HAL Id: hal-01081713
https://hal.archives-ouvertes.fr/hal-01081713
Submitted on 10 Nov 2014
HAL is a multi-disciplinary open access
archive for the deposit and dissemination of sci-
entic research documents, whether they are pub-
lished or not. The documents may come from
teaching and research institutions in France or
abroad, or from public or private research centers.
L’archive ouverte pluridisciplinaire HAL, est
destinée au dépôt et à la diusion de documents
scientiques de niveau recherche, publiés ou non,
émanant des établissements d’enseignement et de
recherche français ou étrangers, des laboratoires
publics ou privés.
Data Ooading Techniques in Cellular Networks: A
Survey
Filippo Rebecchi, Marcelo Dias de Amorim, Vania Conan, Andrea Passarella,
Raaele Bruno, Marco Conti
To cite this version:
Filippo Rebecchi, Marcelo Dias de Amorim, Vania Conan, Andrea Passarella, Raaele Bruno, et al..
Data Ooading Techniques in Cellular Networks: A Survey. Communications Surveys and Tutorials,
IEEE Communications Society, Institute of Electrical and Electronics Engineers, 2015, 17 (2), pp.580-
603. �10.1109/COMST.2014.2369742�. �hal-01081713�
1
Data Offloading Techniques in Cellular Networks:
A Survey
Filippo Rebecchi, Marcelo Dias de Amorim, Vania Conan, Andrea Passarella,
Raffaele Bruno, and Marco Conti
Abstract—One of the most engaging challenges for mobile
operators today is how to manage the exponential data traffic
increase. Mobile data offloading stands out as a promising and
low cost solution to reduce the burden on the cellular network.
To make this possible, we need a new hybrid network paradigm
that leverages the existence of multiple alternative communication
channels. This entails significant modifications in the way data
is handled, affecting also the behavior of network protocols.
In this paper, we present a comprehensive survey of data
offloading techniques in cellular networks and extract the main
requirements needed to integrate data offloading capabilities into
today’s mobile networks. We classify existing strategies into two
main categories, according to their requirements in terms of
content delivery guarantees: delayed and non-delayed offloading.
We overview the technical aspects and discuss the state of the
art in each category. Finally, we describe in detail the novel
functionalities needed to implement mobile data offloading in the
access network, as well as current and future research challenges
in the field, with an eye toward the design of hybrid architectures.
Index Terms—Mobile data offloading, hybrid networks, WiFi,
delay-tolerant networks, cellular networks.
I. INTRODUCTION
G
LOBAL mobile traffic will boom in the years to come,
thanks to the increasing popularity of smart mobile
devices and the introduction of affordable data plans by
cellular operators. Data hungry mobile applications, such as
audio and video streaming, social sharing, or cloud-based
services, are more and more popular among users. Recently,
analysts from Cisco warned that global mobile data traffic
is expected to grow 18-fold between 2011 and 2018, three
times faster than the overall fixed IP traffic in the same
period [1]. It is also anticipated that 66.5% of this traffic
will be video related (with or without real-time requirements)
by 2017. As today’s most common data access method on
the move, cellular networks are under pressure trying to cope
with this unprecedented data overload. Accommodating this
F. Rebecchi is with LIP6 – UPMC Sorbonne Universits and Thales
Communications & Security, France (filippo.rebecchi@lip6.fr).
M. Dias de Amorim is with CNRS/UPMC Sorbonne Universits, 4 Pl.
Jussieu, 75005 Paris, France (marcelo.amorim@lip6.fr).
V. Conan is with Thales Communications & Security, 4 Av. des Louvresses,
92230 Gennevilliers, France (vania.conan@thalesgroup.com).
A. Passarella, R. Bruno, and M. Conti are with IIT–CNR, Via
Giuseppe Moruzzi, 1, 56124 Pisa, Italy ({a.passarella, raffaele.bruno,
marco.conti}@iit.cnr.it).
c
c
2014 IEEE. Personal use of this material is permitted. Permission
from IEEE must be obtained for all other uses, in any current or future
media, including reprinting/republishing this material for advertising or
promotional purposes, creating new collective works, for resale or
redistribution to servers or lists, or reuse of any copyrighted component
of this work in other works.
growth requires major investments both in the radio access
network (RAN) and the core infrastructures. From a purely
economic perspective, upgrading the RAN is very expensive,
since this approach requires more infrastructure equipment and
thus more investment.
Scarce licensed spectrum hinders the RAN enhancements.
Regulations allow mobile operators to use only a small portion
of the overall radio spectrum, which is also extremely expen-
sive. Users must share the same limited wireless resources.
Adding traffic beyond a certain limit mines the performance
and the quality of service (QoS) perceived by the users. During
peak times in crowded metropolitan environments, users al-
ready experience long latencies, low throughput, and network
outages due to congestion and overload at RAN level [2].
Unfortunately, this trend can only exacerbate in future due to
the predicted mobile data explosion. The problem concerns
primarily network operators because they have to trade-off
customer satisfaction with business profitability, given the
trend toward nearly flat rate business models. In other words,
the exponential increase in traffic flowing in their RAN does
not generate enough additional revenues to be allocated into
further RAN upgrades. This creates what M
¨
olleryd et al. call
the revenue gap [3].
The above-mentioned circumstances fostered the interest
in alternative methods to mitigate the pressure on the cel-
lular network. As a first option, mobile operators solved
this contingency by throttling connection speed and capping
data usage [4]. However, these practices negatively affect the
customer satisfaction. For this reason, alternative approaches
emerged. In this survey, we turn our attention to one of these
solutions, recently attracting increasing interest by the research
community: mobile data offloading. An intuitive approach is
to leverage the unused bandwidth across different wireless
technologies. We consider mobile data offloading as the use
of a complementary wireless technology to transfer data
originally targeted to flow through the cellular network, in
order to improve some key performance indicators.
Although offloading may apply to any network, current aca-
demic and industrial research mostly concerns with offloading
data from cellular networks. Those are the type of networks
that would benefit most from this technique. Note that, for
the sake of tractability, we limit the scope of this survey to
solutions in which mobile terminals are explicitly used as
part of the offloading scheme, either through using multiple
wireless interfaces, or through using non-conventional cellular
2
Cellular
BS
(a) Infrastructure only.
Cellular
BS
Wi-Fi AP
wire
(b) AP-based offloading.
Cellular
BS
(c) Terminal-to-terminal offloading.
Fig. 1. The two major approaches to cellular data offloading compared to the baseline traditional infrastructure-only system (a). Offloading through a wireless
Access Point (b). Offloading through terminal-to-terminal transmissions (c).
techniques (e.g., LTE-D2D).
1
Besides the obvious benefit of
relieving the infrastructure network load, shifting data to a
complementary wireless technology leads to a number of other
improvements, including: the increase of the overall through-
put, the reduction of content delivery time, the extension of
network coverage, the increase of network availability, and
better energy efficiency. These improvements hit both cellular
operators and users; therefore, offloading is often described
in the literature as a win-win strategy [5]. Unfortunately, this
does not come for free, and a number of challenges need
to be addressed, mainly related to infrastructure coordination,
mobility of users, service continuity, pricing, business models,
and lack of standards.
For the reader’s convenience, we depict in Fig. 1 the two
main approaches to offload in cellular networks when com-
pared with the traditional infrastructure-only mode (Fig. 1(a)).
Diverting traffic through fixed WiFi Access Points (AP), as
in Fig. 1(b), represents a conventional solution to reduce
traffic on cellular networks. End-users located inside a hot-
spot coverage area (typically much smaller than the one of a
cellular macrocell) might use it as a worthwhile alternative to
the cellular network when they need to exchange data. Hot-
spots generally provide better connection speed and through-
put than cellular networks [6]. However, coverage is limited
and mobility is in general constrained within the cell. Since
the monetary cost of deploying an array of fixed APs is
far lower than deploying a single cellular base station, the
major worldwide cellular providers such as AT&T, Verizon,
T-Mobile, Vodafone, and Orange have started integrating an
increasing number of wireless APs in their cellular networks to
encourage data offloading [7]. Meanwhile, a growing number
of applications that automatize the offloading process are
proposed for popular mobile devices (mainly iPhone and
1
However, for completeness, we will briefly review other strategies such as
femtocells, cognitive offloading, and multicast in Section VII.
Android based), such as iPass [8] or BabelTen [9].
2
The increasing popularity of smart mobile devices proposing
several alternative communication options makes it possible
to deploy a terminal-to-terminal (T2T) network that relies
on direct communication between mobile users, without any
need for an infrastructure backbone (Fig. 1(c)). This innovative
approach has intrinsic properties that can be employed to of-
fload traffic. T2T-offloading represents a vibrant research topic
that we discuss in detail along the survey. Benefiting from
shared interests among co-located users, a cellular provider
may decide to send popular content only to a small subset of
users via the cellular network, and let these users spread the
information through T2T communications and opportunistic
contacts. Note also that these two forms of offloading (AP and
T2T based) may be employed concurrently, enabling users to
retrieve data in a hybrid mode.
Although mobile data offloading can be – at a very high
level – categorized according to these two classes (i.e., using
fixed hot spots or T2T transmissions between mobile nodes), a
more refined classification of offloading techniques is required
to provide a comprehensive picture. Thus, the main contribu-
tions of this survey are three-fold:
• To categorize existing techniques based on their re-
quirements in terms of content delivery guarantee and
summarize previously published works.
• To describe a general architecture to enable mobile data
offloading with tight or loose delay guarantees.
• To discuss open research and implementation issues.
To the best of our knowledge, it exists up to now only
one work from Aijaz et al. that summarizes existing mobile
data offloading techniques, although from a higher level and
business-oriented point of view [11].
The rest of the survey is structured as follows. In Section II,
we propose a general classification of the available mobile data
2
The ability to switch seamlessly between heterogeneous networks is
referred to as vertical handover [10].
3
TABLE I: A classification of mobile data offloading strategies, along with their research directions and surveyed works.
Delay Requirements
Strategy Non-delayed Delayed
AP-based
AP Deployment and Modeling
[12], [13], [14], [15], [16], [17], [18], [19].
Prediction-Based Offloading
3GPP Standardization
[20], [21], [5], [22], [23], [24], [25], [26], [27].
[28], [29], [30], [31], [32], [33].
Feasibility and AP Deployment
Transport Protocols
[12], [34], [5], [35], [36], [37], [38], [39].
[40], [41], [42], [43].
T2T
Cooperative Distribution
Subset Selection
[44], [45], [46], [47],
[48], [49], [50], [51], [52], [53].
[54], [55], [56], [57], [58], [59], [60], [61].
Architecture
D2D Capabilities Integration
[62], [63], [64], [65], [66], [67], [68], [69].
[70], [71], [72], [73], [74], [75], [76], [77], [78], [79], [80].
offloading techniques. In Sections III and IV, we discuss the
state of the art for each category, offering also an in-depth
literature review. In Section V, we present a reference archi-
tecture and its possible implementation into a real framework.
In Section VI, we discuss the performance evaluation aspects
of different offloading strategies. In Section VII, we examine
other possible solutions to the mobile data explosion problem.
Finally, in Section VIII we discuss open research challenges,
concluding the paper in Section IX.
II. CLASSIFICATION
We may find in the literature various offloading strategies.
In this section, we review the main strategies and provide
a comprehensive categorization of existing solutions. It is
important to pinpoint that mobile data offloading techniques
can be classified depending on the assumptions one can
make on the level of synergy between cellular and unlicensed
wireless networks, as well as the involvement of user terminals
in the offloading process. Beyond the distinction between AP-
based and T2T approaches, another aspect plays a major role
in the categorization. In particular, we take into consideration
the requirements of the applications generating the traffic in
terms of delivery guarantees. For this reason, we also consider
a temporal dimension in the classification, depending on the
delay that the data we want to offload may tolerate upon
delivery. This translates into two additional categories: (i) non-
delayed offloading and (ii) delayed offloading.
We consider these two orthogonal dimensions (delivery
delay guarantees and offloading approach), which correspond
to four possible combinations, as shown in Table I. The biggest
difference between non-delayed and delayed offloading mech-
anisms lies in the way the timeliness of content reception is
handled. In fact, in non-delayed offloading we do not have any
extra delay on the “secondary” interface (considering cellular
the “primary”), while in delayed offloading the network adds
some delay (either associated to the fact that the user has to
wait until it gets close enough to a WiFi AP, or to get messages
through opportunistic contacts).
Non-delayed offloading. In non-delayed offloading, each
packet presents a hard delivery delay constraint defined by
the application, which in general is independent of the net-
work. No extra delay is added to data reception in order
to preserve QoS requirements (other than the delay due to
packet processing, physical transmission, and radio access).
For instance, interactive audio and video streams cannot
sustain any additional delay in order to preserve their real-
time requirements. One has to consider that tolerable latency
for voice connections is around 50 ms (up to one second for
live video streaming). This requirement puts a strain on the
network that should meet this deadline to ensure the proper
functioning of the application. It turns out that non-delayed
offloading is essentially unfeasible in opportunistic networks,
since the accumulated end-to-end delay over the transmission
path may be too high with respect to the strict delivery
requirements. However, if we restrict the analysis to low
mobility scenarios, it is still possible to deliver data with strict
delay guarantees using T2T transmissions or with the aid of a
fixed infrastructure. Non-delayed offloading in most cases may
be difficult to implement if one considers that users are mobile
and able to switch between various access technologies. If
operators want to allow users to be truly mobile and not
only nomadic inside the coverage area, they should focus
on issues such as transparent handover and interoperability
between the alternative access technologies and the existing
cellular infrastructure. For instance, this aspect is not granted
when one considers a basic offloading implementation through
IEEE 802.11 APs. On the other hand, this commitment allows
offloading data such as voice over IP (VoIP) or interactive
applications, obtaining a nearly transparent offloading process.
Delayed offloading. In delayed offloading, content reception
may be intentionally deferred up to a certain point in time,
4
in order to reach more favorable delivery conditions. We
include in this category the following types of traffic: (i)
traffic with loose QoS guarantees on a per-content basis
(meaning that individual packets can be delayed, but the entire
content must reach the user within a given deadline) and
(ii) truly delay-tolerant traffic (possibly without any delay
guarantees). The relaxation in the delivery constraint allows
also moving traffic opportunistically, which, by definition, can
only guarantee a probabilistic delivery time. If data transfer
does not end by the expected deadline, the cellular channel
is employed as a fall-back means to complete the transfer,
guaranteeing a minimal QoS. Despite the loss of the real-
time support due to the added transmission delay, note that
many mobile applications generate content intrinsically delay-
tolerant – just think about smartphone-based applications that
synchronize emails or podcasts in background. Enabling an
alternate distribution method for this content during peak-times
(when the cellular network is overloaded or even in outage)
becomes an interesting extension and represents a fundamental
challenge for offloading solutions.
Eventually, the categorization proposed in Table I may also
take into account additional parameters, such as the role of
mobility in the process. Delayed offloading strategies rely so
much on mobility that we can regard it as a real enabler.
Thanks to mobility, users may reach an IEEE 802.11 AP or
a neighbor that carries the content of interest, increasing the
offload capacity. On the other hand, in non-delayed offloading,
mobility often represents a major obstacle and requires a
substantial effort in order to make things work together.
III. NON-DELAYED OFFLOADING
Non-delayed offloading is the most straightforward and
experimented class of offloading. Data may be real-time and
interactive, thereby enabling the fruition of services such
as video streaming and VoIP. So far, WiFi hot-spots have
represented the most logical solution due to their widespread
diffusion, acceptable performance, and low cost. Operators can
incentivize subscribers to offload by offering unlimited data
through WiFi hot-spots (and leveraging instead on the capped
cellular data). Nevertheless, we can find in the literature
many approaches that exploit T2T content sharing between
neighboring nodes.
From a technical point of view, cellular base stations are de-
signed to cover large macro areas (1-2 km of diameter in urban
areas, and 4-12 km in rural areas), while IEEE 802.11 standard
covers limited areas, in the 30-100 meters range. In contrast,
the transmission rate is usually much faster for wireless local
area networks than cellular technologies. For instance, LTE
can reach a shared 28 Mbit/s peak in favorable conditions, with
a more realistic average of 10 Mbit/s [6]. On the other hand,
IEEE 802.11 standard with its latest amendment can reach
a realistic shared throughput of 40 Mbit/s [6].
3
Therefore,
as suggested in some works, it is possible to take advantage
of this complementarity, combining properly the strengths of
different technologies [12], [13].
3
The advertised throughput is around 100 Mbit/s for LTE and 300 Mbit/s
for IEEE 802.11n.
Proposals to employ T2T offloading make use instead of a
multitude of wireless technologies. An additional classification
could divide T2T approaches in two extra categories: (i)
solutions that rely on alternative unlicensed communication
technologies to establish direct communications (out-of-band)
and (ii) solutions that dedicate part of the licensed cellular
band to T2T communications (in-band). In the out-of-band
category, IEEE 802.11 and Bluetooth are common choices
since they are the most popular wireless technologies present
on smart mobile devices today. Other works propose alterna-
tive wireless technologies, such as IEEE 802.15, or other less
known high-speed short-range communication medium, such
as TransferJet [81], WiGig [82] or FlashLinQ [83]. For the
in-band offloading category, recent developments of the 3GPP
LTE-Advance standard (Rel-12) propose to integrate T2T com-
munication capabilities into future cellular architectures (better
known as device-to-device D2D) [84]. However, to date, T2T
technologies using unlicensed band (like WiFi and Bluetooth)
are the only realistic candidates for data offloading. This
because the standardization of in-band D2D communications
as an underlay to a cellular network is still in its early stages,
with a time to market expected in several years [60].
A. AP-based
The prevailing AP-based offloading model today is user-
driven, meaning that users must explicitly enable the alterna-
tive access network in order to benefit from an enhanced ex-
perience.
4
This approach is appealing at first, as it requires no
modifications in the network infrastructure; however, common
limitations such as constrained mobility and lack of session
continuity hinder its mass adoption. To pave the way for better
cross-resource utilization and improved customer experience,
the current trend is to let operators have a deeper control of the
offloading process. This eventually raises the question of how
a cellular operator can run a profitable business by shifting
off-network large parts of its traffic.
Providers are more and more looking toward a tighter
integration of alternative access networks and their cellular
infrastructure, as depicted in Fig. 2. The integration process
concerns partnerships between cellular and wireless providers,
common billing and accounting policies, shared subscriber
databases for authentication, authorization, accounting (AAA),
and security provisioning. Two possible network architectures
to date are envisioned to integrate cellular and WiFi access:
loose coupling and tight coupling. In loose coupling, the two
networks are independent and are interconnected indirectly
through an external IP network. Service continuity is provided
by roaming between the two networks. In tight coupling
instead, the two networks share a common core and many
functions, such as vertical and horizontal handover, integrated
management of resources, and common AAA.
1) AP Deployment and Modeling: Several trace-based
analyses demonstrate that the deployment of fixed APs is a
viable method to reduce congestion in cellular networks. These
4
Smart mobile devices already give priority by default to WiFi when a
wireless network results available and WiFi interface is enabled.
![Fig. 12. Coverage metrics in the TOMP framework [53]: (a) the Static Coverage does not take into account any future movement, so nodes are considered in contact or not based on their present position; (b) the Free Space Coverage considers the possible movement of nodes in free space: the future meeting probability is the area of intersection of the two circles that represent the possible movements of the two nodes; (c) the Graph-based Coverage takes into account the underlying structure of road graph to limit the prediction to the road graph.](/figures/fig-12-coverage-metrics-in-the-tomp-framework-53-a-the-12kqbbm7.webp)
![Fig. 13. Network extension in [62]: destination client experiences bad cellular connectivity . After the discovery of a neighbor node with better channel conditions, data is routed through this “proxy client” in the cellular network.](/figures/fig-13-network-extension-in-62-destination-client-12jo5om9.webp)
![Fig. 10. Offloading ratio of delayed AP transfers with various deadlines, 100% of traffic delayed, Seoul dataset [12]. Increased delay-tolerance values result in an increased fraction of data offloaded.](/figures/fig-10-offloading-ratio-of-delayed-ap-transfers-with-various-2zewqien.webp)