile virtual backhaul networks over a multi-
next steps,” in Proc. Eur. Workshop Softw. Defined Netw. (EWSDN), layer (packet/optical) aggregation infrastructure,” in Proc. OSA OFC,
Darmstadt, Germany, Oct. 2012, pp. 13–17. Anaheim, CA, USA, 2016, p. Th1A.1.
[429] O. Gerstel, V. Lopez, and D. Siracusa, “Multi-layer orches-
[450] R. Zheng, W. Yang, and J. Zhou, “Future access architecture: Software-
tration for application-centric networking,” in Proc. Int.
defined access networking,” in Proc. IEEE CCNC, Las Vegas, NV,
Conf. Photon. Switching (PS), Florence, Italy, Sep. 2015,
USA, Jan. 2014, pp. 881–886.
pp. 318–320.
[451] N. Blum, S. Dutkowski, and T. Magedanz, “InSeRt—An intent-based
[430] M. Khaddam, L. Paraschis, and J. Finkelstein, “SDN multi-layer trans-
service request API for service exposure in next generation net-
port benefits, deployment opportunities, and requirements,” in Proc.
works,” in Proc. IEEE Softw. Eng. Workshop, Kassandra, Greece, 2008,
IEEE Int. Conf. OFC, Los Angeles, CA, USA, Mar. 2015, pp. 1–3.
pp. 21–30.
[431] J. Liu et al., “Experimental validation of IP over optical transport
network based on hierarchical controlled software defined networks [452] R. Cohen et al., “An intent-based approach for network virtualiza-
architecture,” in Proc. IEEE ICOCN, Nanjing, China, Jul. 2015, tion,” in Proc. IFIP/IEEE Int. Symp. Integr. Netw. Manag. (IM), Ghent,
pp. 1–3. Belgium, 2013, pp. 42–50.
[432] R. Vilalta et al., “Network virtualization controller for abstraction and [453] C. Prakash et al., “PGA: Using graphs to express and automatically
control of OpenFlow-enabled multi-tenant multi-technology transport reconcile network policies,” ACM SIGCOMM Comput. Commun. Rev.,
networks,” in Proc. OSA OFC, Los Angeles, CA, USA, 2015, pp. 1–3. vol. 45, no. 5, pp. 29–42, Oct. 2015.
[433] R. Jing et al., “Experimental demonstration of hierarchical control over [454] J. Rak, H. Niedermayer, D. Papadimitriou, and P. Romero,
multi-domain OTN networks based on extended OpenFlow protocol,” “Information-driven network resilience: Research challenges and per-
in Proc. IEEE Conf. OFC, Mar. 2015, pp. 1–3. spectives,” Opt. Switching Netw., to be published.
[434] Z. Zhu et al., “Demonstration of cooperative resource allocation [455] OpenFlow 1.4. (2015). [Online]. Available: http://flowgrammable.org/
in an OpenFlow-controlled multidomain and multinational SD-EON sdn/openflow/message-layer/#tab_ofp_1_4
testbed,” J. Lightw. Technol., vol. 33, no. 8, pp. 1508–1514, Apr. 15, [456] D. Hu et al., “Design and demonstration of SDN-based flexible flow
2015. converging with protocol-oblivious forwarding (POF),” in Proc. IEEE
[435] R. Vilalta et al., “Multidomain network hypervisor for abstraction GLOBECOM, San Diego, CA, USA, Dec. 2015, pp. 1–6.
and control of OpenFlow-enabled multitenant multitechnology trans- [457] K. Phemius, M. Bouet, and J. Leguay, “DISCO: Distributed multi-
port networks [invited],” J. Opt. Commun. Netw., vol. 7, no. 11, domain SDN controllers,” in Proc. IEEE Netw. Oper. Manag.
pp. B55–B61, Nov. 2015. Symp. (NOMS), Kraków, Poland, 2014, pp. 1–4.
[436] R. Muñoz et al., “Transport network orchestration for end-to-end [458] P. Öhlén et al., “Data plane and control architectures for 5G transport
multilayer provisioning across heterogeneous SDN/OpenFlow and networks,” J. Lightw. Technol., vol. 34, no. 6, pp. 1501–1508, Mar. 15,
GMPLS/PCE control domains,” J. Lightw. Technol., vol. 33, no. 8, 2016.
pp. 1540–1548, Apr. 15, 2015. [459] V. Chandrasekhar, J. G. Andrews, and A. Gatherer, “Femtocell net-
[437] L. Liu, “SDN orchestration for dynamic end-to-end control of data cen- works: A survey,” IEEE Commun. Mag., vol. 46, no. 9, pp. 59–67,
ter multi-domain optical networking,” China Commun., vol. 12, no. 8, Sep. 2008.
pp. 10–21, Aug. 2015. [460] H. ElSawy, E. Hossain, and M. Haenggi, “Stochastic geometry for
[438] A. Mayoral, R. Vilalta, R. Muñoz, R. Casellas, and R. Martínez, “SDN modeling, analysis, and design of multi-tier and cognitive cellular wire-
orchestration architectures and their integration with cloud computing less networks: A survey,” IEEE Commun. Surveys Tuts., vol. 15, no. 3,
applications,” Opt. Switching Netw., to be published. pp. 996–1019, 3rd Quart. 2013.
2786 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 18, NO. 4, FOURTH QUARTER 2016

[461] M. Louta, P. Zournatzis, S. Kraounakis, P. Sarigiannidis, and Michael P. McGarry (M’98–SM’13) received the
I. Demetropoulos, “Towards realization of the ABC vision: A compara- B.S. degree in computer engineering from the
tive survey of access network selection,” in Proc. IEEE Symp. Comput. Polytechnic Institute of NYU, in 1997, and the M.S.
Commun. (ISCC), 2011, pp. 472–477. and Ph.D. degrees in electrical engineering from
[462] M. Hasan, E. Hossain, and D. Niyato, “Random access for machine- Arizona State University, in 2004 and 2007, respec-
to-machine communication in LTE-advanced networks: Issues and tively. He is currently an Assistant Professor with the
approaches,” IEEE Commun. Mag., vol. 51, no. 6, pp. 86–93, Department of Electrical and Computer Engineering,
Jun. 2013. University of Texas at El Paso. His research inter-
[463] A. Laya, L. Alonso, and J. Alonso-Zarate, “Is the random access chan- ests include optical and software defined networks
nel of LTE and LTE-A suitable for M2M communications? A survey and cloud computing. He was a co-recipient of the
of alternatives,” IEEE Commun. Surveys Tuts., vol. 16, no. 1, pp. 4–16, IEEE ComSoc Best Tutorial Paper Award in 2009.
1st Quart. 2014.
[464] M. P. Anastasopoulos et al., “Optical wireless network convergence
in support of energy-efficient mobile cloud services,” Photon. Netw.
Commun., vol. 29, no. 3, pp. 269–281, 2015.
[465] Z. Hasan, H. Boostanimehr, and V. K. Bhargava, “Green cellular net-
works: A survey, some research issues and challenges,” IEEE Commun.
Surveys Tuts., vol. 13, no. 4, pp. 524–540, 4th Quart. 2011.
[466] Y. Shi, J. Zhang, and K. B. Letaief, “Group sparse beamforming for
green cloud-RAN,” IEEE Trans. Wireless Commun., vol. 13, no. 5,
pp. 2809–2823, May 2014.
[467] J. Wang, X. Chen, C. Phillips, and Y. Yan, “Energy efficiency with
QoS control in dynamic optical networks with SDN enabled integrated
control plane,” Comput. Netw., vol. 78, pp. 57–67, Feb. 2015.
[468] D. Butler, “SDN and NFV for broadcasters and media,” in Proc. IEEE
ECOC, Valencia, Spain, Sep. 2015, pp. 1–3.
[469] J. Ellerton et al., “Prospects for software defined networking and net-
work function virtualization in media and broadcast,” in Proc. IEEE
Tech. Conf. Exhibit., 2015, pp. 1–21. Martin Reisslein (S’97–A’96–M’98–SM’03–F’14)
[470] S. Azodolmolky et al., “SONEP: A software-defined optical network received the Ph.D. degree in systems engineering
emulation platform,” in Proc. IEEE Int. Conf. Opt. Netw. Design from the University of Pennsylvania, in 1998. He is
Model., Stockholm, Sweden, May 2014, pp. 216–221. a Professor with the School of Electrical, Computer,
[471] OFTest—Validating OpenFlow Switches. Accessed on Jul. 14, 2016. and Energy Engineering, Arizona State University,
[Online]. Available: http://www.projectfloodlight.org/oftest/ Tempe. He currently serves as an Associate Editor
[472] C. Rotsos, G. Antichi, M. Bruyere, P. Owezarski, and A. W. Moore, for the IEEE T RANSACTIONS ON E DUCATION,
“An open testing framework for next-generation OpenFlow switches,” Computer Networks, and Optical Switching and
in Proc. 3rd IEEE Eur. Workshop Softw. Defined Netw. (EWSDN), Networking.
Budapest, Hungary, 2014, pp. 127–128.
[473] M. Jarschel, F. Lehrieder, Z. Magyari, and R. Pries, “A flexible
OpenFlow-controller benchmark,” in Proc. IEEE Eur. Workshop Softw.
Defined Netw. (EWSDN), Darmstadt, Germany, 2012, pp. 48–53.
[474] M. Jarschel, C. Metter, T. Zinner, S. Gebert, and P. Tran-Gia,
“OFCProbe: A platform-independent tool for OpenFlow controller
analysis,” in Proc. IEEE Int. Conf. Commun. Electron. (ICCE), Danang,
Vietnam, 2014, pp. 182–187.

Akhilesh S. Thyagaturu received the B.E. degree

from Visveswaraya Technological University, India,
in 2010, and the M.S. degree in electrical engineer-
ing from Arizona State University, Tempe, in 2013,
where he is currently pursuing the Ph.D. degree. He
was an Engineer with Qualcomm Technologies Inc.,
San Diego, CA, USA, from 2013 to 2015.
Wolfgang Kellerer (M’96–SM’11) has been a Full
Professor with the Technische Universität München,
heading the Chair of Communication Networks
with the Department of Electrical and Computer
Engineering, since 2012. He was the Director and
the Head of wireless technology and mobile network
Anu Mercian received the bachelor’s degree in elec- research with NTT DOCOMO’s European Research
tronics engineering from the College of Engineering, Laboratories and DOCOMO Euro-Laboratory, for
Trivandrum, India, and the master’s and Ph.D. over ten years. His research focuses on concepts for
degrees in electrical engineering from Arizona the dynamic control of networks (software defined
State University (ASU), Tempe. She is currently networking), network virtualization and network
a Sr. Systems Engineer with Hewlett Packard function virtualization, application-aware traffic management, and machine-
Enterprise, Networking Research and Development, to-machine communication, device-to-device communication, and wireless
Palo Alto, CA, USA. She is also a Research Affiliate sensor networks with a focus on resource management toward a concept for
with ASU and Lawrence Berkeley Laboratory as 5th generation mobile communications in wireless networks. His research
well as a contributor with the OpenDayLight and resulted over 200 publications and 29 granted patents in the areas of mobile
Open Networking Foundation Communities. networking and service platforms. He is a member of ACM and the VDE ITG.