
1
INTRODUCTION
With an ever increasing number of mobile
devices and the resulting explosive growth of
mobile traffic, 5G networks call for various tech-
nology advances to transmit traffic more effec-
tively while changing the world by interconnect-
ing a tremendous number of mobile devices.
However, mobile devices have limited communi-
cation and computation capabilities in terms of
computation power, memory, storage, and ener-
gy. In addition to the broadband bandwidth sup-
port from 5G, cloud computing needs to be
utilized to enable mobile devices to obtain virtu-
ally unlimited dynamic resources for computa-
tion, storage, and service provisioning that will
overcome the constraints in smart mobile devices
[1]. Thus, the combination of 5G and cloud com-
puting technology is paving the way for more
attractive applications that involve compute-
intensive task execution at “small” devices car-
ried by mobile users who enjoy “large”
capabilities enabled through the new technolo-
gies.
ABSTRACT
As mobile devices are equipped with more
memory and computational capability, a novel
peer-to-peer communication model for mobile
cloud computing is proposed to interconnect
nearby mobile devices through
various short range radio com-
munication technologies to form
mobile cloudlets, where every
mobile device works as either a
computational service provider or a client of a
service requester. Though this kind of computa-
tion offloading benefits compute-intensive appli-
cations, the corresponding service models and
analytics tools are remaining open issues. In this
paper we categorize computation offloading into
three modes: remote cloud service mode, con-
nected ad hoc cloudlet service mode, and oppor-
tunistic ad hoc cloudlet service mode. We also
conduct a detailed analytic study for the pro-
posed three modes of computation offloading at
ad hoc cloudlet.
With the support of mobile cloud computing
(MCC) [2], a mobile user basically has one more
option to execute the computation of their appli-
cations, i.e., offloading the computation to the
cloud [3]. Thus, one principal problem is: under
what conditions should the mobile user offload
t
he required computation to the cloud [4]. An
illustrative scenario of computation offloading to
a remote cloud is shown in Fig. 1a, where the
user is covered by WiFi. Since the terminal
device at the user end has limited resources, i.e.,
hardware, energy, bandwidth, etc., the cellphone
itself is infeasible to finish some compute-inten-
sive task. Instead, the data related to the compu-
tation task can be offloaded to the remote cloud
v
ia WiFi which is free of charge. After the exe-
cution of the computation task in the remote
cloud, the computation result will be sent back
to the user’s cellphone. We call this conventional
computation offloading mode “remote cloud ser-
vice” (RCS), as shown in Fig. 1. Typically, WiFi
is used in RCS and user mobility is strictly limit-
ed to areas covered by WiFi. In the case that a
mobile user moves into an outdoor environment
without WiFi coverage, the computation task
must be offloaded to the remote cloud via a cel-
lular network, which results in high communica-
tion cost. Compared with traffic offloading [5, 6],
applications involving computa-
tion offloading usually are more
delay-sensitive. Thus, how to
design a cost-effective computa-
tion offloading service mode
while achieving good quality of experience
(QoE) becomes an important problem to be
solved in this work.
As mobile devices are equipped with increased
memory and computational capability, a novel
peer-to-peer communication model for MCC is
proposed to interconnect nearby mobile devices
through various short-range radio communica-
tion technologies (e.g., WiFi, Bluetooth, etc.) to
form mobile cloudlets, where a mobile device
can work as either a computational service
provider or a client of a service requester [7]. In
recent years, direct short range communications
between cellular devices, known as device-to-
device communications (D2D), has been widely
studied, with a comprehensive discussion of relat-
ed usage cases and business models given in [8].
In [9], a fundamental question while utilizing a
cloudlet is proposed as follows: whether and
under what conditions a mobile cloudlet is feasi-
ble for providing mobile application services. To
answer the question, cloudlet size, a cloudlet
node’s lifetime and reachable time are defined.
However, computation offloading is only allowed
when a computation node (i.e., a cloudlet service
requester) stays in contact with a service node
(i.e., a cloudlet service provider) [9]. In this
paper, we denote the previous computation
offloading mode for cloudlet assisted service as
“connected ad hoc cloudlet service” (CCS).
Under CCS mode, the computation task is
offloaded to another local cloudlet node when
there are available D2D connectivities during
the whole procedure, including computation
offloading, computation execution, and compu-
tation feedback. Though local D2D connectivity
might be unavailable due to network dynamics,
ON THE COMPUTATION OFFLOADING AT
AD HOC CLOUDLET:
A
RCHITECTURE AND SERVICE MODES
With the development of mobile devices in terms of increased memory and computational capabili-
ty, a novel peer-to-peer communication model for mobile cloud computing is proposed to intercon-
nect nearby mobile devices through various short range radio communication technologies to
form mobile cloudlets, where every mobile device works as either a computational service
provider or a client of service requester.
Min Chen, Yixue Hao, Yong Li, Chin-Feng Lai, and Di Wu
L
s
b
a
sk
e
edb
a
ck
d subtask
t to Rachel
D connecton
Offload
subtask
Offload
subtask
l
Suri
Eva
......
123
3
4
1234
4
Cloud
COMMUNICATIONS
T A N D A R D
SS
Min Chen and Yixue Hao
are with Huazhong
University of Science and
Technology.
Yong Li is with Tsinghua
University.
Chin-Feng Lai is with
National Chung Cheng
University.
Di Wu is with Sun Yat-sen
University.
IEEE Communications Magazine — Communications Standards Supplement • June 2015
0163-6804/15/$25.00 © 2015 IEEE