Content uploaded by Qiang Duan
Author content
All content in this area was uploaded by Qiang Duan on Dec 03, 2020
Content may be subject to copyright.
IEEE Network • Novembe/December 20 20
148 0890-8044/20/$25.00 © 2020 IEEE
AbstrAct
The wide applications of virtualization and
service-oriented principles in various emerging
networking technologies introduce a trend of net-
work cloudification that enables network systems
to be realized based on cloud technologies and
allows network services to be provisioned follow-
ing the cloud service model. Network cloudifica-
tion together with the critical role of networking
in the latest cloud/edge computing technologies
leads to the convergence of networking and
cloud/edge computing, which calls for a holistic
vision across the fields of networking and com-
puting that may shape relevant technology devel-
opments. In this article, we attempt to sketch a
big picture to reflect the current status of on-go-
ing research toward network-cloud/edge con-
vergence. We first describe the notion of such
convergence and present an architectural frame-
work for converged network-cloud/edge systems.
Then, we survey the state of the art of enabling
technologies for network-cloud/edge conver-
gence by reviewing recent progress in relevant
standardization and technology developments in
representative research projects. We also discuss
challenges that must be fully addressed for real-
izing the convergence of networking and cloud/
edge computing and identify some opportunities
for future research in this exciting interdisciplinary
field.
IntroductIon
A main technical strategy employed by the net-
working research community for building the
future Internet lies in leveraging the virtualiza-
tion and service-oriented principles in network
architecture and service models. Virtualization,
which essentially separates system functions from
their implementations, has been adopted as a key
principle for cloud computing. Network virtualiza-
tion decouples network service provisioning from
data transport/process capabilities, thus enabling
virtual networks customized for multi-tenant
requirements to share a common infrastructure. A
milestone of network virtualization is the Network
Function Virtualization (NFV) architecture devel-
oped by ETSI, which realizes network functions
as software instances that can be deployed on
commodity servers and storage [1].
The service-oriented principle realized through
the Everything-as-a-Service (XaaS) paradigm (e.g.,
IaaS, SaaS, PaaS, and so on) is also a key to cloud
computing. When applied in networking this prin-
ciple allows abstraction of network resources and
functions as self-contained components that can
be exposed, accessed, and composed for service
provisioning through loose-coupling interfaces.
NFV embraces the service-oriented principle and
supports XaaS at various levels including NFV
Infrastructure-as-a-Service (NFVIaaS), Virtual Net-
work Function-as-a-Service (VNFaaS), and Vir-
tual Network-as-a-Service (VNaaS). 3GPP has
proposed a service-based architecture for 5G net-
works that encapsulates network functions as ser-
vice components to support network slicing [2].
The virtualization and service-oriented prin-
ciples in network design introduce a trend of
cloudification in networking that enables network
systems to be realized based on cloud technol-
ogies and network services to be provisioned
following the cloud service model. On the other
hand, networking has been the foundation for
cloud data centers and a key to cloud service
provisioning. The merging edge computing par-
adigm, which essentially embeds decentralized
cloud capabilities into network infrastructures
mainly at the network edge, requires networking
as an indispensable component [3].
The trend of network cloudification and the
critical role of networking in cloud/edge comput-
ing are enabling the convergence of these two
fields that used to be relatively independent. Such
convergence leads to a holistic vision across net-
working and computing that allows integrated
resource/function management across network
and cloud/edge systems and unified provision-
ing of network and cloud/edge services [4]. The
convergence may benefit stakeholders in the net-
work-cloud/edge ecosystem through improved
resource utilization, more flexible service man-
agement, and enhanced service performance.
The interdisciplinary nature of such convergence
may stimulate innovations in networking/com-
puting technologies, and therefore has attract-
ed extensive interest from both academia and
industry. Although research progress toward net-
work-cloud/edge convergence has been report-
ed in various literatures, we feel that a survey
reflecting a big picture of ongoing research in this
exciting field would be beneficial to the research
community.
Convergence of Networking and Cloud/Edge Computing: Status, Challenges, and
Opportunities
Qiang Duan, Shangguang Wang, and Nirwan Ansari
ACCEPTED FROM OPEN CALL
Digital Object Identifier:
10.1109/MNET.011.2000089
Qiang Duan is with Pennsylvania State University; Shangguang Wang is with Beijing University of Posts and Telecommunications;
Nirwan Ansari is with New Jersey Institute of Technology.
DUAN_LAYOUT.indd 148DUAN_LAYOUT.indd 148 11/19/20 3:59 PM11/19/20 3:59 PM
Authorized licensed use limited to: Penn State University. Downloaded on December 03,2020 at 23:15:34 UTC from IEEE Xplore. Restrictions apply.
IEEE Network • Novembe/December 20 20 149
In this article, we attempt to review the current
status of research on network-cloud/edge conver-
gence and identify some challenges/opportunities
for future work. We first describe the notion of
network-cloud/edge convergence and present
an architectural framework. Then, we summarize
recent advances in related standardization and
review key technologies developed in some rep-
resentative projects. We discuss challenges and
research opportunities and draw conclusions in
the final section
convergence of
networkIng And cloud/edge computIng
The ongoing network cloudification is replacing
specially designed network appliances with data
center-like systems constructed with commodity
servers and storage upon which virtual network
functions can be deployed as software instances
and composed as service components for service
provisioning. Therefore, networks are being trans-
formed from infrastructures dedicated to data
communications to versatile platforms for sup-
porting both networking and computing services.
On the other hand, network cloudification adopts
the cloud service model for network service deliv-
ery, which allows the data centers and edge serv-
ers originally built for cloud/edge computing to
be utilized for hosting virtual network functions.
Networking and cloud/edge computing,
which used to focus on data communications
and data process/storage respectively, are now
merging together, thus calling for a holistic vision
across these two fields for various technology
developments. For example, the heterogeneous
infrastructure resources for data transport, pro-
cess, and storage should be managed by unified
mechanisms and utilized through a common
abstraction layer by various virtual functions. The
services delivered to end users should be com-
posite network-cloud/edge services provisioned
through federated orchestration of diverse net-
work/compute functions offered by different ser-
vice providers. In this article, we refer to the trend
in technology developments toward a merging
field of networking-cloud/edge computing with a
holistic vision of system design, management, and
operation as the convergence of networking and
cloud/edge computing.
An architectural framework for a converged
network-cloud/edge system is depicted in Fig. 1.
The infrastructure layer in this framework consists
of multiple admin-domains operated by different
infrastructure providers. Each admin-domain may
be composed of multiple tech-domains, each of
which comprises a certain type of infrastructure
for networking, computing, or storage. The vir-
tualization layer provides a common abstraction
of the heterogeneous computing and networking
infrastructure resources and exposes the virtual
resources via the IaaS paradigm. Various virtual
network functions (VNFs) and virtual compute
functions (VCFs) are realized on the virtual func-
tion layer by leveraging infrastructure services.
VNFs and VCFs are orchestrated as service com-
ponents on the service layer to provision com-
posite services for supporting multi-tenant users/
applications.
This framework indicates that common virtu-
alization and service-oriented abstraction across
networking and computing domains form a foun-
dation for network-cloud/edge convergence.
Infrastructure-level virtualization offers a general
abstraction of heterogenous network-compute
resources that allows VNFs/VCFs to be realized
through leveraging various infrastructure ser-
vices. Function-level virtualization abstracts both
VNFs and VCFs as service components that
can be orchestrated through common mecha-
nisms to enable federated network-cloud/edge
service management. Upon such a foundation,
inter-domain cooperations among networking
and computing systems play a crucial role in net-
work-cloud/edge convergence, which demands
new technologies for i) unified resource manage-
ment across tech-domains comprising heterog-
enous network and compute infrastructures; ii)
FIGURE 1. An Architectural Framework for convergence of networking and cloud/edge computing.
Data Plane
multi-tenant
user/application
Control/Management Plane
Infrastructure Virtualization Layer
Infrastructure Layer
Virtual Network
Function
VNF-1
composite
network-cloud/edge
service-1
Virtual Function
management / orchestration
Service Layer
admin domain-1
infrastructure resource
management / orchestration
admin domain-n
Network/Compute IaaS
Virtual Function Layer
Virtual Function-as-a-Service
unified network-cloud/edge
service management /
orchestration
network
resources
compute/storage
resources
network
resources
compute/storage
resources
Virtual Compute
Function
VCF-1
Virtual Network
Function
VNF-n
Virtual Compute
Function
VCF-m
composite
network-cloud/edge
service-k
multi-tenant
user/application
multi-tenant
user/application
DUAN_LAYOUT.indd 149DUAN_LAYOUT.indd 149 11/19/20 3:59 PM11/19/20 3:59 PM
Authorized licensed use limited to: Penn State University. Downloaded on December 03,2020 at 23:15:34 UTC from IEEE Xplore. Restrictions apply.
IEEE Network • Novembe/December 20 20
150
federated service management across admin-do-
mains operated by different network and cloud/
edge service providers; and iii) holistic life-cycle
management of virtual network/compute func-
tions and network/cloud/edge services.
Network-cloud/edge convergence is expect-
ed to benefit various skateholders in the entire
ecosystem of networking and cloud/edge com-
puting. The holistic vision for architectural design,
resource management, system operations, and
service provisioning across networking and com-
puting fields may significantly improve resource
utilization and service performance, lower capital/
operational costs and generate new revenues, and
introduce new service models that stimulate inno-
vations and create new business opportunities.
In the following two sections, we attempt to
present the current status of research on net-
work-cloud/edge convergence by reviewing
recent progress in relevant standardization and
developments in some representative projects.
stAndArdIzAtIon progress towArd
network-cloud/edge convergence
etsI network functIon vIrtuAlIzAtIon (nfv)
ETSI ISG NFV is a main driving force for network
cloudification and offers some key technologies
for enabling network-cloud/edge convergence. In
the NFV architecture, network and compute infra-
structures are abstracted by a common virtualiza-
tion layer through which virtual resources may be
leveraged for realizing VNFs. Management and
Orchestration (MANO) is responsible for service/
resource management and orchestration. Virtual-
ized Infrastructure Manager (VIM) in MANO sup-
ports unified management of network-compute
resources. The NFV Orchestrator (NFVO) orches-
trates VNFs as service components for end-to-end
service provisioning [5].
Recent developments in NFV further enhance
its support for network-cloud/edge convergence.
NFV Release-3 [6] includes enhancements for
service provisioning spanning multiple admin-do-
mains and new interaction mechanisms between
MANOs for inter-domain service orchestration.
NFV now supports cloud-native VNF implemen-
tations by leveraging container-based virtualiza-
tion and the micro-services architecture. MANO
has also been enhanced for deploying service
components across the access, transport, and
core networks; this is critical for composite net-
work-cloud/edge service provisioning.
etsI multI-Access edge computIng (mec)
The MEC architecture developed by IETF ISG
MEC essentially embeds decentralized cloud
capabilities in the network infrastructure at the
network edge and thus can be regarded as a rep-
resentative case of convergence between net-
working and computing. The MEC architecture
comprises three levels: networks, MEC host, and
MEC system. The compute, storage, and network
resources at the MEC host are abstracted by a
common virtualization layer and leveraged by
the MEC platform for running MEC applications
[3]. Such a virtualization layer enables unified
resource abstraction and management across net-
work and compute domains thus forming a basis
for network-cloud/edge convergence.
The true impact of edge computing relies on
its interaction with networking including orches-
tration between computing and networking capa-
bilities. Recent developments in MEC support
integration of MEC and NFV, which allows MEC
applications and VNFs to share the same virtual
infrastructure and NFV MANO to be leveraged
for MEC management and orchestration. With
MEC-NFV integration, recent progress in NFV
for enabling network-cloud/edge convergence is
embraced in the MEC architecture.
Ietf servIce functIon chAInIng (sfc)
The SFC notion developed by IETF also offers
enabling technologies for network-cloud/edge
convergence. An SFC is defined as an abstract
view of service comprising a set of required ser-
vice functions and their execution order. The
IETF SFC architecture comprises service classifi-
cation functions (SCFs), service function forward-
ers (SFFs), and service functions (SFs) as the key
components, and provides a mechanism to form
service paths for traffic steering through SFs [7].
It has been a consensus that SFC and NFV com-
plement each other. NFV enables flexible SFC
through virtualization and orchestration of service
functions while SFC offers a service provisioning
model in NFV.
Recent developments in SFC include a hier-
archical SFC architecture that supports ser-
vice provisioning across multiple tech- and/or
admin-domains, which may facilitate orchestration
of network and cloud/edge services toward their
convergence. Another important piece of work-
in-progress is SFC in edge computing, including
distributed service discovery and management
mechanisms that are applicable in edge comput-
ing environments.
mef lIfecycle servIce orchestrAtIon (lso)
MEF LSO attempts to enable flexible service
orchestration across multiple provider networks as
well as different tech-domains [8]. The LSO archi-
tecture comprises four layers from top to bottom,
business applications, service orchestration func-
tionality, infrastructure control and management,
and element control and management, with ser-
vice-oriented abstraction between the layers. LSO
supports inter-domain service orchestration via
APIs between business applications in different
domains and interfaces between domain orches-
trators.
The flexible inter-domain service orches-
tration enabled by LSO supports composite
network-cloud service provisioning, which is con-
sidered as a main application scenario of LSO. In
addition, the Cloud Service Architecture devel-
oped by OpenCloud Connect1 considers applica-
tion and connectivity as the key elements of the
inter-cloud interface for integrating cloud applica-
tions and network connections.
ETSI ISG NFV is a main driving force for network cloudification and oers some key technologies
for enabling network-cloud/edge convergence. In the NFV architecture, network and compute
infrastructures are abstracted by a common virtualization layer through which virtual resources
may be leveraged for realizing VNFs.
1 OpenCloud Connect is an
independent organization
under the MEF umbrella.
DUAN_LAYOUT.indd 150DUAN_LAYOUT.indd 150 11/19/20 3:59 PM11/19/20 3:59 PM
Authorized licensed use limited to: Penn State University. Downloaded on December 03,2020 at 23:15:34 UTC from IEEE Xplore. Restrictions apply.
IEEE Network • Novembe/December 20 20 151
3gpp/5g-ppp 5g network slIcIng
The 5G network architecture developed in the
3GPP/5G-PPP community identifies network
slicing as a key to realizing multi-tenant virtual
networks upon shared network-compute infra-
structures for supporting various vertical applica-
tions [9]. The 5G architecture adopts NFV with
the SDN paradigm as the foundation and leverag-
es virtualization, softwarization, programmability,
and service-based architecture as enablers of net-
work slicing. The VNFIaaS, VNFaaS, and VNaaS
paradigms are also applied in 5G architecture for
service provisioning.
Network slicing offers a promising approach to
convergence of networking and cloud/edge com-
puting. 5G architecture emphasizes an end-to-end
slicing perspective that spans over tech-domains
comprising network/compute infrastructures and
admin-domains operated by network, edge, and
cloud service providers. Toward this direction,
3GPP/5G-PPP is developing a common abstrac-
tion layer for unifying network-compute resource
virtualization and an inter-domain orchestra-
tion framework for supporting composite net-
work-cloud/edge service provisioning. The 5G
network is expected to be a main platform for
edge service delivery and thus forms a base for
network-cloud/edge convergence.
representAtIve projects relAted to
network-cloud/edge convergence
unIfy
UNIFY is an EU-funded FP7 project that aims at
unifying cloud systems and carrier networks in a
common architecture [10]. The UNIFY architecture
comprises an infrastructure layer (IL), an orches-
tration layer (OL), and a service layer (SL). The IL
consists of multiple tech-domains of network and
cloud infrastructures and each domain has its own
controller. The OL contains a resource orchestra-
tor (RO) upon a controller adaption (CA) sub-lay-
er. The CA provides an interface between each
domain controller and the RO, which is respon-
sible for inter-domain resource orchestration to
support end-to-end service provisioning.
UNIFY employs a hierarchical structure for
resource management across infrastructure
domains. The RO is a global orchestrator that
coordinates a set of domain controllers for com-
posite network-cloud service provisioning. IaaS
is applied on the CA sub-layer to enable com-
mon abstraction of heterogeneous infrastructure
resources. The unified resource management of
network-compute infrastructures in UNIFY pro-
vides a foundation for network-cloud/edge con-
vergence. On the other hand, the assumption of
a single administrator in UNIFY limits its capability
of inter-domain service-orchestration.
t-novA
The objective of the T-NOVA project is to devel-
op an architecture for provisioning, managing,
monitoring, and optimizing VNFs over integrated
network-compute infrastructures for composite
service delivery [11]. The T-NOVA architecture
comprises the layers of infrastructure, infrastruc-
ture management, orchestration, and marketplace
from bottom to top. The infrastructure man-
agement layer coordinates the management of
datacenter infrastructures and inter-datacenter
network connections. The orchestration layer
comprises a service orchestrator and a resource
orchestrator for end-to-end service provisioning
spanning over different tech-domains of network-
ing and computing.
Following the service-oriented principle,
T-NOVA realizes a Network Function-as-Service
(NFaaS) paradigm that enables VNFs to be pub-
lished as service components that can be selected
and composed by a broker on the marketplace
layer. The brokerage platform allows end-to-end
services to be provisioned through composing
VNF service components. Although T-NOVA
currently focuses on network services, its unified
management of heterogeneous infrastructure
resources and brokerage/orchestration across dif-
ferent service domains support network-cloud/
edge convergence.
cord
CORD (Central Office Rearchitected as a
Data-center) is an ONF (Open Network Foun-
dation) project for reinventing architecture of
central offices at the network edge by leveraging
data center technologies [12]. The CORD proj-
ect unifies NFV, SDN, and cloud technologies
on both the infrastructure and service layers. The
infrastructure layer consists of servers/storage for
hosting VNF instances and a leaf-spine network
fabric. OpenStack, Docker and Kubernetes are
employed for virtual infrastructure management.
The network fabric is implemented as an SDN
with an ONOS controller. Service orchestration
is realized by the XOS module that composes the
infrastructure services provided by OpenStack/
Kubernetes, control services provided by ONOS,
and other network and cloud services.
CORD adopts the XaaS paradigm for unified
resource abstraction and service orchestration.
VNFaaS allows VNFs to be deployed in the same
way as cloud services upon the infrastructure layer
via the IaaS paradigm. The CORD architecture sup-
ports container-based VNF instances required by
the micro-services architecture. Through cloudifica-
tion of central offices located at the network edge,
CORD may significantly enhance the capability of
carrier networks for hosting edge applications, thus
supporting the convergence of networking and
cloud/edge computing. Based on the CORD plat-
form, ONF has developed a reference architecture
M-CORD (https://www.opennetworking.org/m-
cord/) as a cloud-native solution for virtualization
of RAN and mobile core (including vEPC) in 5G
networks to enable mobile edge applications/ser-
vices using a micro-services architecture.
5g exchAnge (5gex)
A main objective of the 5GEx project is to realize
an exchange framework for orchestration of net-
work and cloud resources over multiple tech- and
The 5G architecture adopts NFV with the SDN paradigm as the foundation and leverages vir tualization,
softwarization, programmability, and service-based architecture as enablers of network slicing. The
VNFIaaS, VNFaaS, and VNaaS paradigms are also applied in 5G architecture for service provisioning.
DUAN_LAYOUT.indd 151DUAN_LAYOUT.indd 151 11/19/20 3:59 PM11/19/20 3:59 PM
Authorized licensed use limited to: Penn State University. Downloaded on December 03,2020 at 23:15:34 UTC from IEEE Xplore. Restrictions apply.
IEEE Network • Novembe/December 20 20
152
admin-domains in 5G networks [13]. The 5GEx
architecture is based on a three-layer model com-
prising:
• Multi-operator wholesale relationship that
enables cooperation among service provid-
ers.
• Multi-vendor inter-operation with orchestra-
tion capability over multiple tech-domains.
• An infrastructure layer of network, compute,
and storage resources.
5GEx employs a decentralized cascade approach
for inter-provider service orchestration; each pro-
vider acts as a service reseller to customers but
the delivered services may contain sub-services
and/or resources from other providers.
5GEx introduces a Slice-as-a-Service (SlaaS)
paradigm, which combines NFVIaaS, VNFaaS,
and connectivity services for composite net-
work-cloud/edge service provisioning. Vari-
ous technologies employed in 5GEx, including
fully software-driven design, network-compute
resource integration in individual services, and
automatic resource trading and orchestration
for service delivery, overcome the segregation
between networking and computing, and may
thus greatly facilitate network-cloud/edge conver-
gence.
5g-trAnsformer
The 5G-Transformer project aims to transform
today’s mobile networks into an SDN/NFV-based
networking-computing platform upon which net-
work slices can be constructed for supporting var-
ious vertical industries [14]. 5G-Transform takes a
twofold technical approach: enable customized
network slices for meeting vertical requirements,
and integrate networking-computing resources
throughout the virtual infrastructure from the
access network to the core network and then to
cloud data centers. The 5G-Transformer archi-
tecture comprises a Vertical Slicer (5G-VS) for
creating and managing network slices, a Service
Orchestrator (5G-SO) for service orchestration
across tech- and/or admin-domains, and a Mobile
Transport and Computing Platform (5G-MTP) as
the underlying virtual infrastructure.
5G-Transform architecture supports service
deployment upon multiple tech-domains and ser-
vice orchestration across admin-domains, which
are key enablers for network-cloud/edge conver-
gence. 5G-Transform employs a hierarchical struc-
ture for unified resource management, in which
an orchestrator coordinates the managers in dif-
ferent tech-domains comprising network, com-
pute, and storage infrastructures. 5G-Transform
supports federation between the service orches-
trators in different admin-domains at both the ser-
vice and resource levels for composite service
provisioning. 5G-Transform includes edge applica-
tion management in its architecture, thus support-
ing the integration of MEC in 5G networks.
compArIson of representAtIve reseArch projects
A comparison of the reviewed projects shows
some trends in key technologies for net-
work-cloud/edge convergence, as summarized
in Table 1.
Service-oriented abstraction is employed in all
the projects for achieving common virtualization
of networking and computing resources/func-
tions. We noticed a trend of applying the XaaS
paradigm from underlying infrastructures (IaaS
in UNIFY) to virtual functions (VNFaaS in T-NO-
VA) and then to network services and slices
(XaaS in CORD and Slice-as-a-Service in 5GEx
and 5G-Transform), each providing a higher-level
abstraction based on all the lower-level abstrac-
tions beneath it.
For unified resource management across het-
erogeneous tech-domains, all the reviewed proj-
ects follow a hierarchical structure as illustrated
by Fig. 2 in which each tech-domain has a control-
ler for managing a certain type of infrastructure
resources and a global orchestrator coordinates
the tech-domain controllers for inter-domain
resource management.
A variety of approaches have been pro-
posed for federated service management across
admin-domains (service providers).2 T-NOVA
employs a centralized broker for selecting and
composing VNFs offered by different provid-
ers to provision composite services. The more
recent 5G-Exchange and 5G-Transform projects
both advocate a decentralized model for service
orchestration/federation across the admin-do-
mains owned by different service providers, as
shown in Fig. 3.
Combining the service-level (inter-admin-do-
main) and resource-level (inter-tech-domain)
orchestration in these projects shows a trend
toward a hybrid structure with decentralized
service federation between admin-domains and
TABLE 1. Comparison of key enabling technologies for network-cloud/edge convergence developed in representative projects.
Project Service-oriented abstraction Unified resource management Federated service orchestration Convergence scope
UNIFY Infrastructure-as-a-Service Hierarchical orchestration Limited to single admin-domain Network-cloud
T-NOVA VNF-as-a-Service Hierarchical orchestration Centralized brokerage Network-cloud
CORD Everything-as-a-Service Hierarchical orchestration Limited to single admin-domain Network-edge
5G-Exchange Slice-as-a-Service Hierarchical orchestration Cascade inter-provider orchestration Network-edge-cloud
5G-Transformer Slice-as-a-Service Hierarchical orchestration Inter-provider federation Network-edge-cloud
FIGURE 2. Hierarchical structures employed in the UNIFY and CORD projects for
unified management of heterogeneous resources in a single admin-domain.
Controller Adaption
Resource
Orchestration
compute
(a) resource orchestration in UNIF Y
compute/storage IaaS
domain
controller
network
domain
controller
storage
domain
controller
OpenStack
Docker/Kubernete
Vms /containers,
containers-in-VMs
VNFaaS, V NaaS
SDN control apps
ONOS controller
virtual networks in SDN
Service Orchestrator
XOS
(b) service orchestration in CORD
2 The UNIFY and CORD
projects focus on integrating
heterogeneous infrastruc-
tures resources within an
admin-domain and thus lack
service orchestration capabil-
ity across admin-domains.
DUAN_LAYOUT.indd 152DUAN_LAYOUT.indd 152 11/19/20 3:59 PM11/19/20 3:59 PM
Authorized licensed use limited to: Penn State University. Downloaded on December 03,2020 at 23:15:34 UTC from IEEE Xplore. Restrictions apply.
IEEE Network • Novembe/December 20 20 153
hierarchical resource orchestration across tech-do-
mains within individual admin-domains.
The reviewed projects also reflect an extension
of the network-cloud/edge convergence scope
from network core to network edge. UNIFY and
T-NOVA focus on unifying carrier networks and
cloud data centers. CORD naturally integrates
networking and computing at the network edge.
Both 5GEx and 5G-Transformer extends the con-
vergence to the whole consortium of access-trans-
port-core networks with edge servers (embedded
in access/transport networks) and cloud data cen-
ters (attached to core networks).
chAllenges And reseArch opportunItIes
chAllenges
Heterogeneity in Resources and Func-
tions: Heterogeneity in infrastructure resources
and service functions is a main challenge to net-
work-cloud/edge convergence. The resources in
different tech-domains of network, compute, and
storage impose heterogeneous implementations
but need to be integrated into a common infra-
structure layer with holistic management. Virtu-
al functions for networking and computing may
have quite different features and requirements. For
example, VNFs often require shorter latency, larg-
er throughput, and higher reliability as compared
to typical cloud/edge functions. A wide variety
of service chains/network slices are constructed
to meet the highly diverse requirements of multi-
tenant users. End-to-end service provisioning in a
converged network-cloud/edge environment typ-
ically spans across autonomous admin-domains
operated by different service providers. Therefore,
how to fully utilize heterogenous virtual resources
and functions across tech- and admin-domains for
composite service provisioning to meet diverse
user requirements becomes a challenging problem.
Scalability in System and Function Design: Net-
work-cloud/edge convergence pushes the scalabili-
ty requirement of the system design to a new level.
Virtualization across the network-compute infra-
structures enables various functions to be deployed
as software instances and thus significantly increas-
es the number of function modules involved in the
system. The micro-services architecture supported
by container-based virtualization further decom-
poses virtual functions to finer-grained service
components interconnected through network con-
nections. This presents new challenges to system
scalability due to the increased number of function
components and the communication overheads
among them. In addition, network-cloud/edge
convergence leads to a common service platform
for meeting the diverse multi-tenant requirements,
which incurs a large number of network slices to
be constructed and managed through the orches-
tration of numerous network-compute service
components. Therefore, engineering a scalable
design of system architecture and control/manage-
ment is a challenging open issue.
Flexibility, Agility, and Performance of Ser-
vice Provisioning: Convergence of networking
and cloud/edge computing expects flexible and
agile provisioning of composite services with
performance guarantees. In addition to the het-
erogeneous resources/functions in a large scale
converged network-cloud/edge system, integration
of decentralized computing capabilities in networks
with finer-grained function components enabled by
the micro-services architecture introduces even
more dynamism in various aspects, including avail-
ability, capacity, mobility, and lifespan of both host-
ing infrastructures and virtual function instances.
All of these factors together call for more sophisti-
cated service management, among which inter-do-
main federation for end-to-end provisioning of
composite services is particularly challenging. Infor-
mation exchange between autonomous domains,
collaboration between service providers, holistic
management of resources and services are all
challenging problems that demand thorough inves-
tigation. Service performance guarantee, which
has been an important issue in network virtualiza-
tion, becomes even more challenging due to net-
work-cloud/edge convergence. Optimal mapping
from service requirements to resource allocation,
flexible inter-domain resource management for ser-
vice delivery, effective evaluation and verification
of end-to-end service performance are all open
issues to be fully studied.
Integration of 5G Network and Edge Com-
puting: The 5G network provides an environ-
ment in which edge computing may be widely
deployed; therefore, integrating 5G network and
edge computing forms a representative scenario
of network-cloud/edge convergence. The unprec-
edented complexity of the 5G network introduces
challenges to its integration with edge computing.
Such complexity comes from various aspects, for
example, the dense and heterogeneous network
functions, highly diverse applications, ultra-low
latency requirements for vehicle communications,
growing demand for location-based services with
FIGURE 3. Decentralized structures for service orchestration/federation across
admin-domains in the 5GEx and 5G-Transformer projects.
Multi-domain
Orchestrator
service provider A
(a) inter-domain service orchestration in 5G-Exchange
service provider B
Infrastructure
tech-domain
controller
(b) service / resource federation across admin-domains in 5G-Transformer
Multi-domain
Orchestrator
service provider B
Multi-domain
Orchestrator
service provider C
tech-domain
controller
tech-domain
controller
service provider A Infrastructure
inter-domain
service
orchestration
tech-domain
controller
tech-domain
controller tech-domain
controller
tech-domain
controller
resource
federation
admin-domain 1
owned by provider A
VNFs V CFs VNF s VCFs
network slice 1
VNFs V CFs VNF s VCFs
network slice 2
service
federation
service
federation
admin-domain 2
owned by provider B
DUAN_LAYOUT.indd 153DUAN_LAYOUT.indd 153 11/19/20 3:59 PM11/19/20 3:59 PM
Authorized licensed use limited to: Penn State University. Downloaded on December 03,2020 at 23:15:34 UTC from IEEE Xplore. Restrictions apply.
IEEE Network • Novembe/December 20 20
154
high positioning accuracy, and so on. Therefore,
deploying edge computing within 5G network
becomes an important research area where com-
prehensive solutions are to be developed for
addressing the challenges of heterogeneity, scal-
ability, flexibility, and performance.
reseArch opportunItIes
Architecture Design of Converged Net-
work-Cloud/Edge Systems: Architectural design
forms a technical foundation for enabling net-
work-cloud/edge convergence. Combining the
software-defined principle with service-oriented
virtualization in architecture design may offer
promising approaches to addressing the afore-
mentioned challenges. Software-defined network-
ing essentially decouples the data and control
planes to enable a network operating system with
programmability. Integrating the software-defined
principle into the virtualization-based architecture
allows a global programmable control platform
respectively for the infrastructure layer and service
layer, which may significantly enhance composite
service provisioning. Therefore, this direction pro-
vides opportunities for future research. Inter-do-
main service federation is an important aspect of
architectural design that presents multiple options,
including hierarchical structure with a single mas-
ter orchestrator, peer-to-peer structure with a
cluster of domain orchestrators, and hybrid struc-
ture with multiple federated hierarchical systems.
Thorough evaluation of these options is also an
important subject that deserves more future work.
Resource and Service Modelling for Unified
Abstraction: Developing and standardizing a com-
mon set of models for unified abstraction of the
resources/services across network and cloud/
edge domains plays a key role in addressing the
heterogeneity challenge to network-cloud/edge
convergence. Following the virtualization princi-
ple of decoupling services from infrastructures,
abstraction models should be standardized on
both the resource and service layers for support-
ing multi-level virtualization (e.g., virtual infrastruc-
tures, virtual functions, and virtual networks). At
the same time, mapping from high-level models
for service specification to low-level models for
resource configuration is also critical. The stan-
dard models should include both functional and
non-functional features (e.g., performance and
capacity requirements) to facility high-perfor-
mance service management. On the other hand,
an appropriate level of information aggregation is
important in order to face the scalability challenge.
Research efforts toward resource/service model-
ing are fairly recent and more study is needed in
this area, thus offering research opportunities.
New Technologies for Composite Service
Management: Service management in a con-
verged networking-cloud/edge computing envi-
ronment calls for novel mechanisms for design,
instantiation, deployment, execution, and opti-
mization of composite services upon integrated
network-compute infrastructures. A key problem
is to achieve holistic service management across
heterogeneous tech- and admin-domains. Uni-
fied resource management across tech-domains
is being studied in various fields such as NFV,
5G network slicing, cloud data centers, and edge
computing, but more work is expected for seam-
less integration of the available technologies in a
large scale converged network-cloud/edge sys-
tem. Less progress has been made so far in feder-
ated service management across admin-domains/
service providers, thus offering more opportuni-
ties for future research. Automation and intelli-
gence are expected to be key attributes of service
management in order to face the challenges of
heterogeneity, scalability, and flexibility, which
demand novel technologies to be developed
probably by leveraging methods in areas such as
game theory, control theory, and machine learn-
ing. Big data analytics and machine learning tech-
niques could be particularly useful to address the
challenges of heterogeneity, scalability, flexibility,
and performance introduced by integrating edge
computing within 5G networks [15]. Performance
assurance is another important aspect of service
management. Analytical evaluation and experi-
mental verification of end-to-end performance of
composite network-cloud/edge services is also an
important problem for future research.
conclusIons
The virtualization and service-oriented principles
applied in networking technologies enable cloud-
based networking while the latest developments
in cloud/edge computing lead to a network-based
computing paradigm. Convergence of network-
ing and cloud/edge computing calls for a holistic
vision across these two fields and thus may signifi-
cantly impact relevant technology developments.
In this article, we first described the notion of net-
work-cloud/edge convergence with an architectur-
al framework of converged network-cloud/edge
systems. Then, we gave a brief survey on represen-
tative works to reflect the start-of-the-art research
toward network-cloud/edge convergence, includ-
ing recent progress in relevant standardization
and technology developments in representative
research projects. We also discussed challenges
to realizing network-cloud convergence and iden-
tified some opportunities for future research. We
found that although exciting progress has been
made toward convergence of networking and
cloud/edge computing, this interesting interdisci-
plinary area is still in its infancy, thus offering rich
research opportunities. We believe that cross-fertil-
ization among multiple fields, for example network
virtualization, cloud-native networking, network
slicing, cloud/edge computing, and micro-services
architecture, with a holistic vision of network-com-
pute convergence may trigger technology inno-
vations that will significantly enhance the future
information infrastructure.
Acknowledgment
This work was partially supported by the National
Key Research and Development Program of China
(2018YFE0205503); the National Natural Science
Foundation of China (61922017); and the Funds
for Creative Research Groups of China (61921003).
We believe that cross-fertilization among multiple fields, for example network vir tualization,
cloud-native networking, network slicing, cloud/edge computing, and micro-services architecture,
with a holistic vision of network-compute convergence may trigger technology innovations
that will significantly enhance the future information infrastructure.
DUAN_LAYOUT.indd 154DUAN_LAYOUT.indd 154 11/19/20 3:59 PM11/19/20 3:59 PM
Authorized licensed use limited to: Penn State University. Downloaded on December 03,2020 at 23:15:34 UTC from IEEE Xplore. Restrictions apply.
IEEE Network • Novembe/December 20 20 155
references
[1] B. Yi et al., “A Comprehensive Survey of Network Function
Virtualization,” Computer Networks, vol. 133, 2018, pp.
212–262.
[2] 3GPP, “TS 23.501: System Architecture for the 5G System,
version 15.2.0,” June 2018.
[3] ETSI, “Mobile-Edge Computing (MEC) Framework and Refer-
ence Architecture, version 2.1.1,” Jan. 2019.
[4] Q. Duan and S. Wang, “Network Cloudification Enabling
Network-cloud/Fog Service Unification: State of the Art and
Challenges,” Proc. 2019 IEEE World Congress on Services
(SERVICES), IEEE, 2019, pp. 153–59.
[5] ETSI, “Network function virtualization (NFV) architectural
framework version 1.2.1,” Dec. 2014.
[6] ETSI, “NFV Realease 3 Definition version 0.14.0,” Sept. 2019.
[7] IETF, “RFC 7665: Service Function Chaining (SFC) Architec-
ture,” Oct. 2015.
[8] MEF, “Lifecycle Service Orchestration (LSO) Reference
Architecture and Framework,” Mar. 2016.
[9] 5G-PPP, “View on 5G Architecture version 3.0,” June 2019.
[10] B. Sonkoly et al., “Unifying Cloud and Carrier Network
Resources: An Architectural View,” Proc. 2015 IEEE Global
Commun. Conf. (GLOBECOM 2015), 2015, pp. 1–7.
[11] M.-A. Kourtis et al., “T-NOVA: An Open-Source MANO
Stack for NFV Infrastructures,” IEEE Trans. Network and Ser-
vice Management, vol. 14, no. 3, 2017, pp. 586–602.
[12] L. Peterson et al., “Central Office Re-Architected as a Data
Center,” IEEE Commun. Mag., vol. 54, no. 10, 2016, pp.
96–101.
[13] G. Biczok et al., “Manufactured by Software: SDN-Enabled
Multi-Operator Composite Services with the 5G Exchange,”
IEEE Commun. Mag., vol. 55, no. 4, 2017, pp. 80–86.
[14] A. De la Oliva et al., “5G-Transformer: Slicing and Orches-
trating Transport Networks for Industry Verticals,” IEEE Com-
mun. Mag., vol. 56, no. 8, 2018, pp. 78–84.
[15] Q. Liu, T. Han, and N. Ansari, “Learning-Assisted Secure
End-to-End Network Slicing for Cyber-Physical Systems,” IEEE
Network Mag., vol. 34, no. 2, 2020.
bIogrAphIes
Qiang Duan Qiang Duan (S’00–M’03–SM’17) is a professor
of information sciences and technology at the Pennsylvania
State University Abington College. His current research interests
include next generation Internet, software-defined networking,
network function virtualization, and cloudnative networking. He
has (co-)authored three books, six book chapters, and more than
100 journal articles and conference papers in these areas. He is
an editor for multiple research journals and has been regularly
serving as a reviewer for various IEEE transactions and magazines.
He received IEEE Communications Society Outstanding Reviewer
Award in 2015. He has also served on the TPCs for numerous
international research conferences including GLOBECOM, ICC,
ICCCN, WCNC, AINA, ICNC, etc. He is a senior member of IEEE.
Shangguang Wang Shangguang Wang received his Ph.D.
degree from Beijing University of Posts and Telecommunica-
tions (BUPT) in 2011. He is currently a professor and deputy
director at the State Key Laboratory of Networking and Switch-
ing Technology, BUPT. He has published more than 100 papers,
and participated in organizing many international conferences
as a general chair or PC chair. His research interests include
edge computing, service computing, and cloud computing. He
is a senior member of IEEE, and the Editor-in-Chief of the Inter-
national Journal of Web Science.
Nirwan Ansari Nirwan Ansari (S’78–M’83–SM’94–F’09) is Dis-
tinguished Professor of electrical and computer engineering at
NJIT. He is also a Fellow of the National Academy of Inventors.
He has (co-)authored three books and more than 600 technical
publications. He has also been granted more than 40 U.S. pat-
ents. He has guest-edited a number of special issues covering
various emerging topics in communications and networking. He
has served on the editorial/advisory board of over 10 journals
including as Associate Editor-in-Chief of IEEE Wireless Commu-
nications Magazine. His current research focuses on green com-
munications and networking, cloud computing, drone-assisted
networking, and various aspects of broadband networks.
DUAN_LAYOUT.indd 155DUAN_LAYOUT.indd 155 11/19/20 3:59 PM11/19/20 3:59 PM
Authorized licensed use limited to: Penn State University. Downloaded on December 03,2020 at 23:15:34 UTC from IEEE Xplore. Restrictions apply.
