SPECIAL SECTION ON ULTRA-DENSE CELLULAR NETWORKS

Received September 14, 2015, accepted October 2, 2015, date of publication December 9, 2015,
date of current version December 16, 2015.
Digital Object Identifier 10.1109/ACCESS.2015.2499271

Software-Defined Network Function

Virtualization: A Survey
YONG LI1 , (Member, IEEE), AND MIN CHEN2 , (Senior Member, IEEE)
1 StateKey Laboratory on Microwave and Digital Communications, Tsinghua National Laboratory for Information Science and Technology,
Department of Electronic Engineering, Tsinghua University, Beijing 100084, China
2 School of Computer Science and Technology, Huazhong University of Science and Technology, Wuhan 430074, China

Corresponding author: M. Chen ([email protected])

This work was supported in part by the National Basic Research Program of China (973 Program) under Grant 2013CB329105, in part by
the National Natural Science Foundation of China under Grant 61300224, Grant 61301080, Grant 61171065, Grant 61273214,
Grant 91338203, and Grant 91338102, and in part by the International Science and Technology Collaboration Program under
Grant 2014DFT10070 through the Ministry of Science and Technology, China, and National Natural Science Foundation of
China under Grant 61572220.

ABSTRACT Diverse proprietary network appliances increase both the capital and operational expense of ser-
vice providers, meanwhile causing problems of network ossification. Network function virtualization (NFV)
is proposed to address these issues by implementing network functions as pure software on commodity and
general hardware. NFV allows flexible provisioning, deployment, and centralized management of virtual
network functions. Integrated with SDN, the software-defined NFV architecture further offers agile traffic
steering and joint optimization of network functions and resources. This architecture benefits a wide range
of applications (e.g., service chaining) and is becoming the dominant form of NFV. In this survey, we present
a thorough investigation of the development of NFV under the software-defined NFV architecture, with an
emphasis on service chaining as its application. We first introduce the software-defined NFV architecture
as the state of the art of NFV and present relationships between NFV and SDN. Then, we provide a historic
view of the involvement from middlebox to NFV. Finally, we introduce significant challenges and relevant
solutions of NFV, and discuss its future research directions by different application domains.

INDEX TERMS Software-defined networks, network function virtualization, middlebox, service chain,
network virtualization.

I. INTRODUCTION the underlaying hardware appliances, NFV provides flexible

Current network services rely on proprietary appli- provisioning of software-based network functionalities on top
ances and different network devices that are diverse and of an optimally shared physical infrastructure. It addresses the
purpose-built [1]–[3]. This situation induces the so-called problems of operational costs of managing and controlling
network ossification problem, which prevents the operation these closed and proprietary appliances by leveraging low
of service additions and network upgrades. To address this cost commodity servers.
issue and reduce capital expenditures (CapEx) and operat- On the other hand, with the development of Software-
ing expenditures (OpEx), virtualization has emerged as an Defined Networking (SDN) and as more abstractions are
approach to decouple the software networking processing introduced into network architectures [9]–[11], the trend
and applications from their supported hardware and allow of integrating SDN with NFV (the software-defined NFV
network services to be implemented as software [4]–[6]. architecture) to achieve various network control and manage-
Leveraging virtualization technologies, ETSI Industry ment goals has seen an noticeable growth. SDN when applied
Specification Group proposed Network Functions to NFV can help in addressing the challenges of dynamic
Virtualization (NFV) to virtualize the network functions that resource management and intelligent service orchestration.
are previously carried out by some proprietary dedicated Through NFV, SDN is able to create a virtual service envi-
hardware [7], [8]. By decoupling the network functions from ronment dynamically for a specific type of service chain,

2169-3536
2015 IEEE. Translations and content mining are permitted for academic research only.
2542 Personal use is also permitted, but republication/redistribution requires IEEE permission. VOLUME 3, 2015
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Y. Li, M. Chen: Software-Defined NFV: A Survey

consequently the dedicated hardware and complex labor work NFV in Section V, and discuss its future research directions
to provide a new coming service request is avoid. In con- by different application domains in Section VI. Finally, we
junction with the use of SDN, NFV further enables real-time conclude the paper in Section VII.
and dynamic function provisioning along with flexible traffic
forwarding. II. SOFTWARE-DEFINED NETWORK
Software-defined NFV leverages network virtualization FUNCTION VIRTUALIZATION
and logically centralized intelligence to minimize the To reduce CapEx and OpEx introduced by diverse proprietary
service providing cost and maximize the utilization of net- appliances, NFV was proposed to exploit and take advantage
work resource. In this case, the obtained higher resource of the virtualization technology. NFV allows network opera-
utilization will introduce less investigation on the hardware tors and service providers to implement network functions in
equipments, which on the other hand simplifies networking software, leveraging standard servers and virtualization tech-
operations. Moreover, by automating current manually inten- nologies, instead of run on purpose-built hardware. Recent
sive network configuration, provisioning, and management, trends of increased user information demands, explosion of
the time and operation complexity are significantly reduced traffic and diverse service requirements further drive NFV
and manual errors are dramatically decreased, which offers to be integrated with SDN, forming the software-defined
better scalability. On the other hand, especially in large- NFV architecture. This architecture offers great flexibility,
scale networks, deploying and providing a new kinds of programmability and automation to the operators in service
service usually results in a long and repeated process that provisioning and service model.
requires long cycles of validation, verifying, and testing.
By automating the control, managing and orchestration A. NETWORK FUNCTION VIRTUALIZATION
of the NFV related infrastructure, the deploying time Diverse and fixed proprietary appliances make the service
and operation cost for network configuration and opera- deployment and testing increasingly difficult. NFV was
tion changes for these new services will be significantly proposed as a key technology to benefit IT virtualization
shortened. evolution [4]–[6] by separating the hardware network func-
Service chaining is the main area that software-defined tions from the underlying hardware appliances by transfer-
NFV can play an important role [12], [13]. In the current ring network functions from dedicated hardware to general
networks, a service chain include a set of hardware dedicated software running on commercial off-the-shelf (COTS) equip-
network appliances offering services such as load balancers, ments, i.e., virtual machines [17]–[20]. These software
firewall, Deep Packet Inspection (DPI), Intrusion Detection applications are running on standard IT platforms like
System (IDS), and etc., to support a dedicate networking high-performance switches, service, and storage. By NFV,
processing and applications [14]–[16]. When it comes a new the different network functions can be deployed in differ-
service requirement, new hardware devices must be deployed, ent locations of the networks such as data-centers, network
installed and connected by some order, which is extremely nodes, and end-node of network edge as required. Currently,
time-consuming, complexity, high-cost and error-prone. This the market of NFV includes switching elements, network
kind of networking service providing requires dedicate plan appliances, network services and applications. Here we sum-
of networking changes and outages, which on the other hand mary the commonly used network functions considered for
incurs high OPEX. This situation is exacerbated when a NFV [7], [21].
lot of different kinds of service sequences are dedicated to • Network switching elements [22], i.e., Broadband
different traffic flows by an operator. On the other hand, the Network Gateway (BNG), carrier grade NAT,
architecture of software-defined NFV is able to simplify the Broadband remote access server (BRAS), and routers.
service chain deployment and provisioning. It enables easier • Mobile network devices, i.e., Home Location Register/
and cheaper service providing in the local area network, Home Subscriber Server (HLR/HSS), Serving GPRS
enterprise networks, data center, and Interent service provider Support NodeMobility Management Entity
networks. (SGSNMME), Gateway support node/Packet Data
This survey introduces the state-of-the-art of NFV Network Gateway (GGSN/PDN-GW), RNC, NodeB
and its main challenges within the software-defined and Evolved Node B (eNodeB) [23].
NFV architecture. Service chaining is highlighted and dis- • Virtualized home environments [24], [25].
cussed as a core application of NFV in different contexts. • Tunneling gateway devices, i.e., IPSec/SSL virtual
We further provide guidelines for future developments of private network gateways.
NFV in various application scenarios. In Section II, we • Traffic analysis elements, i.e., Deep Packet Inspec-
introduce the software-defined NFV architecture as the state- tion (DPI), Quality of Experience (QoE) measurement.
of-the-art of NFV and present relationships between NFV • Service Assurance, Service Level Agreement (SLA) [26]
and SDN. Then, we provide a historic view of the involve- monitoring, Test and Diagnostics.
ment from middlebox to NFV in Section III. After survey • Next-Generation Networks (NGN) signaling such as
the current technology of service chain in Section IV, we Session Border Controller (SBCs), IP Multimedia
introduce significant challenges and relevant solutions of Sub-system (IMS).

VOLUME 3, 2015 2543

 Y. Li, M. Chen: Software-Defined NFV: A Survey

• Application-level optimization devices, i.e., Content

Delivery Network (CDNs) [27], load balancers, cache
nodes, and application accelerators.
• Network security devices, i.e., Firewalls [28], intrusion
detection systems, DOS attack detector, virus scanners,
spam protection, etc.
The major advantage of using NFV is to reduce middle-
boxes deployed in the traditional networks to take the advan-
tages of cost savings and bring flexibility. On the other side,
NFV technology also supports the co-exists of multi-tenancy
of network and service functions, through allowing the usage
of one physical platform for different services, applications,
and tenants.

FIGURE 2. SDN architecture.

feature would bring potential benefits of enhanced configu-

ration, improved performance, and encouraged innovation in
network architecture and operations. Especially, SDN offers a
promising alternative for traffic steering by programmatically
configuring forwarding rules [39]. Fig. 2 depicts the SDN
architecture [21], [40], [41]. There are three different layers:
• Application Layer: This layer covers an array of applica-
tions focusing on network services, and they are mainly
software applications communicating with the control
layer.
• Control Layer [42]–[44]: As the core of SDN, the control
FIGURE 1. Illustration of the NFV framework. layer consists of a centralized controller, which logically
maintains a global and dynamic network view, takes
B. NFV FRAMEWORK requests from the application layer, and manages the
ETSI defines the NFV architectural framework (showing network devices via standard protocols.
in Fig. 1) enabling virtualized network functions (VNF) to be • Data-plane Layer: Infrastructure including switches,
deployed and executed on a Network Functions Virtualisation routers and network appliances. In SDN context,
Infrastructure (NFVI), which consists of commodity servers these devices are programmable and support standard
wrapped with a software layer that abstracts and logically interfaces [45].
partitions them [7], [29]. Above the hypervisor layer, a VNF The application layer utilizes the northbound APIs to com-
is typically mapped to one VM in the NFVI. The deployment, municate with the SDN controller, which enable different
execution and operation of VNFs on the NFVI are steered control mechanisms for the networks. The southbound APIs
by a Management and Orchestration (M&O) system [30], define the communication interface between the controller
whose behaviour is driven by a set of metadata describing the layer and data plane devices, which on the other hand enable
characteristics of the network services and their constituent the application to control the forwarding device via this flex-
VNFs. The M&O system includes an NFV Orchestrator in ible and programmable way.
charge of the lifecycle of network services, a set of VNF man-
agers in charge of the lifecycle of the VNFs and a virtualized D. NFV V.S. SDN
infrastructure manager, which can be viewed as an extended NFV and SDN are closely related and highly complemen-
cloud management system responsible for controlling and tary to each other. NFV can serve SDN by virtualizing the
managing NFVI resources [31], [32]. SDN controller (which can be regarded as a network func-
tion) to run on cloud, thus allows dynamic migration of the
C. SOFTWARE-DEFINED NETWORKS controllers to the optimal locations. In turn, SDN serves
Software-Defined Network (SDN) is an important and NFV by providing programmable network connectivity
recently emerging network architecture to decouple the net- between VNFs to achieve optimized traffic engineering and
work control from the data forwarding by directly program- steering [29], [46]. However, NFV and SDN are completely
ming [33]–[36]. With its inherent decoupling of control different from the concepts to the system architecture
plane from data plane, SDN offers a greater control of and functions, which are summarized by the following
a network through programming [37], [38]. This combined aspects:

2544 VOLUME 3, 2015

Y. Li, M. Chen: Software-Defined NFV: A Survey

• NFV is a concept of implementing network functions orchestration system and the controller steer the traffic trav-
in software manner, while SDN is concept of achieving eling through the required and appropriate sequence of VMs
centrally controlled and programmable network archi- and forwarding devices by installing forwarding rules into
tecture to provide better connectivity. them.
• NFV aims at reducing CapEx, OpEx, and space and
power consumption, while SDN aims at providing net- III. FROM MIDDLEBOX TO NFV
work abstractions to enable flexible network control, While NFV receives a large amount of attentions from both
configuration and fast innovation. the industry and academic world, the idea of decoupling
• NFV decouples the network functions from the propri- the software layer and the underlying hardware has been
etary hardware to achieve agile provisioning and deploy- around for many years. Though NFV does not limit in vir-
ment, while SDN decouples the network control plane tualizing middleboxes, the concept of NFV was initiated in
from the data plane forwarding to provide a centralized the context of middlebox. In this section, we introduce the
controller via enabling programmability. evolution from traditional purpose-built middlebox to NFV,
during which consolidated middlebox and software-defined
middlebox acted as transitional paradigms.

A. MIDDLEBOX OVERVIEW
A middlebox, also named network appliance, is a networking
forwarding or processing device that transmits, transforms,
filters, inspects, or control network traffic for purposes of net-
work control and management [2], [54]–[56]. A middlebox
service or function is a method or operation performed by
a network device that needs specific intelligence about the
applications. Typical examples of middleboxes, i.e., network
appliance, include network address translators that modify
packets’ destination and source addresses, and firewalls that
filter unwanted or malicious traffic, and The following are
FIGURE 3. Software-defined NFV system. commonly deployed middleboxes [57]:
1) Network Address Translator (NAT) [58]: NAT is
E. SOFTWARE-DEFINED NFV ARCHITECTURE utilized to replace the source and/or destination
The software-defined NFV system is illustrated in Fig. 3. IP addresses of certain packets that traverse over it.
It consists a control module, forwarding devices and NFV Typically, NAT is deployed to share a single
platform at the edge of the network. The logic of packet IP address by multiple end hosts, i.e., computers: hosts
forwarding is determined by the SDN controller and is ‘‘behind’’ the NAT are assigned a private IP address,
implemented in the forwarding devices through forwarding and their packets destined to Internet will traverse the
tables. Efficient protocols, e.g., Openflow [47]–[51]), can be NAT middlebox that replaces their private address with
utilized as standardized interfaces in communicating between the public address to communicate with the public
the centralized controller and distributed forwarding devices. Internet.
The NFV platform leverages commodity servers to imple- 2) Firewall (FW) [28]: Firewall is utilized to filter traf-
ment high bandwidth NFs at low cost. Hypervisors run on fic according to a set of pre-defined security policies
the servers to support the VMs that implement the NFs. by rejecting packets with specific fields headers of
This platform allows customizable and programmable data the IP and transport, or using more complex policies
plane processing functions such as middlebox of firewalls, of inspecting packets at the application and session
IDSes, proxies, which are running as software within virtual layer.
machines, where NFs are delivered to the network operator 3) Intrusion Detection System (IDS) [59]: IDS is utilized
as pieces of pure software. to monitoring the network to detect security anomalies.
The SDN controller [43], [44], [52], [53] and the NFV Since it does not filter data in real-time, they usu-
orchestration system compose the logical control module. ally are capable of more complex packet processing
The NFV orchestration system is in the charge of provi- than hte middlebox of firewalls that need to made the
sioning for virtualized network functions, and is controlled accept/reject decision when the packet arrives.
by the SDN controller through standard interfaces. After 4) Load Balancer (LB) [60]: The middlebox of network
obtain the network topology and policy requirements, the load balancer is to split network traffic across multiple
control module computes the optimal function assignments different servers, with the aims of optimizing resource
(assigning network functions to certain VMs) and translates use, minimizing network response time, maximizing
the the logic policy specifications into optimized routing system throughput, and avoiding overload of other
paths. The function assignments are enforced by the NFV resource.

VOLUME 3, 2015 2545

 Y. Li, M. Chen: Software-Defined NFV: A Survey

5) WAN Optimizer: WAN Optimizer improves bandwidth network forwarding devices like switch/router’s computing
consumption and shorten network transmission latency modules or separate server and machines. Following such
between different endpoint in the WAN. Typically, idea, [63] is proposed to remove the dedicated hardware mid-
they are deployed near the sending or receiving com- dleboxes and move the related network processing services
munication host, and then cache and compress traffic on network platform and standard servers. In order to provide
passing by. efficient in-network services on top of various processing
6) Flow Monitor (FM): The middlebox of flow monitor modules in the network devices, they proposed a flexible
is utilized to collect information of the flows in the control system that integrate the network processing modules
network for the utilization of traffic analysis or trouble and forwarding devices in an automated way.
shooting. It is widely utilized in the data center or
service providers’ networks. C. SOFTWARE-DEFINED MIDDLEBOX
As SDN evolves, the principles of abstracting the archi-
B. CONSOLIDATED MIDDLEBOX tecture of network from the control and data plane have
Traditionally, a new type of middlebox was usually emerging been investigated in various contexts. This idea introduces
as a solution for some specific need, then integrated into the some unique opportunities for the development of middle-
network of infrastructure by the widely deployment. This boxes [64]. Inspired by the idea of SDN, some researchers
deployment approach leads to significant inefficiency in the proposed a software-defined middlebox and corresponding
use and management of infrastructure hardware resources. networking architecture, with the aim of providing fine-
Prior to NFV, researchers turned to the age-old idea of con- grained and programmable control over the Middlebox state
solidation to address the above challenges by systematically and network forwarding. Now, we summary an overview of
re-architecting middlebox infrastructure to exploit opportu- the software-defined middleboxes.
nities for consolidation [1], [61]–[63]. Now, we provide an
overview for the efforts on consolidating middleboxes, which 1) ENABLING MIDDLEBOX INNOVATION [56]
are precursors to the current NFV paradigm. Ref. [56] is an early effort on designing software-centric
middlebox, which runs on general-purpose hardware plat-
1) CoMb [61] forms controlled and managed through open APIs. A research
To address the important resource management and con- agenda is proposed with the target of manage a single or an
trolling problems that arise in exploiting the benefits of ensemble of middleboxes. To enable fast middlebox innova-
middlebox deployment, CoMb is proposed by consolidating tion, this work explore an approach through three different
individual middleboxes through decoupling the software and strategies: software-centric implementations of middlebox
hardware, which enables software-based implementations of that decouple hardware from the software; multiple software-
middlebox to deploy and run on a the general and consoli- based middlebox are implemented on a shared general
dated hardware platform. On the other hand, CoMb consol- hardware platform; and, finally centralized controlling and
idates the management of different middlebox into a single management with open APIs to provide, control and manage
centralized controller, which takes a unified and network- the deployment of the middlebox.
wide configurations and controlling for policy requirements
across the overall traffic and applications. This is in con- 2) OpenMB [65]
trast to today’s approach where the middleboxs is controlled OpenMB consists of somehow modified middleboxes by
and managed separately. CoMb addressed these important exposing a southbound API for importing/exporting the com-
resource control and management challenges, which results plicate states of middlebox, where the centralized controller
in reducing network provisioning cost and overhead in the implements the open API to define how state can be set and
deployment and operation of middlebox devices. accessed. OpenMB-enabled middleboxes allow a variety of
dynamic scenarios to be realized without influence on the
2) APLOMB [1], [62] correctness or performance of middleboxes, which is crucial
APLOMB is proposed to enable the traffic processing in the to continued innovation in software-defined middlebox.
third-party middlebox device and service providers running
in the data centers and cloud. APLOMB allows enterprise 3) xOMB [66]
networks, as well as individual end hosts, to tunnel their xOMB (Extensible Open MiddleBox) provides pro-
traffic to and from a cloud service, which applies middlebox grammable, flexible and scalable middleboxes on the plat-
processing to their traffic. In this way, it avoids the costly and form of general hardware like servers and operating systems
management cost of administering middleboxes in a local- to achieve high efficiency flow controlling. It utilize
region network. general programmable processing approaches with
user-defined modules for network packet parsing, data trans-
3) INTEGRATE MIDDLEBOXES INTO NETWORK [63] forming, and flow forwarding. By these design, xOMB
There has been a trend to reduce the middleboxes by deploy- shows how middleboxes can be utilized to support different
ing the network services and related processing into the services.

2546 VOLUME 3, 2015

Y. Li, M. Chen: Software-Defined NFV: A Survey

IV. SERVICE CHAINING support efficient forwarding for a large number of applica-
Service chaining is an important model for network service tions and subscribers.
providers, in which NFV plays an important role. It is utilized
to organize the service function deployment, where the ability 3) FLOWTAG [76]
of specifying an ordered list of service processing for the The dynamic, traffic-dependent, and hidden actions of mid-
service’s traffic flows [67] is provided. A chain defines the dleboxes make it hard to systematically enforce and verify
required processing or functions and the corresponding order network-wide policies, and to do network diagnosis. Flowtag
that should be applied to the data flow. These chains require is a complement for SDN based service chaning approaches,
integration of service policy and the above applications to dealing with the dynamic changes imposed by middleboxes.
achieve optimal resource utilization. FlowTags-enhanced middleboxes export tags to provide the
Traditional service chaining mainly rely on manual con- required network context. On the other hand, the SDN con-
figuration which is tedious, error-prone and clumsy. SDN trollers is able to configure the operations of tag generation
provides new capability steer traffic dynamically based on and consumption by the FlowTags APIs. These operations
user requirements. However, hardware-based middleboxes benefit restore bindings between packets and their origins,
limit the benefit of SDN due to their fixed functionalities and guarantee that packets of flow follow policy-required
and deployment. NFV is a good enabler for SDN. With the paths. This approach requires minimal changes in middle-
ability of dynamic function provisioning offered by NFV and boxes and the overhead of FlowTags is comparable to tradi-
the centralized control of SDN, new opportunities emerge tional SDN mechanisms.
in service chaining. Better performance and resource uti-
lization can be achieved with the software-defined NFV B. SERVICE CHAINING IN THE SOFTWARE-DEFINED
architecture. NFV ARCHITECTURE
SDN and NFV together have the potential to benefit ser-
A. SDN&MIDDLEBOX BASED SERVICE CHAINING vice operators satisfy user service level agreements, accu-
SDN offers the flexible control approach and enables rately monitor and control network traffic, which further
dynamic traffic forwarding, and these style of traffic control reduces the minimize operating cost [77]. On one hand,
for middlebox-specific flow can realize flexible and efficient NFV moves network functions out of dedicated hardware
service chaining with no need to generate any placement or boxes to the software based on general hardware plat-
introduce some constraints on middleboxes, which are on the form. On the other hand, SDN moves control functions
other hand easily supported by current SDN standards [73]. out of the hardware and places it in the software con-
Three are some important works in this topic, which are troller. Therefore, the service deployment and service chains
introduced below. can be provided and reconfigured in the controller. In this
scenario, not only flexible and dynamic operations are
allowed, the chance for operation error and events will be
1) SYMPLE [74]
much smaller because the network controller has an overall
SYMPLE (Software-defIned Middlebox PoLicy Enforce- view, which benefits reducing the probability of inconsistent
ment) is a software-defined policy enforcement layer for configurations.
traffic steering. It enables the network managers and opera- Moving the required network functions into software
tors to specify a high-level abstractions of logical middlebox means that deploying the service chain no longer requires
routing policy, and it then further automatically translates the acquiring dedicated middlebox. In this case, the network
policy into control rules with the knowledge of the physical functions execute as the software running on virtual machines
network topology, forwarding device capacities, and resource with the control of a hypervisor, which enable flexibil-
constraints of the whole networks. Without modifying any ity computational and networking resource provision. Thus,
middleboxes and network devices, SYMPLE offers efficient since the computational capacity can be increased when it
data plane for packet processing, and automatically dealing is required, there’s no need to over-provision. On the other
with specifiable packet modifications, which is more mod- hand, software-defined NFV service chaining also benefits
est compared to ongoing and parallel work developing new the network upgrade process. For geographically distributed
visions for SDN or middleboxes. networks, upgrading network devices requires a large amount
of cost. Moreover, the error happening in the network updates
2) StEERING [75] and re-configuration can bring down the entire network may
StEERING, short for SDN inlinE sERvices and forward- outage on interconnecting providers’ networks. However,
iNG, is a scalable framework for dynamically routing traf- with the software-defined NFV, service providers is able
fic through any sequence of middleboxes. With simple to create new chains without radically changing hardware.
centralized configuration, StEERING can explicitly steer Finally, service operator can utilize these service chaining
different types of flows through the desired set of mid- techniques to placing themselves, instead of the third party
dleboxes, scaling at the level of per-subscriber and per- provider. With intelligent service chaining, complexity of
application policies. Built on top of SDN, StEERING can resource provisioning is significantly reduced. Thus service

VOLUME 3, 2015 2547

 Y. Li, M. Chen: Software-Defined NFV: A Survey

providers can deliver services on demand without the help of of view. Below we discuss some important related works on
third parties. function virtualization.
The software-defined NFV architecture is still in research
phase. A unified control and orchestration framework is 1) DPDK [80]
required to integrate the SDN controller, forwarding elements DPDK is a set of libraries and drivers for fast packet process-
and virtual network functions. Moreover, due to the exis- ing for the network functions. It could be run on a wide range
tence of dynamic function and resource provisioning, this of processors. However, the DPDK system has some limi-
framework should also provide coordinated control of both tation to support virtualization, and it along cannot support
network forwarding state and network functions’ states [78]. flexible, high performance functionality in the environment
Fig. 4 illustrates an example of the service chaining process. of NFV.
Taking user policies as inputs, the control module assigns the
2) NetVM [81]
NFs fulfilling these services in an optimal way and mean-
while the optimal routing paths of all policies are selected NetVM is a software platform for running diversity network
taking account of the resource constraints. Then the service functionality at line-speed based on the general commodity
functions are chained by the centralized controller and the hardware. It takes advantage of DPDK’s high throughput
traffic flows are steered according to the service chains. packet processing capabilities, and further enables flexible
traffic steering and overcomes the performance limitations of
hardware switching. Thus, It provides the capability to sup-
port network functions chains by flexible, high-performance
network elements.

3) ClickOS [57], [82]

ClickOS is a high-performance, virtualized software network
function platform. It provides small, booting quickly, and
little delay virtual machines, and over one hundred of them
can be concurrently run while guaranteeing ine-rate pipe on
the general commodity server. To achieve high performance,
ClickOS relies an extensive overhaul of Xen’s I/O subsystem
to speed up the networking process in middleboxes. ClickOS
is proof that software solutions alone are enough to signif-
icantly speed up virtual machine processing, to the point
where the remaining overheads are dwarfed by the ability to
FIGURE 4. Service chaining in the software-defined NFV architecture.
safely consolidate heterogeneous middlebox processing onto
the same hardware.
V. CHALLENGES AND PROBLEMS OF NETWORK
FUNCTION VIRTUALIZATION B. PORTABILITY
NFV is an important innovation and a promising approach for The NFV framework is expected to support the loading,
the service operators and providers. However, it also faces executing and moving VNFs across different but standard
several challenges. In this section, the corresponding chal- servers in multi-vendor environments. This capability is
lenges, open problems, and related solutions are summarized known as portability [83]. These virtualized network func-
with the classification organized in Table 1. tions defeats the portability goal and key benefits of NFV,
namely the capability of multi-tenancy and resource isolation.
A. FUNCTION VIRTUALIZATION Furthermore, once instantiated, a NF leveraging SR-IOV can-
The virtualized functions should meet performance require- not be migrated to another server. The portability challenge is
ments to support packet processing at line-rate for multiple how to achieve high performance leveraging hardware accel-
tenants [29], [79]. First, since neither the hypervisors nor the erators and at the same time have hardware independent NFs.
virtual machines have been optimized for the processing of This approach ensures that the VNFs are OS-independent
middlebox, obtaining high performance, i.e., high I/O speed, and resource isolation is also guaranteed since the VNFs are
fast packet processing, short transmission delays, etc, from executed on independent VMs and are decoupled from the
standard servers is the main challenge for function virtual- underlying OS by the hypervisor layer.
ization. Further, as a server may implements a large amount
of functionality, their platforms should host a wide range of C. STANDARD INTERFACES
virtual machine and software packages. Finally, NFV hard- NFV rely on existing infrastructure to touch the customer.
ware and software platforms should support multi-tenancy, In this case, it is also highly unlikely that a upgrade of
because they are concurrently run by software belonging to the physical network or entire operational support systems
the different operators. These co-located VNFs should be will be feasible. This is a management software integration
isolated not only from a security but also a performance point challenge with the interfaces between NFV and underlying

2548 VOLUME 3, 2015

Y. Li, M. Chen: Software-Defined NFV: A Survey

TABLE 1. NFV challenges.

infrastructure [84], [85]. On the other hand, the interfaces the performance of service chaining [87], [88]. Both auto-
between the centralized controller and VNFs should also be mated provisioning and placement require a global view
standardized. To smoothly bridge NFV with upper and lower of the resources and a unified control and optimization
layers, the VNFs and the underlaying computing platform system with various optimization engines running in it.
should be described by standard templates that enable flexible Another issues is to translate higher-level policies, which
controlling and management. Thus, north- and south-bound is generated from the resource allocation and optimiza-
need to be developed. North-bound interactions are used to tion mechanisms, into lower level configurations [90], [91].
control and manage functions to different types of instances, Templates and standards should be developed to guarantee
e.g., physical servers, VM and VNFs. Since network func- automated and consistent translation. For example, when
tions need service-oriented APIs to be controlled directly there is a need to achieve high-level goal of reducing the
or indirectly, each network service has a specific operation networking transmission delay, the optimization engine may
policy and SLA. Moreover, VNFs could use the north-bound require an algorithm to provision and place virtual functions
API for the requests. On the other hand, the south-bound ensuring that the least overall transmission delay is achieved.
API are utilized to communicate with the NFVI and request Conversely, when we require to achieve the minimum max-
information from other framework entities. Thus, how to imum link utilization, it would need a different optimization
design a flexible and efficiency API for both the north-bound engine with a different algorithm. For more effective opera-
and south-bound communications are important problems in tion and control, the optimization approach should support
the research and development of NFV technologies. real-time swap to make provisioning and placements that
dynamically match the high-level policies from the operator
D. FUNCTION DEPLOYMENT and application.
Fine-grained deployment, control and management of net-
work functions are needed in the context of NFV-enabled E. TRAFFIC STEERING
network nodes, for various optimization purposes [86]–[88]. SDN offers new agility of traffic steering by allowing the
Thus, many challenges are related to algorithm and system network operators and service providers to specify a logical
design regarding of function deployment. control policy, and then it automatically translates this into
One of these challenges is to automatically provide net- data plane forwarding rules. Prior to this, the routing paths are
work and function process resources according to the usage carefully selected by the optimization framework taking into
of the resources involved [89]. A similar and probably account the physical topology, link capacities, and network
even more important challenge is to achieve automatic resource constraints. Solid work has been done on traffic
placement and allocation of the VNFs, since the place- steering in hardware based middlebox systems. However,
ment and assignment of the VNFs significantly impact in the software-defined NFV architecture, traffic steering is

VOLUME 3, 2015 2549

 Y. Li, M. Chen: Software-Defined NFV: A Survey

jointly optimized with NF deployment that can achieve better policies of services and resources, in addition to orchestration
composition. However, the unified optimization paradigm process. It aims to decouple the creation of management
also makes the optimization problem difficult to solve since information from the way it is presented. CloudNFV concerns
more variables are introduced and twisted. To achieve online two challenges of NFV orchestration in cloud, namely to
computing of traffic steering, heuristic algorithms should be embed services and network functions into physical/virtual
designed to reduce the computing complexity. infrastructures and the trade-off between automating SLA
and price negotiation.
VI. APPLICATIONS AND FUTURE WORK
Software-defined NFV technology is in the usage of delivery 2) THE REALTIME CLOUD [102]
significant benefits in niche applications today, while its full The realtime cloud relies on combing cloud, NFV and service
scale use and benefits have yet to be achieved. In this section, provider SDN. Together, these technologies together enables
we look at what should happen in the next phase of software- more fluid, more dynamic and more responsive to new ser-
defined NFV development following the journey and suc- vice needs. By enabling efficient control and management,
cess the concept has enjoyed so far. We describe the major and orchestration across network resources and applications,
domains that are expected to dominate the software-defined network-enabled Cloud, NFV and SDN together are able
NFV scenario over next few years. to help operators ensure they provide efficient and scale
services.
A. CLOUD COMPUTING 3) CLOUDBAND [103]
Cloud computing [92]–[94] enable globally distributed ser- CloudBand is Alcatel’s end-to-end NFV solution and plat-
vices and enterprises to quickly deploy, manage and opti- form. Being open and multi-vendor, it supports the strin-
mize their computing infrastructure dynamically. Partitioning gent needs of carriers and speedup the evolution to NFV.
or replicating a service across multiple globally distributed By introducing the CloudBand ecosystem, Alcatel is making
instances allow these services to move closer to the users it available to the NFV community for free with the goal
thus providing richer user experiences, avoid infrastructure of fostering collaboration and experimentation, which enable
bottlenecks, and implement fault tolerance [35], [95]–[98]. to accelerate NFV adoption and create new business and
NFV is an enabler of such dynamic service provisioning. application opportunities.
By replacing service elements with virtual network func-
tions, New functions can be added or improved by updat- B. MOBILE NETWORK
ing a software image, rather than waiting for a vendor to On the other hand, NFV considers all network functions
develop and manufacture a dedicated appliance. Furthermore, for virtualization through well-defined standards, i.e., in
while integrated with SDN, service providers can express and mobile network, NFV targets at virtualizing mobile core
enforce application traffic management policies and appli- network and the mobile-network base station [104]–[111].
cation delivery constraints at the required level of granu- NFV also benefits data centers owned by mobile service
larity [89], [99], [100]. providers [112], including mobile core network, access net-
NFV allows service providers to provide better working and mobile cloud networks.
services to the users by dynamically changing their deploy- For the core networks, which is the most important part
ment topologies or traffic allocations based on user access of mobile networks [113], [114], NFV allows the cellular
patterns, user mobility, infrastructure load characteris- providers to adopt a network more akin to the data centers,
tics, infrastructure failures and many such situations that which consist of a fabric simple forwarding devices, with
may cause service degradation, disruption or churn [85]. most functionality executed in commodity servers that are
Similarly, replicated service instances might need to be close to the base stations. Some network functions can even
moved/instantiated/released to mask infrastructure failures, be fulfilled by packet-processing rules installed directly in
load conditions, or optimize the deployment based on access the switches [105], [108], [110]. In the system, a logically-
patterns and social interaction graphs. NFV, as well, can centralized controller is able to steer the network traffic
provide intelligent infrastructure support for such dynamic through the required network functions to realize service
service deployment scenarios. Moreover, since NFV offers chaining.
good support for multi-tenant usage, it is available for wide For the access networks, the base station is also consid-
area dynamic multi-cloud environments that can be shared ering to utilize the virtuliazation technology [118], [119].
by multiple providers to implement their specific distributed Thus, SDN and NFV are applied to the wireless access
service delivery contexts. networks [115]–[117], [120] to sharing their remote base-
Below we enlist some important pioneering works trying station infrastructure to achieve better coverage and services
to implement NFV in clouds. with the minimum investment of CAPEX and OPEX.

1) CloudNFV [101] C. ENTERPRISE NETWORK

CloudNFV is a multi-vendor consortium, which mainly aims NFV will no doubt to be widely utilized in the enterprise
to build an unified data model that incorporates data and [122]–[125]. Network managers would like to consume as

2550 VOLUME 3, 2015

Y. Li, M. Chen: Software-Defined NFV: A Survey

much or as little of the network as they need, but there is a [12] A. Friis-Christensen, R. Lucchi, M. Lutz, and N. Ostländer, ‘‘Service
gap between what enterprise customers want and what service chaining architectures for applications implementing distributed geo-
graphic information processing,’’ Int. J. Geographical Inf. Sci., vol. 23,
providers can offer today, which can be address by NFV. no. 5, pp. 561–580, 2009.
It enables the dynamic provisioning of virtual network [13] R. Lemmens, R. De By, M. Gould, A. Wytzisk, C. Granell, and
services on commodity servers within minutes instead of P. Van Oosterom, ‘‘Enhancing geo-service chaining through deep service
descriptions,’’ Trans. GIS, vol. 11, no. 6, pp. 849–871, 2007.
months. [14] A. Greenberg et al., ‘‘A clean slate 4D approach to network control and
NFV for the enterprise will require their platform to management,’’ ACM SIGCOMM Comput. Commun. Rev., vol. 35, no. 5,
become more comfortable embracing software L4-7 services, pp. 41–54, 2005.
[15] C. Tschudin and R. Gold, ‘‘Selnet: A translating underlay network,’’
as well as changes in their operation models. An under- Uppsala Univ., Uppsala, Sweden, Tech. Rep. 2003-020, 2001.
standing of how to optimize performance with DPDKs, and [16] D. A. Joseph, A. Tavakoli, and I. Stoica, ‘‘A policy-aware switching layer
for data centers,’’ ACM SIGCOMM Comput. Commun. Rev., vol. 38, no. 4,
potentially even looking at programmable hardware, will be pp. 51–62, 2008.
needed as well. Another challenge is the time and process it [17] J. R. Santos, Y. Turner, G. Janakiraman, and I. Pratt, ‘‘Bridging the gap
takes to re-architect monolithic services appliances that were between software and hardware techniques for I/O virtualization,’’ in
Proc. USENIX Annu. Tech. Conf., 2008, pp. 29–42.
predominantly deployed for north-south traffic. This can be [18] N. Egi, A. Greenhalgh, M. Handley, M. Hoerdt, F. Huici, and L. Mathy,
achieved by the way that there may be many more appliances, ‘‘Towards high performance virtual routers on commodity hardware,’’ in
but each supporting smaller workloads and be optimized for Proc. ACM CoNEXT Conf., 2008, Art. ID 20.
[19] M. Al-Fares, A. Loukissas, and A. Vahdat, ‘‘A scalable, commodity data
east/west traffic. center network architecture,’’ ACM SIGCOMM Comput. Commun. Rev.,
vol. 38, no. 4, pp. 63–74, 2008.
VII. CONCLUSION [20] A. Greenhalgh, F. Huici, M. Hoerdt, P. Papadimitriou, M. Handley,
and L. Mathy, ‘‘Flow processing and the rise of commodity network
In this work, we investigate a comprehensive overview of hardware,’’ ACM SIGCOMM Comput. Commun. Rev., vol. 39, no. 2,
NFV within the software-defined NFV architecture. We intro- pp. 20–26, 2009.
duce NFV its relationship with SDN. We also look at the [21] S. K. N. Rao, ‘‘SDN and its use-cases-NV and NFV,’’ Network, vol. 2,
p. H6, 2014
history of NFV, presenting how middleboxes evolve to virtual [22] A. Doria et al. ‘‘General switch management protocol (GSMP) V3,’’
network functions. In particular, we choose service chain- Tech. Rep., 2002, doi: http://dx.doi.org/10.17487/RFC3292.
ing as a typical application of NFV. Furthermore, software- [23] T. Wu, L. Rui, A. Xiong, and S. Guo, ‘‘An automation PCI allocation
method for eNodeB and home eNodeB cell,’’ in Proc. IEEE 6th Int. Conf.
defined NFV challenges and possible solutions are presented. Wireless Commun. Netw. Mobile Comput. (WiCOM), Sep. 2010, pp. 1–4.
Finally, promising research areas are illustrated and future [24] A. Berl, H. de Meer, H. Hlavacs, and T. Treutner, ‘‘Virtualization
in energy-efficient future home environments,’’ IEEE Commun. Mag.,
directions are presented. vol. 47, no. 12, pp. 62–67, Dec. 2009.
[25] R. Mortier et al., ‘‘Control and understanding: Owning your home net-
REFERENCES work,’’ in Proc. IEEE 4th Int. Conf. Commun. Syst. Netw. (COMSNETS),
Jan. 2012, pp. 1–10.
[1] J. Sherry, S. Hasan, C. Scott, A. Krishnamurthy, S. Ratnasamy, and
[26] H. Ludwig et al., ‘‘Web service level agreement (WSLA) language
V. Sekar, ‘‘Making middleboxes someone else’s problem: Network pro-
specification,’’ IBM Corp., New York, NY, USA, Tech. Rep., 2003,
cessing as a cloud service,’’ ACM SIGCOMM Comput. Commun. Rev.,
pp. 815–824.
vol. 42, no. 4, pp. 13–24, 2012. [27] F. T. Leighton and D. M. Lewin, ‘‘Content delivery network using edge-
[2] Z. Wang, Z. Qian, Q. Xu, Z. Mao, and M. Zhang, ‘‘An untold story of of-network servers for providing content delivery to a set of participating
middleboxes in cellular networks,’’ ACM SIGCOMM Comput. Commun. content providers,’’ U.S. Patent 6 553 413, Apr. 22, 2003.
Rev., vol. 41, no. 4, pp. 374–385, 2011. [28] E. D. Zwicky, S. Cooper, and D. B. Chapman, Building Internet Firewalls.
[3] M. Walfish, J. Stribling, M. Krohn, H. Balakrishnan, R. Morris, and Sebastopol, CA, USA: O’Reilly Media, 2000.
S. Shenker, ‘‘Middleboxes no longer considered harmful,’’ in Proc. 6th [29] H. Hawilo, A. Shami, M. Mirahmadi, and R. Asal, ‘‘NFV: State of
Symp. OSDI, vol. 4. 2004, p. 15. the art, challenges, and implementation in next generation mobile net-
[4] G. Schaffrath et al., ‘‘Network virtualization architecture: Proposal and works (vEPC),’’ IEEE Netw., vol. 28, no. 6, pp. 18–26, Nov./Dec. 2014.
initial prototype,’’ in Proc. 1st ACM Workshop Virtualized Infrastruct. [30] A. Gember et al. (2013). ‘‘Stratos: A network-aware orchestra-
Syst. Archit., 2009, pp. 63–72. tion layer for virtual middleboxes in clouds.’’ [Online]. Available:
[5] N. M. M. K. Chowdhury and R. Boutaba, ‘‘A survey of network http://arxiv.org/abs/1305.0209
virtualization,’’ Comput. Netw., vol. 54, no. 5, pp. 862–876, [31] J. Case, M. Fedor, M. Schoffstall, and J. Davin, A Simple Network
2010. Management Protocol (SNMP), document 1157, 1989.
[6] N. M. M. K. Chowdhury and R. Boutaba, ‘‘Network virtualization: State [32] A. Leinwand and K. F. Conroy, Network Management: A Practical Per-
of the art and research challenges,’’ IEEE Commun. Mag., vol. 47, no. 7, spective (Unix and Open Systems Series), vol. 1, 2nd ed. Reading, MA,
pp. 20–26, Jul. 2009. USA: Addison-Wesley, 1996.
[7] R. Guerzoni et al., ‘‘Network functions virtualisation: An introduction, [33] B. A. A. Nunes, M. Mendonca, X.-N. Nguyen, K. Obraczka, and
benefits, enablers, challenges & call for action,’’ in Proc. SDN OpenFlow T. Turletti, ‘‘A survey of software-defined networking: Past, present,
World Congr., 2012, pp. 1–16. and future of programmable networks,’’ HAL, Bengaluru, India,
[8] F. Yue. (2013). ‘‘Network functions virtualization—Everything old is new Tech. Rep. hal-00825087, 2013.
again,’’ Tech. Rep. [Online]. Available: http://www.f5.com/pdf/white- [34] W. Xia, Y. Wen, C. H. Foh, D. Niyato, and H. Xie, ‘‘A survey on software-
papers/service-provider-nfv-white-paper.pdf defined networking,’’ Commun. Surveys Tuts., vol. 17, no. 1, pp. 27–51,
[9] A. Manzalini et al., ‘‘Software-defined networks for future networks Jun. 2014.
and services: Main technical challenges and business implications,’’ [35] R. Jain and S. Paul, ‘‘Network virtualization and software defined net-
Tech. Rep., 2014. working for cloud computing: A survey,’’ IEEE Commun. Mag., vol. 51,
[10] S. H. Yeganeh, A. Tootoonchian, and Y. Ganjali, ‘‘On scalability of no. 11, pp. 24–31, Nov. 2013.
software-defined networking,’’ IEEE Commun. Mag., vol. 51, no. 2, [36] Open Networking Fundation, ‘‘Software-defined networking: The new
pp. 136–141, Feb. 2013. norm for networks,’’ Open Netw. Found., ONF White Paper, 2012.
[11] X. Ge, H. Cheng, M. Guizani, T. Han, ‘‘5G wireless backhaul networks: [37] M. Casado, T. Koponen, S. Shenker, and A. Tootoonchian, ‘‘Fabric:
Challenges and research advances,’’ IEEE Netw., vol. 28, no. 6, pp. 6–11, A retrospective on evolving SDN,’’ in Proc. 1st Workshop Hot Topics
Nov. 2014. Softw. Defined Netw., 2012, pp. 85–90.

VOLUME 3, 2015 2551

 Y. Li, M. Chen: Software-Defined NFV: A Survey

[38] N. Foster et al., ‘‘Frenetic: A network programming language,’’ ACM [64] A. Gember, P. Prabhu, Z. Ghadiyali, and A. Akella, ‘‘Toward software-
SIGPLAN Notices, vol. 46, no. 9, pp. 279–291, 2011. defined middlebox networking,’’ in Proc. 11th ACM Workshop Hot Topics
[39] N. Handigol, S. Seetharaman, M. Flajslik, N. McKeown, and R. Johari, Netw., 2012, pp. 7–12.
‘‘Plug-n-serve: Load-balancing Web traffic using OpenFlow,’’ in Proc. [65] A. Gember, R. Grandl, J. Khalid, and A. Akella, ‘‘Design and implemen-
ACM SIGCOMM Demo, 2009, pp. 1–2. tation of a framework for software-defined middlebox networking,’’ in
[40] A. Doria et al., Forwarding and Control Element Separation (ForCES) Proc. ACM SIGCOMM Conf. (SIGCOMM), 2013, pp. 467–468.
Protocol Specification, document 5810, 2010. [66] J. W. Anderson, R. Braud, R. Kapoor, G. Porter, and A. Vahdat, ‘‘xOMB:
[41] B. Raghavan, M. Casado, T. Koponen, S. Ratnasamy, A. Ghodsi, and Extensible open middleboxes with commodity servers,’’ in Proc. 8th
S. Shenker, ‘‘Software-defined Internet architecture: Decoupling archi- ACM/IEEE Symp. Archit. Netw. Commun. Syst., 2012, pp. 49–60.
tecture from infrastructure,’’ in Proc. 11th ACM Workshop Hot Topics [67] P. Quinn and T. Nadeau, Service Function Chaining Problem Statement,
Netw., 2012, pp. 43–48. document draft-quinn-sfc-problem-statement-02, 2013.
[42] R. Bifulco, R. Canonico, M. Brunner, P. Hasselmeyer, and F. Mir, [68] Cisco Catalyst 6500 Series Switches Solution, Cisco Syst., Inc., San Jose,
‘‘A practical experience in designing an OpenFlow controller,’’ in CA, USA, 2015.
Proc. IEEE Eur. Workshop Softw. Defined Netw. (EWSDN), Oct. 2012, [69] B. Davie and Y. Rekhter, MPLS: Technology and Applications.
pp. 61–66. San Mateo, CA, USA: Morgan Kaufmann, 2000.
[43] A. Tootoonchian, S. Gorbunov, Y. Ganjali, M. Casado, and R. Sherwood, [70] I. Stoica, D. Adkins, S. Zhuang, S. Shenker, and S. Surana, ‘‘Internet
‘‘On controller performance in software-defined networks,’’ in Proc. indirection infrastructure,’’ ACM SIGCOMM Comput. Commun. Rev.,
USENIX Workshop Hot Topics Manage. Internet, Cloud, Enterprise Netw. vol. 32, no. 4, pp. 73–86, 2002.
Services (Hot-ICE), vol. 54. 2012, pp. 1–6. [71] R. Rao, ‘‘Multi-service network switch with policy based routing,’’
[44] M. Monaco, O. Michel, and E. Keller, ‘‘Applying operating system princi- U.S. Patent 6 789 118 B1, Sep. 7, 2004.
ples to SDN controller design,’’ in Proc. 12th ACM Workshop Hot Topics [72] M. Casado, M. J. Freedman, J. Pettit, J. Luo, N. McKeown, and
Netw., 2013, Art. ID 2. S. Shenker, ‘‘Ethane: Taking control of the enterprise,’’ ACM SIGCOMM
[45] G. Lu et al., ‘‘Serverswitch: A programmable and high performance Comput. Commun. Rev., vol. 37, no. 4, pp. 1–12, 2007.
platform for data center networks,’’ in Proc. NSDI, vol. 11. 2011, [73] J. Blendin, J. Rückert, N. Leymann, G. Schyguda, and D. Hausheer,
pp. 1–14. ‘‘Position paper: Software-defined network service chaining,’’ in Proc.
[46] D. Drutskoy, E. Keller, and J. Rexford, ‘‘Scalable network virtualization 3rd EWSDN Workshop, Sep. 2014, pp. 109–114.
in software-defined networks,’’ IEEE Internet Comput., vol. 17, no. 2, [74] Z. A. Qazi, C.-C. Tu, L. Chiang, R. Miao, V. Sekar, and M. Yu,
pp. 20–27, Mar./Apr. 2013. ‘‘SIMPLE-fying middlebox policy enforcement using SDN,’’ in Proc.
[47] N. McKeown et al., ‘‘OpenFlow: Enabling innovation in campus net- ACM SIGCOMM Conf. (SIGCOMM), 2013, pp. 27–38.
works,’’ ACM SIGCOMM Comput. Commun. Rev., vol. 38, no. 2, [75] Y. Zhang et al., ‘‘StEERING: A software-defined networking for
pp. 69–74, 2008. inline service chaining,’’ in Proc. 21st IEEE ICNP, Oct. 2013,
[48] ONF Solution Brief, ‘‘OpenFlow-enabled SDN and network functions pp. 1–10.
virtualization,’’ Open Netw. Found., 2014. [76] S. K. Fayazbakhsh, L. Chiang, V. Sekar, M. Yu, and J. C. Mogul,
[49] E. Ng, ‘‘Maestro: A system for scalable OpenFlow control,’’ TSEN, ‘‘Enforcing network-wide policies in the presence of dynamic middlebox
Rice Univ., Houston, TX, USA, Maestro-Tech. Rep. TR10-08, actions using FlowTags,’’ in Proc. 11th USENIX Symp. NSDI, 2014,
2010. pp. 533–544.
[50] C. Rotsos, N. Sarrar, S. Uhlig, R. Sherwood, and A. W. Moore, [77] M. Xia, M. Shirazipour, Y. Zhang, H. Green, and A. Takacs, ‘‘SOLuTIoN:
‘‘OFLOPS: An open framework for OpenFlow switch evaluation,’’ in SDN-based optical traffic steering for NFV,’’ in Proc. 3rd Workshop Hot
Passive and Active Measurement. Berlin, Germany: Springer, 2012, Topics Softw. Defined Netw., 2014, pp. 227–228.
pp. 85–95. [78] A. Gember-Jacobson et al., ‘‘OpenNF: Enabling innovation in net-
[51] R. Wang, D. Butnariu, and J. Rexford, ‘‘OpenFlow-based server load work function control,’’ in Proc. ACM Conf. SIGCOMM, 2014,
balancing gone wild,’’ in Proc. 11th USENIX Conf. Hot Topics Manage. pp. 163–174.
Internet, Cloud, Enterprise Netw. Services, 2011, p. 12. [79] H. Masutani et al., ‘‘Requirements and design of flexible NFV network
[52] A. Dixit, F. Hao, S. Mukherjee, T. V. Lakshman, and R. Kompella, infrastructure node leveraging SDN/OpenFlow,’’ in Proc. Int. Conf. Opt.
‘‘Towards an elastic distributed SDN controller,’’ in Proc. Netw. Design Model., May 2014, pp. 258–263.
2nd ACM SIGCOMM Workshop Hot Topics Softw. Defined Netw., 2013, [80] Intel Data Plane Development Kit, Intel, Santa Clara, CA, USA, 2015.
pp. 7–12. [81] J. Hwang, K. K. Ramakrishnan, and T. Wood, ‘‘NetVM: High
[53] N. Gude et al., ‘‘NOX: Towards an operating system for networks,’’ ACM performance and flexible networking using virtualization on com-
SIGCOMM Comput. Commun. Rev., vol. 38, no. 3, pp. 105–110, 2008. modity platforms,’’ in Proc. 11th USENIX Symp. Netw. Syst. Design
[54] D. Joseph and I. Stoica, ‘‘Modeling middleboxes,’’ IEEE Netw., vol. 22, Implement. (NSDI), 2014, pp. 1–12.
no. 5, pp. 20–25, Sep./Oct. 2008. [82] J. Martins, M. Ahmed, C. Raiciu, and F. Huici, ‘‘Enabling fast, dynamic
[55] M. Allman, ‘‘On the performance of middleboxes,’’ in Proc. 3rd ACM network processing with ClickOS,’’ in Proc. 2nd ACM SIGCOMM
SIGCOMM Conf. Internet Meas., 2003, pp. 307–312. Workshop Hot Topics Softw. Defined Netw., 2013, pp. 67–72.
[56] V. Sekar, S. Ratnasamy, M. K. Reiter, N. Egi, and G. Shi, ‘‘The middlebox [83] D. King and C. Ford, ‘‘A critical survey of network functions
manifesto: Enabling innovation in middlebox deployment,’’ in Proc. 10th virtualization (NFV),’’ in Proc. iPOP, 2013, pp. 1–21.
ACM Workshop Hot Topics Netw., 2011, Art. ID 21. [84] W. Shen, M. Yoshida, T. Kawabata, K. Minato, and W. Imajuku,
[57] J. Martins et al., ‘‘ClickOS and the art of network function virtualization,’’ ‘‘vConductor: An NFV management solution for realizing end-to-end
in Proc. 11th USENIX Symp. Netw. Syst. Design Implement. (NSDI), virtual network services,’’ in Proc. 16th Asia-Pacific Netw. Oper. Manage.
Seattle, WA, USA, 2014, pp. 459–473. Symp. (APNOMS), Sep. 2014, pp. 1–6.
[58] Network Working Group Published in Computer Communication [85] M. Yoshida, W. Shen, T. Kawabata, K. Minato, and W. Imajuku,
Review, K. Egevang, and P. Francis, ‘‘The IP network address translator ‘‘MORSA: A multi-objective resource scheduling algorithm for NFV
(NAT),’’ Tech. Rep. RFC 1631, May 1994. infrastructure,’’ in Proc. 16th APNOMS, Sep. 2014, pp. 1–6.
[59] C. H. Rowland, ‘‘Intrusion detection system,’’ U.S. Patent 6 405 318, [86] M. Bouet, J. Leguay, and V. Conan, ‘‘Cost-based placement of virtual-
Jun. 11, 2002. ized deep packet inspection functions in SDN,’’ in Proc. IEEE Military
[60] M. K. Bowman-Amuah, ‘‘Load balancer in environment services pat- Commun. Conf. (MILCOM), Nov. 2013, pp. 992–997.
terns,’’ U.S. Patent 6 578 068, Jun. 10, 2003. [87] M. Scholler, M. Stiemerling, A. Ripke, and R. Bless, ‘‘Resilient deploy-
[61] V. Sekar, N. Egi, S. Ratnasamy, M. K. Reiter, and G. Shi, ‘‘Design and ment of virtual network functions,’’ in Proc. 5th Int. Congr. Ultra
implementation of a consolidated middlebox architecture,’’ in Proc. 9th Modern Telecommun. Control Syst. Workshops (ICUMT), Sep. 2013,
USENIX Conf. NSDI, 2012, pp. 323–336. pp. 208–214.
[62] J. Sherry, ‘‘Future architectures for middlebox processing services on the [88] S. Clayman, E. Maini, A. Galis, A. Manzalini, and N. Mazzocca,
Internet and in the cloud,’’ M.S. thesis, Dept. EECS, Univ. California, ‘‘The dynamic placement of virtual network functions,’’ in Proc. IEEE
Berkeley, Berkeley, CA, USA, 2012. Netw. Oper. Manage. Symp. (NOMS), May 2014, pp. 1–9.
[63] J. Lee, J. Tourrilhes, P. Sharma, and S. Banerjee, ‘‘No more middlebox: [89] R. Cannistra et al., ‘‘Enabling autonomic provisioning in SDN cloud
Integrate processing into network,’’ ACM SIGCOMM Comput. Commun. networks with NFV service chaining,’’ in Proc. Opt. Fiber Commun.
Rev., vol. 40, no. 4, pp. 459–460, 2010. Conf. Exhibit., Mar. 2014, pp. 1–3, paper Tu2I-4.

2552 VOLUME 3, 2015

Y. Li, M. Chen: Software-Defined NFV: A Survey

[90] T. Benson, A. Akella, and D. A. Maltz, ‘‘Mining policies from enterprise [116] Y. Niu, Y. Li, M. Chen, D. Jin, and S. Chen, ‘‘A cross-layer design
network configuration,’’ in Proc. 9th ACM SIGCOMM Conf. Internet for software defined millimeter-wave mobile broadband system,’’ IEEE
Meas. Conf., 2009, pp. 136–142. Commun. Mag., pp. 1–18, Sep. 2015.
[91] N. Kang, J. Reich, J. Rexford, and D. Walker, ‘‘Policy transformation [117] Y. Li, P. Hui, D. Jin, and S. Chen, ‘‘Delay-tolerant network protocol
in software defined networks,’’ in Proc. ACM SIGCOMM Conf. Appl., testing and evaluation,’’ IEEE Commun. Mag., vol. 53, no. 1,
Technol., Archit., Protocols Comput. Commun., 2012, pp. 309–310. pp. 258–266, Jan. 2015.
[92] M. Armbrust et al., ‘‘A view of cloud computing,’’ Commun. ACM, [118] Z. Zhu et al., ‘‘Virtual base station pool: Towards a wireless network
vol. 53, no. 4, pp. 50–58, 2010. cloud for radio access networks,’’ in Proc. 8th ACM Int. Conf. Comput.
[93] Z. Liu, Y. Li, D. Jin, L. Su, and L. Zeng, ‘‘M2cloud: Software defined Frontiers, 2011, Art. ID 34.
multi-site data center network control framework for multi-tenant,’’ [119] G. Bhanage, I. Seskar, R. Mahindra, and D. Raychaudhuri, ‘‘Virtual
ACM SIGCOMM Comput. Commun. Rev., vol. 43, no. 4, pp. 517–518, basestation: Architecture for an open shared WiMAX framework,’’ in
2013. Proc. 2nd ACM SIGCOMM Workshop Virtualized Infrastruct. Syst.
[94] L. Qian, Z. Luo, Y. Du, and L. Guo, ‘‘Cloud computing: An overview,’’ Archit., 2010, pp. 1–8.
in Cloud Computing. Berlin, Germany: Springer, 2009, pp. 626–631. [120] B. Pfaff, J. Pettit, T. Koponen, K. Amidon, M. Casado, and S. Shenker,
[95] H. Wang, Y. Li, Y. Zhang, and D. Jin, ‘‘Virtual machine migration ‘‘Extending networking into the virtualization layer,’’ in Proc. HotNets,
planning in software-defined networks,’’ in Proc. IEEE Conf. INFOCOM, 2009, pp. 1–6.
Apr./May 2015, pp. 487–495. [121] B. Rimler and N. Rasmussen, ‘‘Mobile data center,’’
[96] J. Liu, Y. Li, D. Jin, L. Su, and L. Zeng, ‘‘Traffic aware cross-site virtual U.S. Patent 2006 0 082 263 A1, Apr. 20, 2006.
machine migration in future mobile cloud computing,’’ Mobile Netw. [122] Y. Zhou, X. Yang, Y. Li, D. Jin, L. Su, and L. Zeng, ‘‘Incremental
Appl., vol. 20, no. 1, pp. 62–71, Feb. 2015. re-embedding scheme for evolving virtual network requests,’’ IEEE
[97] M. Chen, Y. Hao, Y. Li, C. Lai, and D. Wu, ‘‘On the computation Commun. Lett., vol. 17, no. 5, pp. 1016–1019, May 2013.
offloading at ad hoc cloudlet: Architecture and service models,’’ IEEE [123] A. K. Nayak, A. Reimers, N. Feamster, and R. Clark, ‘‘Resonance:
Commun., vol. 53, no. 6, pp. 18–24, Jun. 2015. Dynamic access control for enterprise networks,’’ in Proc. 1st ACM
[98] J. Liu, Y. Li, and D. Jin, ‘‘SDN-based live VM migration across datacen- Workshop Res. Enterprise Netw., 2009, pp. 11–18.
ters,’’ in Proc. ACM Conf. SIGCOMM, 2014, vol. 44. no. 4, pp. 583–584. [124] X. Ge, B. Yang, J. Ye, G. Mao, C.-X. Wang, and T. Han, ‘‘Spatial spec-
[99] F. Derakhshan, H. Grob-Lipski, H. Roessler, P. Schefczik, and trum and energy efficiency of random cellular networks,’’ IEEE Trans.
M. Soellner, ‘‘Enabling cloud connectivity using SDN and NFV tech- Commun., vol. 63, no. 3, pp. 1019–1030, Mar. 2015.
nologies,’’ in Mobile Networks and Management. Springer International [125] Z. Liu, Y. Li, B. Cui, L. Su, D. Jin, and L. Zeng, ‘‘GrainFlow:
Publishing, 2013, pp. 245–258. Enable testing for future Internet architectures by per-bit customization,’’
[100] M. Chen, S. Gonzalez, Q. Zhang, and V. Leung, ‘‘Code-centric RFID Comput. Netw., vol. 69, pp. 121–132, Aug. 2014.
system based on software agent intelligence,’’ IEEE Intell. Syst., vol. 25,
no. 2, pp. 12–19, Mar./Apr. 2010.
[101] CloudNFV. [Online]. Available: http://www.cloudnfv.com/, accessed
2015. YONG LI (M’09) received the B.S. degree in
[102] Combining Cloud, NFV, and Service Provider SDN, ‘‘The real-time electronics and information engineering from the
cloud,’’ Ericsson, Stockholm, Sweden, White Paper Uen 284 23-3219, Huazhong University of Science and Technology,
Feb. 2014. Wuhan, China, in 2007, and the Ph.D. degree in
[103] Alcatel-CloudBand. [Online]. Available: http://www.alcatel-lucent.com/ electronics engineering from Tsinghua Univer-
solutions/cloudband, accessed 2015. sity, Beijing, China, in 2012. In 2012 and 2013,
[104] M. Yang, Y. Li, D. Jin, L. Zeng, X. Wu, and A. V. Vasilakos, he was a Visiting Research Associate with
‘‘Software-defined and virtualized future mobile and wireless networks: Telekom Innovation Laboratories and The
A survey,’’ Mobile Netw. Appl., vol. 20, no. 1, pp. 4–18, Feb. 2015. Hong Kong University of Science and Technology,
[105] S. Paul and R. Jain, ‘‘OpenADN: Mobile apps on global clouds using
respectively. From 2013 to 2014, he was a Visiting
OpenFlow and software defined networking,’’ in Proc. IEEE Globecom
Workshops (GC Wkshps), Dec. 2012, pp. 719–723.
Scientist with the University of Miami. He is currently a Faculty Mem-
[106] J. Liu, Y. Li, M. Chen, W. Dong, and D. Jin, ‘‘Software-defined Internet ber with the Department of Electronic Engineering, Tsinghua University.
of Things for smart urban sensing,’’ IEEE Commun. Mag., vol. 53, no. 8, His research interests are in the areas of networking and communications.
pp. 55–63, Sep. 2015.
[107] M. Yang et al., ‘‘Cross-layer software-defined 5G network,’’ Mobile
Netw. Appl., vol. 20, no. 3, pp. 400–409, 2015. MIN CHEN (M’08–SM’09) was an Assistant Pro-
[108] X. Jin, L. E. Li, L. Vanbever, and J. Rexford, ‘‘SoftCell: Scalable and fessor with the School of Computer Science and
flexible cellular core network architecture,’’ in Proc. 9th ACM Conf. Engineering, Seoul National University (SNU),
Emerg. Netw. Experim. Technol., 2013, pp. 163–174.
from 2009 to 2012. He was a Post-Doctoral
[109] M. Moradi, L. E. Li, and Z. M. Mao, ‘‘SoftMoW: A dynamic and scalable
software defined architecture for cellular WANs,’’ in Proc. 3rd Workshop
Fellow with the Department of Electrical and
Hot Topics Softw. Defined Netw., 2014, pp. 201–202. Computer Engineering, University of British
[110] L. E. Li, Z. M. Mao, and J. Rexford, ‘‘CellSDN: Software-defined cellular Columbia (UBC), for three years. Before joining
networks,’’ Dept. Comput. Sci., Princeton Univ., Princeton, NJ, USA, UBC, he was a Post-Doctoral Fellow with SNU for
Tech. Rep., 2012. one and a half years. He is currently a Professor
[111] A. Basta, A. Blenk, M. Hoffmann, H. J. Morper, K. Hoffmann, and with the School of Computer Science and Tech-
W. Kellerer, ‘‘SDN and NFV dynamic operation of LTE EPC gateways nology, Huazhong University of Science and Technology (HUST), where he
for time-varying traffic patterns,’’ in Proc. 6th Int. Conf. Mobile Netw. is also the Director of the Embedded and Pervasive Computing Laboratory.
Manage., Sep. 2014, pp. 63–76. He has authored over 260 paper publications, including over 100 SCI papers,
[112] A. Basta, W. Kellerer, M. Hoffmann, H. J. Morper, and K. Hoffmann, over 50 IEEE TRANSACTIONS/journal papers, six ISI highly cited papers, and
‘‘Applying NFV and SDN to LTE mobile core gateways. The functions one hot paper. He has also authored the books entitled Internet of Things
placement problem,’’ in Proc. 4th Workshop All Things Cellular, Oper., (IoT): OPNET IoT Simulation (HUST Press, 2015) and Big Data Related
Appl., Challenges, 2014, pp. 33–38.
Technologies (Springer Series in Computer Science, 2014) in big data. His
[113] J. De Vriendt, P. Laine, C. Lerouge, and X. Xu, ‘‘Mobile network
evolution: A revolution on the move,’’ IEEE Commun. Mag., vol. 40,
Google Scholars Citations reached over 5300 with an h-index of 34. His top
no. 4, pp. 104–111, Apr. 2002. paper was cited 648 times, while his top book was cited 420 times in 2015.
[114] T. C. Y. Wang and S. H. Moritz, ‘‘Mobile unit tracking system,’’ His research focuses on IoT, mobile cloud, body area networks, emotion-
U.S. Patent 5 365 451, Nov. 15, 1994. aware computing, healthcare big data, cyber physical systems, and robotics.
[115] M. Yang, Y. Li, D. Jin, L. Su, S. Ma, and L. Zeng, ‘‘OpenRAN: He received the best paper award from the IEEE ICC 2012, and the Best
A software-defined ran architecture via virtualization,’’ ACM SIGCOMM Paper Runner-Up Award from QShine 2008.
Comput. Commun. Rev., vol. 43, no. 4, pp. 549–550, 2013.

VOLUME 3, 2015 2553