See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/285648601

Cloudifying the 3GPP IP Multimedia Subsystem for 4G and Beyond: A Survey

Article  in  IEEE Communications Magazine · December 2015

DOI: 10.1109/MCOM.2016.7378432 · Source: arXiv

CITATIONS READS
19 390

4 authors:

Mohammad Abu-Lebdeh Jagruti Sahoo

Concordia University Montreal South Carolina State University
12 PUBLICATIONS   79 CITATIONS    32 PUBLICATIONS   627 CITATIONS   

SEE PROFILE SEE PROFILE

Roch Glitho Constant Wette

Concordia University Montreal Ericsson
267 PUBLICATIONS   2,470 CITATIONS    15 PUBLICATIONS   52 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Optimistic Concurrency Conctrol View project

A Cloud-based Architecture for Multimedia Conferencing Service Provisioning View project

All content following this page was uploaded by Constant Wette on 08 April 2016.

The user has requested enhancement of the downloaded file.

 This paper has been accepted in IEEE Communications Magazine and will appear in January 2016 issue of the Network and
Service Management Series. This is an author copy.

Cloudifying the 3GPP IP Multimedia Subsystem for 4G

and Beyond: A Survey 1
Mohammad Abu-Lebdeh1, Jagruti Sahoo1, Roch Glitho1, Constant Wette Tchouati2
1
CIISE, Concordia University, Montreal, QC, Canada.
2
Ericsson, Montreal, QC, Canada.

Abstract—4G systems have been continuously evolving to cope these scenarios, such as much higher bandwidth, much lower
with the emerging challenges of human-centric and machine-to- latency, and much more stringent reliability and scalability than
machine (M2M) applications. Research has also now started on 5G what is offered today by the evolved 4G systems. For instance,
systems. Scenarios have been proposed and initial requirements 5G systems are expected to attain 10 to 100 times higher user
derived. 4G and beyond systems are expected to easily deliver a
data rate, and 5 times lower end-to-end latency [1]. Another
wide range of human-centric and M2M applications and services
in a scalable, elastic, and cost efficient manner. The 3GPP IP example is the requirement of cost efficiency, which was not a
multimedia subsystem (IMS) was standardized as the service primary concern in 4G. This is certainly due to the recent
delivery platform for 3G networks. Unfortunately, it does not meet emergence of new technologies such as cloud computing that
several requirements for provisioning applications and services in can easily enable cost efficiency.
4G and beyond systems. However, cloudifying it will certainly The 3GPP IP multimedia subsystem (IMS) [2] is a strong
pave the way for its use as a service delivery platform for 4G and candidate for application and service provisioning in 4G and
beyond. This article presents a critical overview of the beyond because it will enable a smooth migration. It was
architectures proposed so far for cloudifying the IMS. There are specified as the application and service delivery platform for
two classes of approaches; the first focuses on the whole IMS
3G networks and was then used at the inception of 4G as the de
system, and the second deals with specific IMS entities. Research
directions are also discussed. IMS granularity and a PaaS for the facto service platform. However, it does not meet all of the
development and management of IMS functional entities are the requirements of 4G and beyond.
two key directions we currently foresee. Cloud computing has emerged as a paradigm for delivering
computing resources (e.g., servers and storage) as a utility. It
Keywords—4G, 4G and beyond, 5G, cloud computing, elasticity, promises many benefits including elasticity, efficiency in
infrastructure as a service (IaaS), IP multimedia subsystem (IMS), resource usage, easy application and service provisioning, and
network function virtualization (NFV), platform as a service (PaaS), cost reduction. It has established the foundations for the
scalability, software as a service (SaaS), virtualization. emergence of network function virtualization (NFV), which
I. INTRODUCTION aims to transform network architectures through the
implementation of network functions (e.g., IMS) in software
Mobile systems have been undergoing a rather fast that can run on industry standard hardware. Cloud and NFV
evolution in the recent times. 4G systems have provided technologies can certainly aid in tackling the IMS shortcomings
increasingly higher bandwidth, lower latency, and more when it comes to the requirements of 4G and beyond mobile
features to meet the more stringent requirements of human- and wireless systems.
centric and machine-to-machine (M2M) communications since There are several approaches for integrating IMS and cloud
their inception during the second half of the last decade. This technologies. Gouveia et al. [3] illustrate these approaches by
constant innovation has paved the way for the growth of future presenting scenarios in a 4G network setting. In the first group
human-centric and M2M applications and is now leading us to of scenarios, IMS is re-engineered using cloud technologies. In
the 5G era. the second group, IMS is used to access applications and
METIS is a European project that aims to lay the services implemented in clouds. In this article, “cloudifying
foundation of the 5G concept to fulfill the requirements of the IMS” means re-engineering IMS using cloud technologies.
beyond-2020 connected information society and to support new This corresponds to the first group of scenarios. Readers
usage scenarios. It identifies five service and application interested in the use of IMS to access applications and services
scenarios that 5G will have to support, namely: amazingly fast, implemented in the cloud can consult [4].
great service in a crowd, best experience follows you, super This article is a survey on IMS cloudification for 4G and
real time reliable communications, and ubiquitous things beyond. It provides a critical review of the architectures for
communicating [1]. Several requirements are derived from cloudyfying IMS that have been proposed in the literature and

1
This article is an extended version of a paper presented at NTMS 2014 under the title “Cloudifying the 3GPP IP Multimedia
Sub-system: Why and How?”
 This paper has been accepted in IEEE Communications Magazine and will appear in January 2016 issue of the Network and
Service Management Series. This is an author copy.

Figure 1. Simplified IMS architecture.

further discusses research directions. NFV based architectures reference point to obtain users’ profiles. As a SIP proxy, S-
are included in our review. The architectures proposed for IMS CSCF forwards specific types of SIP messages to the
cloudification thus far focus on either the entire IMS system or appropriate application server.
on specific entities. We start by introducing IMS, cloud HSS is another key component of the architecture. It is the
computing and NFV, also outlining the requirements of IMS central database of the mobile network that contains user-
cloudification for 4G and beyond. The third section reviews the related information, such as subscription, location, and
architectures that focus on the entire IMS system. In the fourth identification information. It supports the network entities’
section, we discuss the architecture that focuses on specific IMS functions (e.g., mobility) and service provisioning. Several IMS
entities. The fifth section focuses on research directions, and we functional entities at both IMS service and control layers
conclude in the final section. interact with it using the diameter protocol.
The SIP application server (SIP AS) is a SIP-based server
II. BACKGROUND INFORMATION ON IMS, CLOUD that implements the logic of IMS services. The SIP AS interacts
COMPUTING / NFV AND REQUIREMENTS ON IMS with HSS to obtain users’ profiles via the Sh reference point.
CLOUDIFICATION An example of an IMS service is the presence service, which
A. IMS accepts, stores and distributes presence information via SIP
messages.
IMS is an overlay control layer on top of an IP transport
The 3GPP IMS specification provides scalability through
layer required for the seamless and robust provisioning of IP
the distribution of components such as the CSCF and the HSS.
multimedia services to end-users. It is made up of a service
However, despite this provision, scalability remains a key issue
layer and a control layer. The service layer includes application
in IMS, as articulated in [5]. This is due to the fact that SIP is a
servers, such as a presence server. The key functional entity of
text-based protocol. Signaling delay may not be sustainable
the control layer is the call state control function (CSCF). It uses
when several CSCFs and application servers are deployed. In
the session initiation protocol (SIP) to control multimedia
addition to the scalability issue, there is actually no provision in
functions.
IMS for meeting the cost efficiency requirement of 4G and
Fig. 1 depicts a simplified architecture for IMS network.
beyond mobile and wireless communications.
There are three types of CSCF: proxy-CSCF (P-CSCF),
interrogating-CSCF (I-CSCF) and serving-CSCF (S-CSCF). P- B. Cloud Computing And NFV
CSCF is the first point of contact for the IMS user equipment Cloud computing has emerged as a viable delivery model
(UE) within an IMS network. It acts as a stateful SIP proxy for IT resources. It leverages visualization technology to enable
when routing SIP signaling messages going to and from an IMS on-demand network access to a shared pool of configurable
UE. It is allocated to the IMS UE and does not change for the resources (e.g., networks, servers, storage, applications, and
duration of the registration. I-CSCF is the first contact point for services) with self-service provisioning and administration. It
external IMS networks. It is a stateless SIP proxy that selects has three main service models: infrastructure as a service
an S-CSCF for IMS UE and routes incoming SIP signaling (IaaS), platform as a service (PaaS), and software as a service
messages to the selected S-CSCF. Serving-CSCF (S-CSCF) is (SaaS).
the central node of the signaling plane of an IMS network. It IaaS offers end-users computing resources such as
acts as a stateful SIP registrar and proxy in an IMS network. As processing, storage, and network as a service over a network.
a SIP registrar, it registers IMS users and maintains the binding End-users can dynamically provision and de-provision
between the public user identity and the user profile. It also resources according to their need. Service providers use PaaS
interacts with the home subscriber server (HSS) via the Cx
 This paper has been accepted in IEEE Communications Magazine and will appear in January 2016 issue of the Network and
Service Management Series. This is an author copy.

to provision applications and services that are offered as SaaS

on a pay-per-use basis to end-users or other applications. PaaS
eases the provisioning process by adding levels of abstraction
to the infrastructure offered as IaaS. PaaS solutions vary widely
in the capabilities they offer. However, they all have the basic
capability to deploy applications on IaaS.
The NFV technology offers a new way to design, deploy
and manage network services. It decouples network functions
that are implemented in software from the underlying
proprietary hardware and runs the software as applications (i.e.,
virtual network functions [VNFs]) on commercial off-the-shelf
(COTS) hardware [6]. The shift towards software-based
network functions leads to flexibility as the VNFs can be easily
deployed in various locations, updated, and scaled without the
need to change the hardware.
NFV was developed to benefit the networks from
virtualization technology to consolidate and run VNFs on
COTS hardware such as servers and switches. It promises many
benefits to the Telco industry such as flexibility, openness,
network services agility, and reduced capital expenditures
(CAPEX) and operational expenditures (OPEX) [6].
Although related, cloud computing and NFV are different
concepts. Cloud computing refers to the concept of delivering
the computing resource as a service whereas NFV focuses on
migrating the network functions to run on COTS hardware.
However, by leveraging cloud computing, NFV can take Figure 2. NFV architectural framework [8].
advantage of the benefits of cloud computing and bring it to the
Telco industry. The benefits include elasticity, resource
efficiency, and even more reduced CAPEX and OPEX than following requirements to be the most pertinent for cloudifying
NFV on its own. the IMS for 4G and beyond:
The NFV architectural framework [8], as being 1) Elastic scalability: IMS today relies solely on using pre-
standardized by the European Telecommunications Standards allocated and over-provisioned functional entities to meet the
Institute (ETSI), is depicted in Fig. 2. It comprises NFV expected demand peak. New capacity requires significant
infrastructure (NFVI), VNFs, and NFV management and efforts to manually add new equipment to the system. On the
orchestration layers. NFVI provides the environment in which other hand, a cloudified IMS should take advantage of the
VNFs can execute. It provides the compute capabilities elasticity of the cloud to adapt dynamically to the growing or
comparable to an IaaS, although usually with much more shrinking of the load requirements by adjusting the allocated
stringent performance requirements. It also supports the resources in a fine-grained manner. Additionally, it should be
dynamic network connectivity between VNFs, which can be able to handle smoothly a massive number of IMS UEs. Indeed,
achieved by leveraging emerging technologies such as 10-100 times more devices are expected to be connected to 5G
software-defined networking (SDN). The virtualized compared to today.
infrastructure manager performs resource management and 2) Latency: 4G and beyond will support a wide variety of
allocation. The VNF manager handles VNF life cycle human-centric and M2M applications that will tolerate different
management (e.g., instantiation, scaling, and termination). The values of latency. Some of these applications can tolerate
VNF orchestrator is mainly responsible for the life cycle latencies on the order of a few seconds while others have stricter
management of the network services, which usually includes latency requirements than what exists today. For instance,
several VNF instances.
teleprotection is a mission-critical application for power
C. Requirements utilities. It includes real-time monitoring and alerting
The IMS was designed for 3G with human-centric functionalities that require transferring the messages with about
applications in mind; however, 4G and beyond aim at catering 8 milliseconds delay on the application layer [1]. The cloudified
for both human-centric applications and M2M applications IMS should be able to support the applications that require
(e.g., smart grid). This calls for a redesign of the IMS, and cloud different levels of latency. This includes the applications that
computing is the ideal basis since it enables features such as have very strict latency requirements compared to today. It also
scalability and efficiency in resource usage. We consider the should be able to maintain the required latency under a high
load.
 This paper has been accepted in IEEE Communications Magazine and will appear in January 2016 issue of the Network and
Service Management Series. This is an author copy.

3) Resource efficiency: Today, IMS is installed with over-

provisioning of resources to accommodate the peak demand.
However, the shift towards on-demand capacity makes resource
efficiency more critical, since inefficiency would be translated
directly into higher running cost (i.e., OPEX) with the pay-per-
use pricing model.
4) Follow-me: The basic idea behind the “follow-me”
concept is that cloud services follow the end-users during their
movement [7]. Mobile operators will use multiple IaaSs that are
geographically distributed and interconnected [7]. IMS and
IMS services could be deployed in different locations to offer
better user experience. Therefore, as soon as the end-user
moves, the optimal application server for providing the IMS
service may change. In the future, the service should follow the
end-user and should always be accessed from the application
server and through the IMS functional entities which ensure the
best user experience. Nowadays, P-CSCF and S-CSCF entities
are allocated to the IMS UE and do not change for the duration
of the registration. Through this period, end-users access their
IMS services through these assigned entities. Therefore, to have
service mobility in this model, the IMS UE should de-register
from the assigned IMS entities and then register again which Figure 3. Simplified virtualized IMS.
will cause service interruption.
Requirements are unfortunately often in conflict, and our allocation algorithm can dynamically allocate and de-allocate
proposed requirements are no exception to that tendency. virtual central processing unit (vCPU) and memory resources
Appropriate trade-offs will need to be made when new to VMs according to the current workload. The algorithm aims
architectures are designed. Let us illustrate this by to allow the platform to satisfy the carrier-grade response time
demonstrating the conflicts between elastic scalability, latency, requirement, achieve high resource utilization and reduce cost.
and resource efficiency. It is clear that today’s granularity level Additionally, the algorithm assumes that each VM boots
(i.e., 3GPP functional entities) is an impediment to elastic with an initially allocated vCPU and memory. Each VM also
scalability. However, refining that level of granularity through has a predefined maximum amount of vCPU and memory that
the splitting of the functional entities will usually lead to an can be allocated. When the resource utilization exceeds a
additional cost (e.g. management complexity, inter sub- predefined threshold, the system adds one vCPU or more
functional entities communications). These costs may (or may memory if the VM has not reached the maximum allowed
not) offset the gains expected from the refining. In addition, the resources. If the physical machine (PM), which hosts the VM,
splitting may prevent latency requirements from being met. does not have enough resources to scale the resources of the
Optimal splitting, therefore, becomes the key. We further VM, the algorithm performs live migration of the VM to
elaborate on this in the research directions section. another PM with enough resources. The algorithm can also
elastically scale the number of the active PMs in the cloud
III. APPROACHES THAT DEAL WITH THE ENTIRE IMS
infrastructure automatically, according to the workload. It aims
This section reviews the approaches that focus on the whole to achieve high resource utilization and reduce power
IMS system in the light of the requirements set forth in section consumption costs.
II.C. In these approaches, a common pool of resources is The proposed resource allocation algorithm can elastically
dynamically allocated to IMS functional entities. Fig. 3 scale IMS vertically to adapt to the workload whereas
provides an illustration. The physical computation, storage, and horizontal scalability is not tackled. However, the stateful
networking resources are virtualized. This allows for an IMS architecture for many of the IMS functional entities (e.g., S-
with a set of interacting virtual functional entities (i.e., CSCF) hinders the implementation of horizontal scalability.
functional entities that rely on virtualized resources). Table 1 For instance, it would be difficult to terminate an S-CSCF
summarizes the review findings. instance when it handles an ongoing call because this would
A. Virtualized IMS require transferring the stored state to another S-CSCF instance.
The authors propose a resource allocation algorithm to achieve
In [9], Lu et al. propose a cloud platform for the IMS core high resource utilization. However, resource efficiency may not
network that runs IMS entities on cloud-based virtual machines be maximal since the optimal splitting is not considered and the
(VMs). The proposed platform supports dynamic resource default splitting (i.e., IMS functional entities as defined today)
allocation and disaster protection. The proposed resource is used. The authors also do not evaluate the latency achieved
 This paper has been accepted in IEEE Communications Magazine and will appear in January 2016 issue of the Network and
Service Management Series. This is an author copy.

by their architecture. Furthermore, they do not tackle follow- A. HSS

me requirement. However, it remains an issue in the proposed Few works propose virtualized and cloud-based HSS
design due to the static assignment of IMS functional entities architectures. Yang et al. [11] propose the distribution of HSS
for a specific IMS UE at the registration process. into a resource layer and a management layer. The resource
B. IMS as a Service layer is implemented in the cloud and simulations are
performed to demonstrate performance gains. Although their
Carella et al. [10] propose three architectures for cloud-
proposed solution enables an independent scaling of resource
based virtualized IMS using NFV: Virtualized-IMS, Split-IMS,
and management layers, elastic scalability is not tackled. It is
and Merge-IMS. In the Virtualized-IMS architecture, each IMS
also not possible to evaluate the resource efficiency as the
functional entity is implemented as software that runs on a
optimal splitting is not tackled. The performed experiment
single VM. The interfaces with external components are not
shows that the latency is high. Furthermore, follow-me is not
changed. The Split-IMS moves the state of the subscribers,
tackled in their work.
which is maintained in many IMS entities (e.g., P-CSCF and S-
In [12], Paivarinta et al. use home location register (HLR) to
CSCF), to an external functional entity called Shared-Memory.
evaluate whether cloud technologies can meet the carrier-grade
This makes the IMS entities stateless. A load balancer is
requirements. HLR is the primary subscriber database for
positioned as an entry point for the new stateless entities to
mobile networks up to the 3GPP release 4 standards, and today
distribute the load.
is considered a subset of HSS. The proposed architecture uses
Additionally, the Merge-IMS architecture groups the main
HBase NOSQL database as HLR storage and deploys it on
four entities of IMS (i.e., P-CSCF, S- CSCF, I-CSCF, and HSS)
Amazon EC2 IaaS. It utilizes the telecommunication
and deploys them into one VM called IMS-VM. It introduces
application transaction processing (TATP) benchmarking tool
the IMS-Locator entity, which assigns the subscribers to a
to measure the performance of the HBase database under load,
particular IMS-VM instance during the registration process. All
which is typical in telecommunications. Unfortunately, the
HSS entities in IMS-VM instances share the same database to
authors do not tackle elastic scalability. They also do only
store subscriber information.
discuss the storage of HLR and do not discuss the HLR
The Virtualized-IMS architecture can scale using the
application logic. Thus, it is not possible to evaluate the
procedures already standardized by 3GPP to some extent.
resource efficiency. In addition, the performed experiment
However, the scalability is limited due to the stateful
shows that the latency increases proportionally to the
architecture. The Split-IMS architecture separates functional
throughput. Follow-me requirement has not been tackled.
entities’ logic and state so the logic can scale easily by
instantiating new stateless entities and adding them to the load B. Presence service
balancer. However, the scalability is not in a fine-grained In [13], Belqasmi et al. propose an early architecture for a
manner. For instance, to scale HSS, a full-fledged HSS (e.g., virtualized presence service for the future Internet. Although
storage and all reference points) should be instantiated. It is also the scalability is not tackled, it is ensured through the use of
important to verify the optimality of the proposed splitting, and presence service substrates. However, the authors do not tackle
how it affects the performance and resource efficiency. the level of granularity of the substrates. It is therefore rather
Moreover, the Merge-IMS architecture scales by creating a new difficult to assess whether the architecture could scale in a fine-
IMS-VM that has all items (i.e., full-fledged IMS). Thus, the grained manner and whether resource efficiency could be
elastic scalability is limited since it is difficult to scale in due to ensured. The latency and follow-me requirements are not
the granularity level (i.e., IMS-VM) and the stateful tackled.
architecture. Quan et al. [14] also focus on presence service. They
The authors do not provide performance metrics to evaluate propose a cloud-based implementation of presence service. The
the architectures’ latency. They also do not tackle the optimal Eucalyptus cloud open source software is used, and the whole
splitting of the IMS functional entities. Therefore, it is hard to presence server is deployed on a VM. The authors do not tackle
assess whether resource efficiency could be met. Although elasticity scalability, optimal splitting, and follow-me.
follow-me requirement has not been tackled, none of the Moreover, the evaluation shows that the architecture’s latency
proposed architectures can satisfy it without re-architecting the is high and increases proportionally to the throughput.
IMS.
V. RESEARCH DIRECTIONS
IV. APPROACHES THAT DEAL WITH SPECIFIC IMS ENTITIES
Research on IMS cloudification has started. This section
This section reviews the approaches that focus on specific provides insightful directions for future studies. In this article,
IMS entities. The main IMS entities that have attracted the we focus on two research issues as illustrations. In the first
attention of researchers are the HSS and the presence service. section, we will discuss challenges related to the IMS
This is probably due to the fact that they are much less complex granularity level. This discussion includes both architectural
than other nodes such as the CSCF. Table 1 summarizes the and algorithmic issues. The second section focuses on PaaS for
review findings. IMS.
 This paper has been accepted in IEEE Communications Magazine and will appear in January 2016 issue of the Network and
Service Management Series. This is an author copy.

Requirements
Architectures
Elastic Scalability Latency Resource Efficiency Follow-me

Virtualized IMS [9] Partly NO Partly NO

IMS as a Service
Partly NO NO NO
(Virtualized-IMS)
IMS as a Service
[10] Partly NO NO NO
( Split-IMS )
IMS as a Service
Partly NO NO NO
( Merge-IMS )
[11] NO NO NO NO
HSS
[12] NO NO NO NO

[13] Partly NO NO NO
Presence
[14] NO NO NO NO

Table 1. Summary of the evaluated approaches vs. the identified requirements for cloudifying the IMS for 4G and beyond.

The decomposition also gives rise to architectural and

A. Reconsidering The Granularity Level of IMS
algorithmic challenges. At the architecture level, if the new sub-
Each IMS network functional entity as defined by 3GPP functional entities interact with each other, then there is a need
contains a set of functions as one deployable and scalable unit. to design new interfaces. The interfaces should be very
These entities are often stateful which hinders elastic scalability lightweight to minimize the extra cost induced by the
and resiliency in the cloud. We believe it is worthy, in the cloud communication. On the other hand, they also need to be reliable
environment, to investigate the possibility of having finer and scalable.
granularity for IMS network functional entities to achieve finer At the algorithmic level, there is a need to identify the
control, elastic scalability, and better resiliency. optimal granularity for the sub-functional entities that can
A good starting point may be to separate the functional achieve the intended benefits (if possible). A key challenge is
entities’ logic and data (or state). It should be noted that 3GPP to determine the fine-grained atomic operations of each coarse-
has also stipulated this separation [15] primarily for data grained IMS network functional entity (e.g., HSS and presence
consistency purposes, but has also mentioned better scalability server), and the degree of relationship between these operations
as a potential advantage. This brings about the challenge of and the associated cost (e.g., memory and processing).
leveraging cloud technologies (e.g., distributed cache) to ensure Resource inefficiency could be translated into the cost of
equivalent performance characteristics. unused resources for given operations. This could be, for
A next step will be to consider decomposing the IMS instance, translated into a graph theory problem with a weighted
network functional entities’ logic into smaller sub-functional undirected graph formed by representing atomic operations as
entities, leading to finer control over the distinct functions. vertices. In this model, two vertices would be joined by an edge
However, this decomposition may not be priceless. Indeed, it if they are related, and they need to communicate. It could then
increases the management complexity and may have a negative be solved by formulating it as an optimal clustering problem
impact on latency. The cloud can help to alleviate the where each cluster is represented as a set of vertices. The
management complexity by leveraging PaaS to automate IMS’s objective would be to maximize the sum of intra-cluster
life cycle. However, many challenges have to be addressed at communication cost, minimize the sum of resource costs of all
the PaaS level to make it a reality. We elaborate more on this in clusters, and minimize the sum of the inter-cluster
the next section. As for the latency, placement algorithms are a communication cost. Of course, there would be constraints
potential avenue to minimize effectively the latency and cross- such as latency. The optimal clustering problem can be solved
network traffic. For instance, the algorithms may place the for each coarse-grained functional entity independently. It can
related functions on VMs hosted on the same physical server, be shown that the optimal clustering problem is non-
so they communicate through a virtual switch which leads to deterministic polynomial-time hard (NP-hard) when the
lower latency compared to the communication over the number of atomic operations is large and hence requires
network. efficient heuristics to solve it. The design of these heuristics is
 This paper has been accepted in IEEE Communications Magazine and will appear in January 2016 issue of the Network and
Service Management Series. This is an author copy.

an important research direction. More importantly, clustering suffer from nontrivial latency in detecting problems and making
algorithms such as hierarchical clustering and K-means decisions. On the other hand, the distributed architecture has
clustering can be modified to solve the optimal granularity lower latency. However, it is more complex and gives rise to
problem. the challenge of maintaining end-to-end service visibility.
Another challenge is related to the capacity of these functions.
B. Towards a PaaS for IMS
The number of functional entities that need to be managed is
The Telco industry could leverage PaaS to deliver IMS changed over time as the services scale elastically. Thus, the
network functional entities (e.g., CSCFs, HSS, presence, etc.) capacity of these functions should elastically scale to adapt to
or a subset (e.g., only HSS) as SaaS services with pay-per-use system workload. This requires clear definitions of the key
pricing to end-users (i.e., IMS UE) or even to other SaaS performance indicators need to manage the capacity.
services. The PaaS would automate the life cycle of the Another open issue is network orchestration. To best of our
functional entities from deployment to management (e.g., knowledge, today IT PaaS solutions use the networks with best-
monitoring, auto-scaling and auto-healing) and orchestration. effort delivery. On the other hand, quality of service (QoS) is a
For Telco, the PaaS would need to run on multiple, requirement in Telco to guarantee the performance (e.g.,
geographically distributed IaaSs that are interconnected by latency) required by the applications (e.g., multimedia
wide area network (WAN). This would help ensure the service applications). Indeed, Telco PaaS should leverage the IaaS
continuity and reduce latency by deploying closer to end-users. network and WAN capabilities to interconnect the deployed
A key open issue in PaaS is the aspiration for standard functional entities (could be across multiple IaaSs) using
language to describe the SaaS services. This language should transport network that meets specific requirements (e.g.,
be able to describe the structure of these services (i.e., latency and bandwidth). This requires that both IaaS and WAN
functional entities, relationships, requirements, etc.), and their support advanced networking capabilities (e.g., QoS) and
management aspects (e.g., deployment, monitoring, scaling, exposes them via northbound interfaces.
etc.). It should support the deployment and management across
multiple IaaSs so that functional entities could be deployed at VI. CONCLUSIONS
different locations. PaaS could use the services’ description to In this article, we identified the most pertinent requirements
automate their life cycle. Topology and orchestration for cloudifying the IMS for 4G and beyond. We have also
specification for cloud applications (TOSCA) [16] may be a reviewed the architectures proposed thus far for the
good starting point. It is standard to describe cloud applications cloudification of IMS using the identified requirements. These
by means of topology templates and management plans. architectures are classified into two categories: the first focuses
However, the current TOSCA version (version 1.0) does not on the whole IMS system, and the second deals with specific
support all management aspects needed in Telco, such as IMS functional entity. Our evaluation has showed that the
monitoring. existing literature does not meet the requirements of
Another research challenge is the elastic scaling of the SaaS cloudifying the IMS for 4G and beyond. Subsequently, we
services offered by the PaaS. These services often consist of outlined some interesting research issues that still need to be
multiple interconnected functional entities that could be resolved. We have discussed the possibility of decomposing
distributed across multiple IaaSs. The traditional scaling IMS functional entities to achieve elastic scalability and better
approaches in PaaS usually scale the overloaded entity itself resiliency in cloud settings. We have also discussed the main
without considering the impact on other entities in the service. challenges resulting from this decomposition, such as the need
In Telco, these approaches would not be sufficient and efficient for new communication interfaces and optimal granularity.
since there could be a need in many cases to scale and optimize Furthermore, we have identified many challenges at the PaaS
other entities in the service. In fact, there is a need for new smart level. One challenge is the lack of a standard language that can
scaling approaches which consider the end-to-end service (i.e., describe the IMS structure and management aspects. Another
all entities in the service), and is aware of service requirements challenge is the design of elastic management and orchestration
(e.g., latency and resiliency) and surrounding environment functions in a distributed environment.
(e.g., resource availability and network traffic status). These
approaches should evaluate the impact of scaling and then ACKNOWLEDGEMENT
decide accordingly what to scale, where to scale (same IaaS or This work is supported in part by Ericsson and the National
across multiple IaaSs), and what to optimize aiming to meet the Science and Engineering Research Council (NSERC) of
service requirements. Canada.
The PaaS includes management and orchestration functions
(e.g., monitoring, fault management and scaling). These REFERENCES
functions are responsible for automating the life cycle of the
functional entities. However, there will be many challenges in [1] A. Osseiran, F. Boccardi, V. Braun, K. Kusume, P. Marsch, M.
Maternia, O. Queseth, M. Schellmann, H. Schotten, H. Taoka, H.
designing them in a distributed environment. One challenge is Tullberg, M. A. Uusitalo, B. Timus, and M. Fallgren, “Scenarios for 5G
related to the architecture and whether it is centralized or mobile and wireless communications: the vision of the METIS project,”
distributed. The centralized architecture is simpler, but will IEEE Commun. Mag., vol. 52, no. 5, pp. 26–35, May 2014.
 This paper has been accepted in IEEE Communications Magazine and will appear in January 2016 issue of the Network and
Service Management Series. This is an author copy.

[2] 3GPP TS 23.228, “IP Multimedia Subsystem (IMS); Stage 2.” version fellow in the Department of Electrical and Computer
12.8.0, 2015.
Engineering, University of Sherbrooke, Canada. Her research
[3] F. Gouveia, S. Wahle, N. Blum, and T. Magedanz, “Cloud computing
and EPC/IMS integration: new value-added services on demand,” in interests include cloud computing, network functions
Proceedings of the 5th international ICST mobile multimedia virtualization, content delivery networks and wireless sensor
communications conference, 2009, p. 51. networks.
[4] W. Zhang, W. Lei, X. Chen, and S. Liu, “Architecture and Key Issues
Roch Glitho holds a Ph.D. (Tekn. Dr.) in tele-informatics
of IMS-Based Cloud Computing,” in Cloud Computing (CLOUD),
2013 IEEE Sixth International Conference on, 2013, pp. 629–635. (Royal Institute of Technology, Stockholm, Sweden), and
[5] P. Agrawal, J.-H. Yeh, J.-C. Chen, and T. Zhang, “IP multimedia M.Sc. degrees in business economics (University of Grenoble,
subsystems in 3GPP and 3GPP2: overview and scalability issues,” France), pure mathematics (University of Geneva,
IEEE Commun. Mag., vol. 46, no. 1, pp. 138–145, Jan. 2008.
Switzerland), and computer science (University of Geneva). He
[6] B. Han, V. Gopalakrishnan, L. Ji, and S. Lee, “Network function
virtualization: Challenges and opportunities for innovations,” IEEE is an Associate Professor and Canada Research Chair at
Commun. Mag., vol. 53, no. 2, pp. 90–97, Feb. 2015. Concordia University in Montreal (Canada). He is also an
[7] T. Taleb and A. Ksentini, “Follow me cloud: interworking federated adjunct professor at several other universities including
clouds and distributed mobile networks,” IEEE Netw., vol. 27, no. 5,
Telecom Sud Paris (France) and the University of Western
pp. 12–19, Sep. 2013.
[8] ETSI GS NFV, “Network Functions Virtualization (NFV); Cape (South Africa). In the past, he has worked in industry and
Architectural Framework.” V1.2.1, 2014. has held several senior technical positions (e.g. senior
[9] F. Lu, H. Pan, X. Lei, X. Liao, and H. Jin, “A Virtualization-Based specialist, principal engineer, expert) at Ericsson in Sweden and
Cloud Infrastructure for IMS Core Network,” in Cloud Computing
Canada. His industrial experience includes research,
Technology and Science (CloudCom), 2013 IEEE 5th International
Conference on, 2013, vol. 1, pp. 25–32. international standards setting, product management, project
[10] G. Carella, M. Corici, P. Crosta, P. Comi, T. M. Bohnert, A. A. Corici, management, systems engineering and software/firmware
D. Vingarzan, and T. Magedanz, “Cloudified IP Multimedia Subsystem design. He has also served as IEEE distinguished lecturer,
(IMS) for Network Function Virtualization (NFV)-based
Editor-In-Chief of IEEE Communications Magazine, and
architectures,” in 2014 IEEE Symposium on Computers and
Communication (ISCC), 2014, vol. Workshops, pp. 1–6. Editor-In-Chief of IEEE Communications Surveys & Tutorials
[11] T. Yang, X. Wen, Y. Sun, Z. Zhao, and Y. Wang, “A new architecture Journal.
of HSS based on cloud computing,” in 2011 IEEE 13th International Constant Wette Tchouati is a Senior System Developer at
Conference on Communication Technology (ICCT), 2011, pp. 526–
Business Unit Cloud & IP, Ericsson Canada. He holds a M.Sc.
530.
[12] R. Paivarinta and Y. Raivio, “Applicability of NoSQL Databases to in Computer Engineering from Ecole Polytechnique de
Mobile Networks: Case Home Location Register,” in Cloud Computing Montreal (Canada), an M.Eng. In Electronics &
and Services Science, I. Ivanov, M. van Sinderen, and B. Shishkov, Eds. Communication from Ecole Polytechnique de Yaounde
Springer New York, 2012, pp. 225–242.
(Cameroon) and an MBA in Management from HEC Montreal
[13] F. Belqasmi, N. Kara, and R. Glitho, “A Novel Virtualized Presence
Service for Future Internet,” in 2011 IEEE International Conference on (Canada). He joined Ericsson in 2000 and has been involved in
Communications Workshops (ICC), 2011, pp. 1–6. research, product development, innovation projects and new
[14] W. Quan, J. Wu, X. Zhan, X. Huang, and Y. Ma, “Research of presence business development in different areas including Cloud
service testbed on cloud-computing environment,” in 2010 3rd IEEE
architectures, Data Management and Analytics, IoT/M2M, IMS
International Conference on Broadband Network and Multimedia
Technology (IC-BNMT), 2010, pp. 865–869. and Social Media.
[15] 3GPP TS 23.335, “LTE; User Data Convergence (UDC); Technical
realization and information flows; Stage 2.” version 11.0.0 Release 11,
2012.
[16] OASIS Committee Specification, “Topology and Orchestration
Specification for Cloud Applications,” Version 1.0, Nov. 2013.

BIOGRAPHIES
Mohammad Abu-Lebdeh received his B.Sc. degree in
Computer Engineering from An-Najah National University,
Palestine, and M.Sc. degree in Electrical & Computer
Engineering from Concordia University, Canada. He is
currently pursuing his Ph.D. degree in Information & Systems
Engineering at Concordia University. In the past, he worked for
several years as a software engineer. His current research
interests include cloud computing, service engineering, and
next generation networks.
Jagruti Sahoo holds Ph.D. degree in computer science and
information engineering from National Central University,
Taiwan. She is currently a postdoctoral fellow at Concordia
University, Canada. In the past, she has worked as postdoctoral

View publication stats