Journal of Network and Computer Applications 59 (2016) 46–54

Contents lists available at ScienceDirect

Journal of Network and Computer Applications

journal homepage: www.elsevier.com/locate/jnca

Dynamic energy-aware cloudlet-based mobile cloud computing model

for green computing
Keke Gai a, Meikang Qiu a,n, Hui Zhao b, Lixin Tao a, Ziliang Zong c
a
Department of Computer Science, Pace University, NY, USA
b
Software School, Henan University, Kaifeng, Henan 475000, China
c
Department of Computer Science, Texas State University, TX, USA

art ic l e i nf o a b s t r a c t

Available online 11 June 2015 Employing mobile cloud computing (MCC) to enable mobile users to acquire benefits of cloud computing
Keywords: by an environmental friendly method is an efficient strategy for meeting current industrial demands.
Mobile cloud computing However, the restrictions of wireless bandwidth and device capacity have brought various obstacles,
Green computing such as extra energy waste and latency delay, when deploying MCC. Addressing this issue, we propose a
Dynamic program dynamic energy-aware cloudlet-based mobile cloud computing model (DECM) focusing on solving the
Energy-aware additional energy consumptions during the wireless communications by leveraging dynamic cloudlets
Cloudlets (DCL)-based model. In this paper, we examine our model by a simulation of practical scenario and
provide solid results for the evaluations. The main contributions of this paper are twofold. First, this
paper is the first exploration in solving energy waste problems within the dynamic networking
environment. Second, the proposed model provides future research with a guideline and theoretical
supports.
& 2015 Elsevier Ltd. All rights reserved.

1. Introduction optimize the cloud-based infrastructure usages and services for

achieving green computing. This model uses dynamic program-
Mobile cloud computing (MCC) is an emergence of multiple ming to assist cloudlets cloud computing resources within a
Internet-based technologies development, which enables mobile changing operational environment. The intention of DECM
users to acquire benefits of cloud computing and achieve green matches practical demands of mobile industry because various
computing by using their mobile devices (Sabharwal et al., 2013; elements can have major influences on the cloud services quality.
Bonino et al., 2013). The technology mainly derives from three For example, mobile users who are using map services highly rely
hemispheres, including mobile computing, mobile Internet, and on the speed of wireless communications while the mobile
cloud computing. Combing the advantages of multiple techniques devices are rapidly moving. Nevertheless, unstable and inefficient
allows users to offload data processing and storage to the cloud- wireless connections usually shorten the battery life.
based servers (Huang et al., 2011; Kumar and Lu, 2010). However, Many researchers and scholars have done various achievements in
behind the benefits of adopting this approach, the implementa- energy-aware mobile cloud computing in previous research. The
tions of MCC are still facing a few challenges that limit its research is diverse in different perspectives (Gupta and Roy, 2013).
performance, such as energy over consumptions while the wire- Zhu and his team (Yang et al., 2014) developed a real-time tasks
less communications are weak (Guan et al., 2011). Keeping oriented virtualized cloud computing system that was designed to
searching wireless signals can dry out the power of mobile achieve energy-aware scheduling in their recent works. The proposed
devices, which may cause unexpected energy waste (Han et al., solution (Yang et al., 2014) intends to integrate various energy-aware
2011). scheduling algorithms by employing a rolling-horizon optimization
In this paper, we propose an advanced dynamic model, dynamic policy. However, this approach did not consider mobility usage and
energy-aware cloudlet-based mobile cloud computing model the similar research focusing on energy-aware cloud computing
(DECM), which uses cloudlets technique to assign, manage, and systems has been accomplished by other scholars (Mezmaz et al.,
2011; Beloglazov and Buyya, 2010; Berl et al., 2010).
Furthermore, as one of the core techniques in cloud computing,
n
Corresponding author. virtual machine (VM) is considered an efficient approach for
E-mail address: [email protected] (M. Qiu). building up cloud-based datacenter to achieve green computing

http://dx.doi.org/10.1016/j.jnca.2015.05.016
1084-8045/& 2015 Elsevier Ltd. All rights reserved.
 K. Gai et al. / Journal of Network and Computer Applications 59 (2016) 46–54 47

(Aksanli et al., 2012). Nevertheless, VM is only a service represen- The remainder of this paper is follows. We provide explana-
tation approach that does not bring much technical innovations, tions about DECM in Section 2. Following the description of the
even though VM has been broadly applied in deploying green model, an example is given in Section 3 in order to demonstrate
Information Technology (IT) industry, such as green data processing, the implementation of DECM in practice. Section 4 proposes and
storage, and transmissions (Baliga et al., 2011; Lee and Zomaya, evaluates DECM algorithm. The experimental results are repre-
2012; Xiao et al., 2013). sented in Section 5. The conclusions are given in Section 6.
Based on the industrial needs, our advanced deployment
model, DECM, offers a unique mechanism to avoid the energy
waste when users are suffering a complicated and unstable 2. Concepts and model
networking environment. The model is a type of web service that
focuses on efficient communications between user devices and 2.1. Green computing
cloud servers. Figure 1 represents a conceptual model of DECM.
Three main components of DECM include mobile device, cloudlets The principle of DECM can enable green computing because the
with dynamic searching, and cloud computing. The relationship model is designed for mainly reducing energy consumptions, which
between cloudlet and dynamic searching is that the cloudlets matches one of the characteristics of green computing as an energy-
provide an operating platform in which the dynamic search is aware feature (Beloglazov et al., 2012; Mukherjee et al., 2014). The
executed. approaches of achieving green computing are from software and
The procedure of the service delivery is followed by the hardware to management, policy, and legal issues. Our model focuses
directions of the arrows. Mobile devices send the service requests on technical side that leverages a few cloud-related techniques, such
to the closest cloudlets before the requests reach the cloud servers. as VM, wireless networks, and dynamic programming.
The cloudlets allocate the cloud servers for better service perfor- Using our proposed schema can reduce the energy consump-
mance determined by a group of constraints, such as nearby server tion in wireless communications. The system reliability is also
locations or networking stability. Dynamic programming is considered in the proposed schema, which aims to guarantee the
applied in cloudlets for adapting to the constant changing context. performance of the system. Saving the energy costs of the com-
Selecting the best solution directed by the dynamic-based cloud- munications can achieve green computing. The goal of performing
lets is the core component of DECM, which is expected to avoid green computing is to apply DECM to reduce energy consumptions
energy waste when switching cloud servers or wireless networks. on mass mobile devices without weakening the performance of
We develop a motivational example explaining the fundamen- cloud services.
tal methods of adopting DECM. The implementation of the
example is a simulation followed by the proposed model. 2.2. Mobile cloud computing
Figure 2 illustrates the fundamental concepts of DECM. Compare
with classic web services, business logics are refereed by cloudlets MCC is the conceptual architecture that combines three tech-
that use dynamic programming to search efficient cloud services. nologies, including mobile Internet, mobile computing, and cloud
This difference is a core session in our model, which is optimizing computing, to enable mobile users to offload data processing and
the utilizations of cloud resources in mobile cloud. Detailed storage onto clouds via wireless networks and mobile devices
information of fundamental concepts for DECM is given in the (Dinh et al., 2013; Kumar and Lu, 2010; Guan et al., 2011; Song and
following section. Su, 2011). The motivation of applying MCC is gaining benefits of
The main contributions of this paper are twofold. cloud computing technologies by leveraging mobile techniques.
The dynamic networking environment results in more compli-
1. This research is the first attempt on the functionality of cated service deployments and implementations, comparing with
cloudlets in order to achieve energy-aware performances in basic cloud computing.
the dynamic networking environment. The first basic part of MCC is mobile Internet, also known as
2. The results of this research provide theoretical supports and wireless networks, that is a series of computing networks for
explorations. The model may be migrated and applied in achieving wireless communications by connecting network nodes
multiple domains. The model may be able to be migrated and and using wireless protocols. A variety of mobile Internet techni-
applied in multiple industries, which requires further research ques are available for different demands (Chan et al., 2011). Main
for identifying and proving. techniques of wireless networks include wireless personal area
networks (WPAN), wireless local area networks (WLAN), wireless
metropolitan area networks (WMAN), wireless wide area networks
(WWAN), and cellular networks. Among these techniques, the
cellular network is an approach of deploying wireless commu-
nications, such as 3G and long-term evolution (LTE).
Moreover, supported by mobile Internet, mobile computing
(Satyanarayanan, 2011) is a group of techniques for executing wireless
Fig. 1. A high level conceptual model of DECM. communications on both software and hardware sides, such as mobile

Fig. 2. Fundamental concepts for DECM.

48 K. Gai et al. / Journal of Network and Computer Applications 59 (2016) 46–54

devices, protocols, and networking infrastructure. The concept
Public dynamic cloudlets: Public DCLs stand for a public-purpose
includes most current mobile devices, such as portable computers, access to the DCL-based cloud computing services, which may
smartphones, and tablets. One of the restrictions of using mobile support multiple service providers. We give an example that a
computing is that most devices still highly rely on the battery life. A dynamic searching services can be delivered by deploying
high performance usually conflicts with the durations of battery public DCLs since the services maybe supported by multiple
power and less efficient workload generates energy waste that is service providers who present various data mining offerings.
non-environmental friendly.
Furthermore, cloud computing is a core component of MCC
because it provides service deployment framework and mechanisms. 2.4. Web service
The concept of cloud computing is leveraging Internet-based technol-
ogies to perform computer resources sharing and mask technical Web service states and defines the fundamental of DECM. The
complexities so that cloud customers can access computing services concept of web service is a set of techniques for executing
according to their real-time demands without technical obstacles (Gai machine-to-machine communications via wireless networks. As
and Li, 2012). In MCC, both data processing and data storage can be shown in Fig. 2, service representation is given by the client
migrated onto the cloud-based servers, which is an energy-aware applications, which can be supported by application programming
approach for mobile devices while the wireless communications are interface (API). Proxy receives the service requests who claims the
stable and efficient (Miettinen and Nurminen, 2010; Ma et al., 2012). service delivery capability. XML (extensible markup language)
processors provide web services with a mediative service that
integrates different computing languages and ensures that
2.3. Cloudlets with dynamic programming
machines can understand each other by universally being trans-
lated into a mediatorial language.
Term cloudlet refers to a layer connecting mobile devices and
Once the service request is received by DCLs, business logic is
cloud servers in MCC, which plays a mediator role focusing on the
applied to the requests and the further determinations will be
business logic. The concept of the cloudlet addresses a trusted
made by dynamic programming. Simple object access protocol
party that can be either a computer or a cluster of digital
(SOAP) supports the structured information exchanges, aligning
infrastructure with software that provides cloud users with a
with other web protocols, such as hypertext transfer protocol
rapid response and specific customized functionalities (Verbelen
(HTTP). The same process of translating computing languages
et al., 2012; Fesehaye et al., 2012; Soyata et al., 2012). It is a self-
occurred at the XML processors in order to communicate with
management mechanism that is used to strengthen communica-
various cloud servers. The service responses are sent back from the
tions between mobile devices and cloud servers by reducing
physical servers on the cloud through the CDLs, proxy, and service
latency. For example, a cloudlet can be utilized to transfer com-
representation layer, and finally back to mobile devices.
munication information to mobile nodes by wireless networks
The following section represents a motivational example using
(Fesehaye et al., 2012).
DECM in a practical scenario.
In DECM, cloudlets are deployed with dynamic programming
and we conceptualize it as a dynamic cloudlet (DCL). Mobile cloud
users send out the service requests through the virtual machine 3. Motivational example
(VM) attached to the client applications and the requests will
efficiently reach the nearest cloudlet. The access method to the We exhibit an example in this paper in order to further clarify
DCLs resembles the access points to the public Wi-Fi. Before the the description of DECM. The scenario is that mobile users are
service requests are passed to the cloud, most calculations con- using an app that is a cloud-based real-time acquaintances alarm
cerning the resource allocations are done in DCLs according to the system while users travel between two places. The system offers
business rules. Typically, DCLs are designed to complete at least the service that lets mobile users know whether their acquain-
three missions, which include the following: tances are nearby who are using the same function of the app. In a
traditional way, mobile devices communicate with cloud severs on
1. Leveraging dynamic programming to determine and predict wireless networks directly. This approach keeps the communica-
physical machines on the cloud for current and future use. tions in the box containing the specific cloud servers and mobile
2. Calculations of simple applications and quickly respond to end- devices until the signals or connections become weak or
users. disconnected.
3. Predict whether users should switch to other cloudlets and Compare with the traditional manner, DECM offers a dynamic
enable a real-time suggestion for cloudlets switching. wireless communication that uses CDLs to dynamically select and
determine the most nearby and efficient cloud servers based on
The predictions are based on the services content and other the business logic. Figure 3 represents a conceptual work flow
attached information to the service requests. Cloudlets apply describing the method of DECM operations.
dynamic programming to determine which cloud servers they As displayed in Fig. 3, mobile devices, defined as a mobile user layer,
need to connect with. Energy saving is the main purpose of DECM communicate with cloudlets layer by searching the nearest CDLs while
so that all parameters related to predications and determinations mobile users are accessing the cloud services. Cloudlet A that is
are defined for measuring energy costs. assumed the nearest cloudlet at the beginning of the service does
Moreover, according to the various service demands, there are the calculations based on the business logic and start predicting other
two types of DCL deployments. The deployments are defined from potential cloudlet services according to the service contents, such as
the perspective of cloud services, as follows: multiple locations and dynamic service requests. Once Cloudlet A
confirms that Cloudlet B can offer a better service performance due to

Private dynamic cloudlets: This deployment supports one single the constraints, such as geographic distance and networking condi-
cloud service provider for the purpose of increasing service tions, the connections will switch to Cloudlet B so that mobile users
performance and reducing energy consumptions. For example, can obtain greater services by communication Cloudlet B.
private dynamic cloudlets can be employed for global position Moreover, as we mentioned in Section 2, CDLs are responsible
system (GPS) services that is provided by one service provider. for searching the cloud servers that are nearest or can offer better
 K. Gai et al. / Journal of Network and Computer Applications 59 (2016) 46–54 49

service performance. The communications between cloudlets layer The total cost, C Total ðT; xÞ, consists of two components. The first
and cloud computing layer deliver the service requests and component addresses the energy consumption among mobile
responses and responses in a dynamic manner. The cloud server users, cloudlets, and cloud servers specifically occurred on one
that physically provides services is selected and determined by route, defined as a Ei ðt; xÞ. A route is a preference selection of CDLs
CDLs. Figure 4 illustrates an example of the DECM implementation. implementations.
Depicted in Fig. 4, the starting point describes the position Moreover, considering the service performances, the other side
where the service requests being sent. The multiple positions of is P i ðt; xÞ that defines a performance level localized on one route x
mobile devices identify the locations passed by users during the for one specific operating time unit t. This aspect is for avoiding
service delivery process. Positions are connected by the solid the reduction of service quality while selecting an energy-aware
arrow lines, which exhibits the shift routes of users. Solid color route. Our goal is to generate the minimum energy consumption
dots stand for cloudlets that have potentials of providing services under a specific timing constraint that is often considered a
by connecting with mobile devices. Numerous cloudlets nearby workload planning time. Therefore, the ECM problem can be
the mobile devices imply a dynamic surrounding environment. formulated into the following equation:
Broken lines attached to cloudlets indicate the potential wireless
communications of cloud-based service implementations between Minimize C Total ðT; xÞ ð 8 T A R þ ; 8 x A NÞ ð1Þ
mobile devices and cloudlets, which will be determined by the
dynamic program operated by DECM. The total energy consumption is a sum of the products of
energy cost for each model and performance percentage levels.
Eq. (2) represents the method of generating total energy con-
sumptions. For the meanings of the symbols in the equation refer
4. Algorithm to Table 1.

X
n
4.1. DECM algorithm C Total ðT; xÞ ¼ f ðP i ðt; xÞ; Ei ðt; xÞÞ ð 8 i A N; 8 x A N; 8 t A R þ Þ ð2Þ
x¼1
Table 1 lists a group of predefined elements used for the
simulations and formulations of the solution algorithm in this Using a dynamic programing approach on DCLs aims to select
paper. the most efficient communication between mobile devices and
cloud servers. On the basis of Eq. (2), we generate and propose our
Definition 1. ECM: energy cost minimization problem during the algorithm that is given in the following sections.
wireless communications by adopting cloudlets. Given a set of
cloudlets coded by x, 8 x A N and 8 x Z 1, N ¼natural numbers.

Fig. 3. Conceptual workflow of DECM for motivating example.

Fig. 4. An example of cloudlet-based wireless communications implemented by DECM.

50 K. Gai et al. / Journal of Network and Computer Applications 59 (2016) 46–54

Table 1
Main notations used in this paper and the definitions.

Notations Definitions

x Cloudlet node code, referring to the operation manners aligning with the cloudlet nodes
i Deciding which method route will be used, 8 i A N
Mi (Method route) means which cloudlet route will be used, 8 i A N
t Specific latency or timing cost for each node
T (Execution time unit) referring to the time length of delivering services under requests
P i ðt; xÞ (Service performance) represents the level of service performance at the time node t and cloudlet node x, eg. 0.8 explains 80 percent of full performance
Ei ðt; xÞ (Energy consumption) is the energy cost during the wireless communications at the time node t and cloudlet node x
© An operation that calculates (P, E) ¼ (Pi, Ei)©(Pj, Ej); P¼ Pi  Pj; E¼ Ei þEj
CDLi;t ðM i ðxÞÞ Attributes of one CDL with two variables, i and t, and the corresponding Mi(x)
CostCDL Performance and energy consumption of each cloudlet
C Total ðT; xÞ (Total cost consumption) is an outcome of energy total consumptions considering service performances within a specific scheduling demands
f(P i ðt; xÞ, Total cost function with two variables P i ðt; xÞ, Ei ðt; xÞ
Ei ðt; xÞ)

Table 2 consumptions:
Algorithm of dynamic cloudlets-based module for determining cloud servers. 8Xn
>
>
>
> EnergyCloudlet min ðCDLi;t ðM i ðxÞÞ
Input: P i ðt; xÞ, Ei ðt; xÞ, Ti(x), Mi(x), and Nd(x) >
>
Output: Minimum energy consumptions within a specific timing period <x¼1
f ðP i ðt; xÞ; Ei ðt; xÞÞ ¼ n ð4Þ
>
> ∏ P i ðt; xÞ
1: for x’1 to NdðxÞ >
>
>
> x¼1
2: for m’1 to Mi(x) :
3: for t T ’ to T i ðxÞ
8 i A N; 8 x A N; 8 t A R þ ; 0 o t oT
P
4: P m ðt T ; xÞ’ tt T¼ 1 P m ðt; xÞ
5: Em ðt T ; xÞ’Em ðt T ; xÞ Theorem. By applying the recursive function, each element attached
6: do the comparison and cancel the pair performing worse to the matrix CostCDL[CDLi;t ðM x Þ] is the minimum energy cost for
7: end for
selecting cloudlets by examining the energy cost, Ei ðt; xÞ, and service
8: end for
9: end for performance, P i ðt; xÞ.
10: for t T ’1 to Ti(x)
11: /n calculate every CostCDL from Ndð1Þ to NdðxÞ Proof. The assumption of the execution scenario is that the
12: do the comparison and cancel the pair performing worse n/ mobile device connects to the cloud server with one route M1 at
13: end for the beginning of the service. The scenario considers a common
14: return all CostCDL½CDLi;t ðM x Þ situation for starting a service. The following routes have multiple
options, M1, M2,…,Mn, and the routes may have different values for
variables. For example, route M2 may have a longer latency at the
4.2. Recursive formulation first time-unit than M1. At the starting point, the existence of
P 1 ð1; 1Þ and E1 ð1; 1Þ represent that (i¼1, t¼1, and x ¼1, which are
The crucial component of dynamic programming algorithm is inputs to ECM problem.
to generate the recursive formulation, which drills down the large Moving forward to the next time unit, a decision will be made
size problems into smaller sized sub-problems to make the whole depending on the result of the comparison among all routes,
problem solvable. Selecting the best solution to each sub-problem {P 1 ð2; 1Þ, E1 ð2; 1Þ © P 2 ð2; 1Þ, E2 ð2; 1Þ…P Á n ð2; 1Þ, En ð2; 1Þ}. The route
is one of the principles of dynamic programming, which is (s) with a better performance will be selected. The resolution will
applicable for complicated problems, such as polynomial time be made for each time unit by using the same method and finally
problems (Qiu et al., 2014). In this paper, we propose an approach it reaches a minimum total energy cost with quality services in a
of using dynamic programs to optimize operating manners defined timing period, when t¼tT, x¼ xT, and valued {(P 1 ðt T ; xT Þ,
of DCLs. Á
E1 ðt T ; xT Þ)} © {(P 2 ðt T ; xT Þ, E2 ðt T ; xT ))} …fðP n ðt T ; xT Þ; En ðt T ; xT ÞÞg. The
First, aligning with Eq. (2), we propose an efficient schema that final result will be a set of {CDLi;T ðM i ðxÞÞ} that may consist of only
enables a dynamic switch between cloudlets that can have better one or multiple paths. Associating with restrictions and consid-
service offerings. According to the matrix shown in Eq. (3), the ered parameters, our theorem is a proved energy-aware approach
minimum energy consumption with quality services can derive for the implementations of energy-aware solutions. The next
from the recursive formulation. The performance of each cloudlet section describes our main algorithm. Table 2 the dynamic
is symbolized by CostCDL: programming algorithm we used for our proposed mechanism
8 that derives from Eq. (4).□
>
> EnergyCloudletðCDLi;t ðM 1 ÞÞ ¼ CostCDL½CDLi;t ðM 1 Þ
>
>
< EnergyCloudletðCDLi;t ðM 2 ÞÞ ¼ CostCDL½CDLi;t ðM 2 Þ
5. Experiment and results
ð3Þ
>
> ⋮
>
>
: EnergyCloudletðCDLi;t ðM n ÞÞ ¼ CostCDL½CDLi;t ðM n Þ We evaluate our algorithm by using the scenario provided in
Section 3. The evaluation is based on a simulation that is a
Second, with cognitions of the recursive formulation, we define mathematic deduction by comparing DECM approach with the
an energy unit cost function addressing each DCL as EnergyCloudlet traditional cloud computing method. The experiment can simulate
(P i ðt; xÞ, Ei ðt; xÞ), which represents the total energy costs with the both private and public dynamic cloudlet executions. The work-
service performance at each cloudlet. The total cost function is load has been examined on the simulator named DECM-Sim that
formulated in f(P i ðt; xÞ, Ei ðt; xÞ), which is given in Eq. (4). In this is developed by our lab. The simulator is designed to calculate
case, we do not consider the energy cost that occurred while input data and generate the results with using the dynamic
switching cloudlets because of its small amount of energy programming. A few parameters are defined for the comparison
 K. Gai et al. / Journal of Network and Computer Applications 59 (2016) 46–54 51

Table 3
Energy consumptions and performances between different cloudlet-based
modules.

Nd M1 M2 Nd M1 M2

T1 P1 E1 T2 P2 E2 T1 P1 E1 T2 P2 E2

0 1 0.8 9 2 0.7 4 4 2 0.7 8 4 0.9 5

2 0.2 9 3 0.3 4 5 0.3 8 5 0.1 5

1 1 0.9 7 2 0.8 2 5 1 0.8 7 2 0.9 3

2 0.1 7 5 0.2 2 3 0.2 7 4 0.1 3

2 1 0.9 10 2 0.9 5 2 2 0.9 6 4 0.8 2

3 0.1 10 4 0.1 5 3 0.1 6 6 0.2 2

3 2 0.8 8 5 0.9 3 7 1 0.7 8 3 0.9 4

3 0.2 8 7 0.1 3 3 0.3 8 5 0.1 4

Table 4
Minimum total costs with computed performance capacities under various timing
constraints for a DECM.

T (P, E) (P, E) (P, E) (P, E) (P, E)

11 (0.183, 54.0)
12 (0.206, 50.0) (0.229, 54.0) (0.162, 49.0)
13 (0.258, 50.0) (0.183, 45.0) (0.286, 54.0) (0.142, 44.0) (0.229, 49.0)
14 (0.265, 46.0) (0.327, 54.0) (0.321, 50.0) (0.16, 40.0) (0.258, 45.0)
(0.286, 49.0) (0.203, 44.0)
15 (0.368, 50.0) (0.331, 46.0) (0.236, 41.0) (0.409, 54.0) (0.321, 45.0)
(0.229, 40.0) (0.254, 44.0)
16 (0.339, 43.0) (0.459, 50.0) (0.413, 46.0) (0.206, 36.0) (0.368, 45.0)
(0.331, 41.0) (0.286, 40.0) Fig. 5. Experimental results. (a) The result of DECM approach focusing on energy-
17 (0.51, 50.0) (0.473, 47.0) (0.426, 43.0) (0.303, 38.0) (0.459, 45.0) aware performance (the line marked as Emin refers to the most energy-aware path
(0.413, 41.0) (0.327, 40.0) (0.294, 36.0) (0.525, 51.0) (0.18, 35.0) and the line named Emax is the most costly path produced by adopting DECM). 98
18 (0.183, 32.0) (0.459, 41.0) (0.59, 47.0) (0.531, 43.0) (0.265, 33.0) (b) The result of DECM approach focusing on energy-aware performance (the line
(0.473, 42.0) (0.426, 38.0) (0.368, 36.0) 99 marked as Pmax refers to the highest level of wireless communications path and
⋮ ⋮ ⋮ ⋮ ⋮ ⋮ the line named Pmin is the path providing lowest level of quality).
32 (0.9, 29.0) (0.729, 24.0) (0.81, 28.0) (1.0, 37.0)
33 (0.729, 24.0) (1.0, 37.0) (0.9, 28.0)
34 (0.81, 24.0) (0.9, 28.0) (1.0, 36.0) method routes in which a complicated availability of the cloudlet
35 (0.81, 24.0) (1.0, 33.0) (0.9, 28.0)
nodes is given. For 4-method-route scenario, we produce four
36 (0.9, 24.0) (1.0, 32.0)
37 (0.9, 28.0) (1.0, 32.0)
nodes tested in the second experiment. This configuration simu-
38 (0.9, 28.0) (1.0, 32.0) lates the scenario with a few method routes that consists of less
39 (1.0, 28.0) cloudlet nodes settings.
The purpose of the experiment 1 is to examine whether DECM
is applicable for an operating interval associated with specific
and assessment purposes, including timing units (Tx), wireless timing constraints, to examine the relationships between two
communication performances (Px), energy consumptions (Ex), main variables, Px and Ex, and to evaluate whether there is an
mobility routes (Mx), and switching nodes among different routes optimized solution at each timing constraint when two-route is
(n). Data collection is followed by the simulation research deployed. Meanwhile, the second experiment is designed for
methodology. testing the operability of DECM in a complicated CDLs-connected
The configurations of the experiment are defined as follows. environment. Four routes provide various data in order to simulate
Data collection is followed by the simulation research methodol- a complex networking surrounding.
ogy communication standard for our experiment. LTE advance is The parameters in the experiment examining 2 method routes
an advanced standard for deploying LTE, which enables multiple are represented in Table 3 that explicates a variety of cloudlet
mobile devices to be compatible with various LTE networks. The nodes with diverse energy consumptions and performances based
performance is measured by the wireless communication latency on our assumptions. Nodes represent the switches between cloud
time, which are represented by the variable (Px). The minimum servers, which are symbolized as Nd. Addressing the ECM problem
latency time for each route is defined as a full level performance defined in Section 4.1, we present a minimum total energy cost
that is marked as 1.0. The tested mobile devices are smart phones with the performance capacities under various timing constraints
with Android 4.4.2 and up operating system, which support at for a DECM in Table 4 by using the data from Table 3.
least 300 Mbps max speed, e.g. HTC One (M8), Samsung Galaxy We use the data in Table 3 to evaluate our DECM algorithm and
Note 3, and Samsung Galaxy Note 4. gain the experimental results. The issue represented in Table 4 is
A few method routes are simulated in the experiment and the to illustrate optimized selections of switching CDLs under different
workflow method is given in Fig. 3. Two different experiments timing constraints, which are examined by two parameters,
with 2 and 4 method routes are applied for examining our including energy consumption level (E) and performance capacity
algorithm and the feasibility of the model. We assume that there percent level (P). Under this configuration, the latency of the
are 8 nodes in the simulation of 2-method-route. This experi- mobile environment is involved in the parameter P. For example,
mental configuration simulates the scenario that having few under timing constraint 32, there are four optimal options (0.9,
52 K. Gai et al. / Journal of Network and Computer Applications 59 (2016) 46–54

Fig. 6. Experimental results for two-route at different timing constraints. (a) The result of DECM approach stating wireless communication quality level with the
corresponding energy consumption at 17 timing constraints). (b) The result of DECM approach stating wireless communication quality level with the corresponding energy
consumption at 23 timing constraints). (c) The result of DECM approach stating wireless communication quality level with the corresponding energy consumption at 26
timing constraints. (d) The result of DECM approach stating wireless communication quality level with the corresponding energy consumption at 30 timing constraints.

Table 5 Table 6
Energy consumptions and performances for four modules. Minimum total costs with computed performance capacities under various timing
constraints for a DECM.
Nodes M1 M2 M3 M4
T (P, E) (P, E) (P, E) (P, E) (P, E)
T1 P1 E1 T2 P2 E2 T3 P3 E3 T4 P4 E4
4 (0.454, 31.0) (0.101, 26.0) (0.34, 28.0) (0.076, 23.0)
0 1 0.8 9 2 0.7 4 1 0.6 6 3 0.9 9 5 (0.504, 26.0) (0.378, 23.0) (0.113, 22.0) (0.09, 21.0) (0.067, 18.0)
2 0.2 9 3 0.3 4 3 0.4 6 4 0.1 9 (0.567, 31.0)
6 (0.648, 34.0) (0.504, 25.0) (0.448, 21.0) (0.336, 18.0) (0.101, 17.0)
1 1 0.9 7 2 0.8 2 2 0.4 5 1 0.5 7 (0.63, 26.0) (0.078, 16.0)
2 0.1 7 5 0.2 2 3 0.6 5 2 0.5 7 7 (0.72, 29.0) (0.81, 34.0) (0.448, 20.0) (0.7, 26.0) (0.392, 16.0)
(0.118, 15.0) (0.63, 21.0)
2 1 0.9 10 2 0.9 5 1 0.2 5 2 0.3 4 8 (0.729, 28.0) (0.9, 29.0) (0.72, 24.0) (0.392, 15.0) (0.63, 20.0)
3 0.1 10 4 0.1 5 2 0.8 5 3 0.7 4 (0.7, 21.0) (0.56, 16.0)
9 (0.576, 18.0) (0.13, 14.0) (0.81, 23.0) (0.101, 13.0) (1.0, 29.0)
3 2 0.8 8 3 0.7 7 1 0.7 5 4 0.9 2 (0.9, 24.0) (0.56, 15.0) (0.7, 19.0)
3 0.2 8 4 0.3 7 4 0.3 5 5 0.1 2 10 (0.576, 17.0) (0.504, 13.0) (0.151, 12.0) (0.9, 21.0) (0.81, 18.0)
(1.0, 24.0) (0.56, 15.0)
11 (0.9, 18.0) (0.504, 12.0) (0.81, 17.0) (1.0, 21.0) (0.8, 16.0)
(0.72, 13.0)
29.0), (0.729, 24.0), (0.81, 28.0), and (1.0, 37.0). The option (0.9,
12 (1.0, 18.0) (0.8, 13.0) (0.72, 12.0) (0.9, 16.0)
29.0) means the value of P is 0.9 and the value of E is 29.0. 13 (0.8, 12.0) (1.0, 16.0) (0.9, 15.0)
P E . Compare with another option (1.0, 37.0), the option (0.9, 29.0) 14 (1.0, 15.0) (0.8, 12.0) (0.9, 13.0)
has a lower performance capacity with a lower energy consump- 15 (1.0, 13.0) (0.9, 12.0)
tion. The determination can be made by the priority of the user 16 (1.0, 12.0)
17 (1.0, 13.0)
settings.
P 18 (1.0, 18.0)
Deriving from Eq. (4), we calculate E ¼ CostCDL. T refers to
various timing constraints, fT A N∣11 r T r39g. The bolded data
 K. Gai et al. / Journal of Network and Computer Applications 59 (2016) 46–54 53

Fig. 7. Conceptual workflow of DECM for motivating example.

represent a full-level service status with the corresponding One of the experimental results are depicted in Fig. 7. The figure
energy-saving efficiency, e.g. (1.0, 32.0). represents the energy consumptions for different routes with full
There are a few solution options with different energy cost service performances at four timing constraints, namely 10, 12, 14,
performances and service quality levels when the timing con- and 16. At 10 timing constraints, only DECM approach and M1 can
straints are different. For instance, at the timing constraint 12, accomplish all nodes with full service quality, P¼1.0. Similarly, DECM
there are three approaches, (P, E)¼ {(0.206, 50.0), (0.229, 54.0), approach, M1, M3, and M4 can finish all the nodes at timing constraint
(0.162, 49.0)}. Adding more constraints can assist mobile users to 14. According to the demonstration of the figure, DECM performs
further determine the outcomes depending on the demands. better in saving energy than all other method routes when delivering
Figure 5 illustrates two results obtained from our experiment the full service performance.
that are displayed in two graphics, Fig. 5(a) and (b). These graphics In summary, our experimental results have provided a solid
represent the implementations of DECM from both energy saving evidence showing that the proposed DECM is efficient in saving energy
and communication perspectives. as well as higher-level wireless service quality (Verbelen et al., 2012).
Figure 5(a) exhibits the DECM approach focusing on energy-aware Many previous research focused on the improvement of cloud service
performance. X-axis represents a range of timing constraints that can quality throughout different fields, from architectural to operating
accomplish the task, from 11 to 39, and the Y-axis shows the energy levels (Soyata et al., 2012; Satyanarayanan et al., 2009). Jararweh
consumptions. Emin refers to the minimum energy cost and Emax et al. (2014) proposed a scalable cloudlet-based approach that aimed
means the maximum energy consumption. According to Fig. 5(a), the to reduce energy consumptions and latency by scaling up and down
task can be completed in the interval between Emin and Emax lines. the service. This approach was efficient under a specific condition that
The figure depicts that using DECM can save more energy by defining a implies the users can accept a scalable service delivery manner during
longer timing constraint. the usages. Some other research had other focuses, such as using
Distinguishing from Fig. 5(a), (b) describes the scenario examining cloudlets for efficient data allocation (Quwaider and Jararweh, 2015),
the service quality (P) by two boundaries, maximum performance resource allocation (Bohez et al., 2015), or admission control policy
(Pmax) and minimum performance (Pmin). Both Pmax and Pmin have (Hoang et al., 2012). Nonetheless, none of them had the same target as
a positive relationship with timing constraints, T. According to the ours. We intended and explored to leverage use dynamic programming
representation of Fig. 5(a) and (b), energy consumptions, E have a cloudlets to gain energy-aware MCC approach offering a high-perfor-
negative relationship with communication performances, P. mance service quality. Our approach has been verified as a productive
Moreover, using DECM offers various outcomes at different defined method for solving the cloudlet-based service deployment in a
timing constraints, T. Figure 6 provides four scenarios defined by the dynamic circumstance.
varied timing constraints, namely 17, 23, 26, and 30, which are given by
four sub-figures, including Fig. 6(a)–(d). Each figure represents a trend
of wireless communication performances (P) aligning with the energy
consumptions (E). All outcomes from these figures illustrate that P has
a positive relationship with the amount of E and there is always an 6. Conclusions
optimized solution at each timing constraint. The solution is associated
with the requirements of P and E. This paper proposed a novel approach of using dynamic
Dissimilar to the two-route module, we further made an cloudlet-based MCC model, DECM, to gain the benefits of green
experiment for the four routes module and the parameters data computing. The experimental results proved that our approach
are given in Table 5. Displayed in the table, this experiment defines was an effective mechanism that could enable mobile users to
four routes, namely M1, M2, M3, and M4, and four nodes. address green IT within a dynamic complicated wireless environ-
By using DECM-Sim, we generate a set of solutions to each ment. Future research includes two aspects. The first focus is to
route. Table 6 represents optimized selections of switching CDLs examine whether DECM is applicable in multiple industries with
under different timing constraints determined by P and E. The different service requirements. Another aspect is to build up
shortest timing constraints are 4 timing units and the longest structured connections among CDLs to strengthen communica-
constraints are 18 timing units. tions between cloud servers and mobile devices.
54 K. Gai et al. / Journal of Network and Computer Applications 59 (2016) 46–54

References Kumar K, Lu Y. Cloud computing for mobile users: can offloading computation save
energy?. Computer 2010;43(4):51–6. http://dx.doi.org/10.1109/MC.2010.98.
Lee YC, Zomaya AY. Energy efficient utilization of resources in cloud computing
Aksanli B, Venkatesh J, Rosing T. Using datacenter simulation to evaluate green
systems. J Supercomput 2012;60(2):268–80. http://dx.doi.org/10.1007/s11227-
energy integration. Computer 2012;45(9) 0056–64, doi:http://doi.ieeecompu
010-0421-3.
tersociety.org/10.1109/MC.2012.249.
Ma X, Cui Y, Stojmenovic I. Energy efficiency on location based applications in
Baliga J, Ayre RW, Hinton K, Tucker R. Green cloud computing: balancing energy in
mobile cloud computing: a survey. Procedia Comput Sci 2012;10:577–84. http:
processing, storage, and transport. Proc IEEE 2011;99(1):149–67. http://dx.doi.
org/10.1109/JPROC.2010.2060451. //dx.doi.org/10.1016/j.procs.2012.06.074.
Beloglazov A, Buyya R. Energy efficient resource management in virtualized cloud Mezmaza M, Melabb N, Kessacib Y, Leec Y, Talbib E, Zomayac. A parallel bi-objective
data centers. In: Proceedings of the 10th IEEE/ACM international conference on hybrid metaheuristic for energy-aware scheduling for cloud computing sys-
cluster, cloud and grid computing. IEEE Computer Society; Melbourne, Aus- tems. J Parallel Distrib Comput 2011;71(11):1497–508. http://dx.doi.org/
tralia, 2010. p. 826–31. 10.1016/j.jpdc.2011.04.007.
Beloglazov A, Abawajy J, Buyya R. Energy-aware resource allocation heuristics for Miettinen AP, Nurminen JK. Energy efficiency of mobile clients in cloud computing.
efficient management of data centers for cloud computing. Future Gener In: Proceedings of the 2nd USENIX conference on hot topics in cloud comput-
Comput Syst 2012;28(5):755–68. http://dx.doi.org/10.1016/j.future.2011.04.017. ing. USENIX Association; Boston, MA, USA, 2010. p. 4.
Berl A, Gelenbe E, Girolamo MD, Giuliani G, Meer HD, Dang MQ, et al. Energy- Mukherjee A, Gupta P, De D. Mobile cloud computing based energy efficient
efficient cloud computing. Comput J 2010;53(7):1045–51. http://dx.doi.org/ offloading strategies for femtocell network. In: Applications and innovations in
10.1093/comjnl/bxp080. mobile computing (AIMoC). IEEE; Kolkata, India , 2014. p. 28–35. doi:http://dx.
Bohez S, Verbelen T, Simoens P, Dhoedt B. Discrete-event simulation for efficient doi.org/10.1109/AIMOC.2014.6785515.
and stable resource allocation in collaborative mobile cloudlets. Simul Model Qiu M, Chen Z, Ming Z, Qin X, Niu J. Energy-aware data allocation with hybrid
Pract Theory 2015;50:109–29. http://dx.doi.org/10.1016/j.simpat.2014.05.006. memory for mobile cloud systems. IEEE Syst J 2014:1–10. doi:http://dx.doi.org/
Bonino D, Russis LD, Corno F, Ferrero G. JEERP: energy aware enterprise resource 10.1109/JSYST.2014.2345733.
planning. IT Prof 2013;50–6. doi:http://dx.doi.org/10.1109/MITP.2013.22. Quwaider M, Jararweh Y. Cloudlet-based efficient data collection in wireless body
Chan HA, Yokota H, Xie J, Seite P, Liu D. Distributed and dynamic mobility area networks. Simul Model Pract Theory 2015;50:57–71. http://dx.doi.org/
management in mobile internet: current approaches and issues. J Commun 10.1016/j.simpat.2014.06.015.
2011;6(1):4–15. http://dx.doi.org/10.4304/jcm.6.1.4-15. Sabharwal M, Agrawal A, Metri G. Enabling green IT through energy-aware
Dinh HT, Lee C, Niyato D, Wang P. A survey of mobile cloud computing:
software. IT Prof 2013:19–27. doi:http://doi.ieeecomputersociety.org/10.1109/
architecture, applications, and approaches. Wirel Commun Mob Comput
MITP.2012.104.
2013;13(18):1587–611. http://dx.doi.org/10.1002/wcm.1203.
Satyanarayanan M. Mobile computing: the next decade. ACM SIGMOBILE Mob
Fesehaye D, Gao Y, Nahrstedt K, Wang G. Impact of cloudlets on interactive mobile
Comput Commun Rev 2011;15(2):2–10. http://dx.doi.org/10.1145/
cloud applications. In: IEEE 16th international enterprise distributed object
computing conference (EDOC). IEEE; Beijing, China, 2012. p. 123–32. doi:http:// 2016598.2016600.
dx.doi.org/10.1109/EDOC.2012.23. Satyanarayanan M, Bahl P, Caceres R, Davies N. The case for vm-based cloudlets in
Gai K, Li S. Towards cloud computing: a literature review on cloud computing and mobile computing. IEEE Pervasive Comput 2009;8(4):14–23. http://dx.doi.org/
its development trends. In: 4th international conference on multimedia 10.1109/MPRV.2009.82.
information networking and security (MINES). IEEE; Nanjing, China, 2012. p. Song W, Su X. Review of mobile cloud computing. In: IEEE 3rd international
142–6. doi:http://dx.doi.org/10.1109/MINES.2012.240. conference on communication software and networks (ICCSN). IEEE; Xi'an,
Guan L, Ke X, Song M, Song J. A survey of research on mobile cloud computing. In: China, 2011. p. 1–4. doi:http://dx.doi.org/10.1109/ICCSN.2011.6014374.
Proceedings of the 10th IEEE/ACIS international conference on computer and Soyata T, Muraleedharan R, Funai C, Kwon M, Heinzelman W. Cloud-vision: Real-
information science. IEEE Computer Society; Sanya, China, 2011. p. 387–92. doi: time face recognition using a mobile-cloudlet-cloud acceleration architecture.
http://doi.ieeecomputersociety.org/10.1109/ICIS.2011.67. In: IEEE symposium on computers and communications (ISCC). IEEE; Cappa-
Gupta S, Roy K. Comparison of different energy minimization techniques in wireless docia, Turkey, 2012. p. 059–66. doi:http://dx.doi.org/10.1109/ISCC.2012.
sensor network. Energy 2013;75(18), . http://dx.doi.org/10.5120/13348-0576. 6249269.
Han B, Hui P, Srinivasan A. Mobile data offloading in metropolitan area networks. Verbelen T, Simoens P, Turck FD, Dhoedt B. Cloudlets: bringing the cloud to the
ACM SIGMOBILE Mob Comput Commun Rev 2011;14(4):28–30. http://dx.doi. mobile user. In: Proceedings of the 3rd ACM workshop on mobile cloud
org/10.1145/1942268.1942279. computing and services. ACM; Low Wood Bay, UK, 2012. p. 29–36. doi:http://
Hoang DT, Niyato D, Wang P. Optimal admission control policy for mobile cloud dx.doi.org/10.1145/2307849.2307858.
computing hotspot with cloudlet. In: IEEE wireless communications and Xiao Z, Song W, Chen Q. Dynamic resource allocation using virtual machines for
networking conference (WCNC). IEEE; Shanghai, China, 2012. p. 3145–9. doi: cloud computing environment. IEEE Trans Parallel Distrib Syst 2013;24
http://dx.doi.org/10.1109/WCNC.2012.6214347.
(6):1107–17. http://dx.doi.org/10.1109/TPDS.2012.283.
Huang D. Mobile cloud computing. IEEE COMSOC Multimed Commun Tech Comm
Yang L, Zhu X, Chen H, Wang J, Yin S, Liu X. Real-time tasks oriented energy-aware
(MMTC) E-Lett 2011;6(10):27–31.
scheduling in virtualized clouds. IEEE Trans Cloud Comput 2014;2(2):168–80.
Jararweha Y, Tawalbehb L, Ababneha F, Khreishahc A, Dosarib F. Scalable cloudlet-
http://dx.doi.org/10.1109/TCC.2014.2310452.
based mobile computing model. Procedia Comput Sci 2014;34:434–41. http:
//dx.doi.org/10.1016/j.procs.2014.07.051.