Design of Secure Mobile Cloud with Smart Load Balancing

This paper presents a new framework for a smart load balancer to efficiently allocate resources to increase application processing speed for data and request response of memory stored by mobile devices in a secure manner. It investigates current architecture types of mobile cloud networks and presents a testbed to determine new secure methods of data allocation within the network. The paper also discusses resource management, allocation, and provisioning in mobile cloud computing networks.

A Framework and the Design of Secure Mobile Cloud
with Smart Load Balancing
Isaac J. Cushman, Md. Baitul Al Sadi, Lei Chen, Rami J. Haddad
Department of Electrical Engineering, Department of Information Technology
Georgia Southern University
Statesboro, USA
{ic00214, ms12508, lchen, rhadd ad}@georgiasouthern.edu
Abstract—The use of mobile devices has exponentially
expanded in recent years. A device which was made with the sole
purpose of making mobile audio phone calls is now the leading
basis for functionality in the social world. The types of applications
widely vary from audio and video calls, internet browsing,
healthcare applications, to mobile games with online connectivity,
among many others. These applications have expanded the
original idea of what a mobile device could be, however there have
been constant drawbacks to these devices, namely short battery
life and limited available storage memory. Another current issue
that exists with mobile devices is the higher data consumption
when on mobile network data. To solve this problem, it is possible
to use cloud computing to mitigate these large applications and use
less data. Integrating in a mobile cloud system to allocate andstore
these applications will allow for the mobile devices to conserve
battery and memory by avoiding large computational processes.
Another major concern is security breaches resulting in data theft
and/or invasion of privacy. In this paper, we present a new
framework that will allow for a smart load balancer to efficiently
allocate resources to increase application processing speed for
data and request response of memory stored by mobile devices in
a secure manner.
Keywords—Mobile Cloud Framework, Resource Provisioning,
Smart Load balancer, Mobile Resource Management, Security and
Privacy
I. INTRODUCTION
Cloud networks offer many benefits to service providers and
users, however there are notable drawbacks that come along
with these benefits. By offering a cloud network, the service
provider can extend to their user resources on demand through
service packages. Cloud computing networks (CCN) are
designed so that many users can be virtually connected into the
same space, as tenants, and rely on the cloud network to store
and/or utilize their data [1]. CCNs are complex networks
consisting of tenants sharing the same space, but with varying
levels of needed security. A proposed approach to assuring
levels of security to different users is addressed later in this
paper. A quickly evolving branch of CCNs is the Mobile Cloud
Computing (MCC). MCC has the potential to overcome the
constraints of the performance of mobile entities, such as
computational power, storage, bandwidth, heterogeneity and
scalability [2]. The recent mobile standard Long-Term
Evolution (LTE) is supporting the cloud augmentation as new
generation mobile applications are needed to overcome the
limitations of computation [2]. Next generation application data
are no longer static as there is much more diversity in mobile
applications [3]. To handle such dynamic data, dynamic
resource management can be used by dynamic resource
allocation technique in a virtual cloud system [3]. This concept
allows users to avoid having to purchase large packages that
may include many other pieces of software or too much
processing power for the required use. The driving force behind
this is known as “as-a-service”, where software, platforms or
infrastructures are offered to the user virtually. A new business
owner will be able to maintain their entire business operation
on a single machine without needing the complete knowledge
of how to configure and operate their operating systems and
servers as all the backend processes and procedures will take
place on the cloud server side. Resource allocation and data
management within mobile clouds have a variety of challenges
that have previously been researched, most critically of which
are: heterogeneity of data, availability to the network,
offloading, and security and privacy [4].
Mobile devices are not on ly the medium of verbal
interactions but also the intermediate of user accessing,
managing and preserving multimedia data [4]. A mobile device
itself is a source of media-rich application data. From high
definition multimedia to spreadsheet data, from GPS location
information to medical records, from banking information to
regular grocery activates, all are processed in the mobile device.
Much of the data preserved in the mobile device are considered
as private or sensitive data. Hence, mobile data deserve greater
security and privacy. Whether the data is highly important like
banking account information or casual such as daily notes, they
all demand availability, reliability, consistency, redundancy,
integrity and security. Mobile Cloud Computing (MCC) has the
potentiality to meet all such demands and the expectation to
augment the computational limitation of the mobile device. The
computational gain, which is achievable from the mobile cloud
may play an important role to improve the overall performance
of mobile devices [5].
The contributions of this paper investigate current
architecture types of mobile cloud networks and present a
testbed to determine new secure methods of data allocation
2017 5th IEEE International Conference on Mobile Cloud Computing, Services, and Engineering
978-1-5090-6325-3/17 $31.00 © 2017 IEEE
DOI 10.1109/MobileCloud.2017.41
205

within the network. The testbed is in current production as
simulation and then physical servers to deploy the smart load
balancer to service multiple virtual machine tenants. The
remainder of this paper is organized as follows: section II
discusses how OpenStack is used in the design of the cloud
platform; section III provides an insight on current technologies
and research on resource management, allocation and provision
in mobile cloud, as well as dynamic resource allocation; section
IV provides the proposed solution in both simulation and
hardware, and section V draws the conclusions.
II. OPEN STACK CLOUD DESIGN
In many academic cloud deployments, open source allows
the deployment of cloud networks without the need of expensive
licenses. This project uses OpenStack to build and test the cloud
network. OpenStack is open source and was selected in this
project thanks to its large community of developers in both
industrial and academic cloud deployments. A key strength that
comes with developing a computing network using OpenStack
is that the cloud models can have a variety of configurations to
serve a task with excellent flexibility. Some example uses are
public cloud, high throughput computing, web hosting, and
video processing and content delivery, etc. [6]. The architecture
built for this project requires three components: the controller,
compute and network nodes. The controller node is responsible
for running the virtual machine Identity and Image services,
management portion of compute node and the dashboard. The
compute node is responsible for running the hypervisor that
operates tenant virtual machines or instances, and connects
network plug-ins and firewall services. Lastly, the network node
is responsible for providing switching, routing, Network
Address Translation (NAT) and Dynamic Host Configuration
Protocol (DHCP) [7].
The OpenStack cloud design has several key pieces of
software that enable the server to operate a cloud network. Each
of the required software used in this project and their
descriptions are listed below. It should be noted there is optional
software available, however they are not generally required by
the cloud network. The following OpenStack projects, listed by
their project names, are defined in [8]:
Keystone: Authentication and authorization service that
operates as the identity of the cloud network. It connects all other
OpenStack services.
Glance: Operates as the image service for the cloud network;
this software is responsible for creating, editing and
provisioning virtual machines. Glance will manage any of the
users that need resources while in the cloud.
Neutron: Establishes the internal and external bridge
connections between each of the nodes and the other OpenStack
services; provides an API to define networks and build network
topologies and configure network policies.
Nova: Manages the lifecycle of compute instances in an
OpenStack environment. This includes spawning, scheduling
and decommissioning of virtual machines on demand.
Cinder: This software is the block storage on the cloud network;
it provides persistent block storage on the instances created by
Glance.
Horizon: This software is the web-based systemthat allows for
the cloud provider to quickly access and manages each of the
services in the cloud outside of the command line interface.
Horizon can create, edit and remove virtual machines, users, and
hypervisors inside the network.
The controller node requires Keystone, Glance and Horizon to
be installed on it. To create a network connection between the
other two nodes, networking settings on IPv4 are established
and a portion of Nova on the compute node will call toward the
IP address of the controller. Some nodes will have more
software than others operating on them, and therefore it is
necessary to also include more block and object storage onto
one or more of these nodes. This allows for maximu m
processing capabilities for the entire system.
III. RESOURCE MANAGEMENT , ALLOCATION AND
PROVISIONING
Resource allocation within mobile cloud computing
networks has been presented in several different ways, typically
generating a cost function per the efficiency of the required
request. The work in [9] presents an adaptation where the
overall cloud network is not localized and requires mobile
social users, cloud brokers and a mobile cloud. When a request
from a social user is presented to the broker, a cost for the
resources is determined and the request is sent to the cloud.
When the cloud broker negotiates higher or lower costs, the
mobile user would then make the decision to connect. Their
work presents a game theoretic method of resource allocation
for better energy efficiency. Another propos al made by the
authors in [9] aims to reduce the overloading on the cloud by
optimizing user traffic through segmenting the data. In this
manner, incoming tasks can be organized in a more dynamic
order to appear as if there is less traffic coming in. While
solutions developed have aimed to solve specific issues, mobile
cloud lacks a common framework that will dynamically
determine the needs of the system based on the user requests.
To determine the ability of a load balancer to efficiently
handle these problems, a measure of quality of service (QoS) is
conducted. QoS can be considered as severaldifferent measures
dependent on the system that is being observed. In mobile cloud
computing, the important factors of QoS are the ability to
remain connected to the network and the overall throughput of
the data. Network connectivity and reliability among mobile
carriers has significantly increased, however there are still areas
where dead zones exist. Lack of availability in a system where
major computation and storage for a mobile device takes place
becomes a major concern. Mobile cloud is a technology
supporting online dynamic resource allocation enabled
services. Dynamic Load Balancing (BLD) mechanism can be
used to distribute the resources by maintaining scalable
workload among every node in the network. Features like
resource optimization, diminishing of response time and down
time, maximizing the throughput, avoiding of overload can be
obtained by Dynamic Load Balancing techniques [11 ].
206

A. Dynamic Resource Allocation
In resource allocation, one of the challenging parts is to
categorize the mobile resources per its priority factor. The
priority factor can be assigned per its requirement, time
sensitivity and the size or space of the data. For example, if
there is an application in the mobile device that deals with real-
time voice or gaming data, undoubtedly such data is highly time
sensitive. Similarly, certain applications are required to access
and process the data immediately, depending on the time
sensitivity of the data, such as video broadcasting and
streaming. To explain further, a variety of data that is stored on
a mobile device does not require continuous synchronization or
need to be processed immediately when created and therefore
can be stored in mobile cloud storage. Subsequently, this kind
of data can be considered as less prioritized data. Some
applications, such as High Definition (HD) video capturing,
may generate large amount of data, hence they may consume
large amount of storage in mobile devices and therefore may
affect the overall performance of these devices. In this case,
data from the mobile devices can be sent by sensing the
available space in the mobile devices. If the mobile device does
not have enough space, it should send the data to the cloud
immediately. Otherwise, when there is enough space in the
mobile device, a certain predefined schedule can be set to
transfer the data.
With the context of the origin of the mobile data, data can be
categorized as follows:
x User Generated Data: Such data can be referred to as the
data generated by the user according to the requirement of
the user, such as contact information, text messages,
captured photos and videos, created personal notes.
x Application data: All mobile application driven data can be
classified as application data, such like email applications
data, GPS information, map information, social networking
data, various gaming and application data, etc. Some of the
application data may require frequent access as per user
demand basis or application requirement basis.
x System data: All data associated with the system
information, system files, system configuration belong to
this category.
B. Smart Load Balancer
In cloud computing, load balancing is defined as the ability
for the system to take incoming application data from the user,
measure the computational requirements and determine which
of the availability zones it needs to be stored in. It is also
required to handle any incoming data to an application so that
the processing ability of that application is not overloaded [12].
The load balancer will have two main functions, finding the
best location that information should be stored and finding the
best path a request should take to retrieve the information. In
mobile cloud networks, this poses a problem, due to the
heterogeneity of the incoming and outgoing data types and
security. In current load balancing methods, the request from
the user is granted based on the current availability in each of
the zones and if the request can be filled without overcoming
the system. The overall basis of how a cloud load balancer is
deployed can be categorized as either in software or in hardware
[13]. From the related work, it is possible to classify sever key
characteristics that are involved when developing a load
balancer for mobile cloud computing. The first to discuss is the
ability to scale up and down in the network. The work in [14]
focuses on the ability for a load balancer to react to the growth
of web applications. When many more machines are added to
the system, the algorithm for load balancing must adapt to this
change. The next characteristic to observe is time based load
balancing. In the work presented in [15], the proposed
algorithms are round-robin, equally-spread current execution
load algorithm, and active VM load balancing.
In the round-robin algorithm, a randomized list of all the
virtual machines is generated and sorted into a list for
processing. The fallback of this method is that certain nodes can
be consistently missed in very large networks. In equally-
spaced current execution, it was noted that the load balancer
was completely in charge of determining the selection of the
VMs. This system works well in terms of overall execution
time, however as addressed in [15], a minor fault in the load
balancer would cause a catastrophic problem to the entire
system. In the active VM load balancing, all the requests made
by each of the VMs would be logged and the least used VM
would be placed at the top of the priority list when resources
are allocated. The drawback to this system alone is that users in
need of using large amounts of data would have less privilege
in acquiring resources compared to users that do not necessarily
need access. Each of the methods stated above can serve as a
foundation candidate for load balancing with appropriate
modifications .
A smart load balancer will be able to intelligently define the
incoming requests by predicting the needs of the request based
on the data type. It will be possible to utilize the discussed
methods in part within the algorithm of the smart load balancer
to effectively maintain large networks. This proposed method
of a smart load balancer will establish a set level of Need before
Greed (NBG) in the system when requests are made. This
parameter is used to determine whether the request from the
user should be granted based on total resource capacity
required, type of data, or the priority level of the user. When
allocating resources, NBG will consider priority users that
absolutely require the system before any others . Examples of
this would be mobile service providers granting a mass
broadcast of emergency information to all users. The process of
how to determine which category and more specifically the
degree of need or greed the tenant is in is determined based on
total system requirement. The total system requirement will
track the weight of each request coming into the SLB and
determine the total processing power reque sted by the tenant.
Examples of what determines need would be if the request is
for high bandwidth applications such as video broadcasting or
high level signal processing computing; whereas a low-level
greed would be webpage browsing.
The metric of NBG is discussed below. A problem is
created on when and how to fairly distribute available resources
207

to each user. Mobile cloud networks are intrinsically large with
a varying amount of data types. To properly design a load
balancer. The idea of need before greed is a method to establish
a protocol where all users will agree that whoever truly needs
to have the most resources will be granted it first. In the event
of multiple users with a need for data, or when one user has
constantly needed the resources, the system will then establish
an algorithm for fairness. In many cases, the level of fairness in
a system is dependent on the current usage of one user
compared to another. For example, if the system is aware that
one user has been granted a large amount of the resources for
an extended period, that user may end up at the end of the queue
when it sends another request. This proposed method aims to
solve issues seen from load balancers that allocate resource
based on usage or systems that use timing as a metric by
applying this need before greed metric.
IV. PROPOSED SOLUTION
The framework proposed in this research is primar ily
developed using simulation based software. However, a
hardware replication is also currently being developed.
A. Simulation Basis
Since the aim of the mobile cloud is to reduce the load of
computing power and mobile storage from mobile devices, it is
necessary to transfer mobile data from these devices to the
mobile cloud and process them in the cloud.
There are several preceding steps prior to sending mobile
data from mobile devices to the mobile cloud. The foremost step
is to verify the data whether there are any security concerns. For
this model, data are divided into five different security classes
according to its severity and shareability from class-1 to class-5.
Class-1 is categorized with the top secret data, and the following
categories contain less secrete data respectively (Table 1).
T ABLE I. DATA SECURITY CLASSES
Class Data
severity Examples
Security
Class-1
Top Secret
Data
Very personal information
without sharing, e.g. personal
notes, text messages, some
pictures etc.
Security
Class-2 Secret Data
Data shareable with very specific
people or specific group of
people.
Security
Class-3
Confidential
Data
Data shareable with certain group
of people including friends and
family, co-workers, people of
certain community, etc.
Security
Class-4
Restricted
Data
Data shareable with almost
everyone but not certain limited
group.
Security
Class-5
Unclassified
Data
Any kind of application data not
fallen into any of the above
class es.
Data will be processed differently according to its class. For
example, very sensitive data that belong to class-1should not be
sent from mobile devices to mobile cloud, but rather to a private
sector of the mobile cloud through a secure encryption method.
It is also suggested to have a cloud that supports multi-layer
structure, which may enhance a layer of security and share the
processing capacity as well as storage load. The multilayer cloud
structure refers to a cloud infrastructure cconsisting public
cloud, private cloud, hybrid cloud and community-based cloud.
In Fig. 1, the proposed approach of data processing in multilayer
cloud structure per security class is shown.
Once the ascertainment of the security issues is
accomplished, data from all three categories (User Generated
data, Application data and System data) are required to be
separated according to its demand (Fig. 2). Mobile cloud API is
a universal mobile application that is responsible for interfacing
from the mobile device to the mobile cloud. During this process,
mobile API is required to accomplish a number of critical tasks,
including classifying the data, assigning security flag based on
the security classifier, scheduling, assigning the priority of the
data and resource management. Mobile API enables a unified
platform for all operation systems. When mobile API takes the
responsibility to store and process data from mobile cloud, the
applications of the cloud will be operation system independent,
hence enhancing the mobility of mobile applications.
Fig. 1. Mobile Cloud Application Program Interface Architecture
Fig. 2. Data categories located within the Data Classifier Layer
208

Fig. 3. Security Class levels
To construct a model of a cloud computing and simulate the
environment, the JAVA based toolkit CloudSim provides the
framework for cloud infrastructures and services. This extensive
simulation toolkit makes available the major cloud components,
including Cloud Information Service (CIS), Data Center,
Broker, CloudLet, Host, Virtual Machine (VM), etc. Fig. 4
illustrates the Basic model of CloudSim.
Fig. 4. Basic Model of CloudSim
B. Research Tasks
The two main tasks of this research are: the design of a
private cloud network that operates within the university
network and the design of an algorithm that will dictate how
resources will be allocated and processed. The testbed will
allow us to simulate a multi-user cloud environment and
analyze the accessibility of various types of users. This
environment provides the ability to analyze possible security
issues in mobile cloud, including but not limited to protection
of highly confidential data, causes of slow request rates,
maximum capacity of requests at peak times and overall
availability of the system to the users.
C. Experimental Framework
Each of the machines in the implementation runs Ubuntu
Server 14.04 and has OpenStack cloud software installed in
order to store and provision virtual machines. In this model, the
lead node is the controller, which maintains the communications
between database storage of the users, software and permissions
among the nodes. The main control station uses the Ubuntu
Metal As A Service (MAAS) as a means to install software and
updates. MAAS allows for very easy scale-up and scale-down
of physical machines, thanks to the fact that any server
connected is simply seen as clusters of virtual machines. One
cluster contains the nodes and in each node runs the required
software for the cloud. The network topology of the
experimental framework is demonstrated in Fig. 5.
Fig. 5. Network Topology for Private Cloud
The MAAS controller has two network interface cards
(NICs) to keep Internet communication with the university
network IP address and to host the private set of IP addresses.
The cloud computing system is hosted on the private network.
For each machine to be added into the cluster, they are each
given an IP address and the gateway IP is the same as the MAAS
controller private IP. In this model, the MAAS controller is
given a class A private IP address 10.0.0.41 and the gateway IP
for each of the machines is configured with the same IP address.
With each of the nodes on the private gateway, they are each set
to boot from the network connection, such that the MAAS
controller will automatically assign each an IP address while
also running a script file to gather the machine information.
The MAAS controller uses an interface to host each of the
machines that are connected to it. OpenStack Autopilot pulls
from the same pool of machines to create the private cloud.
209

OpenStack cloud allows for the cloud provider to access the
overall configurations through the dashboard login and at the
same time allows for users to log in only with their level of
access (least privilege). From the dashboard, the cloud provider
can issue resources in the form of virtual machines to meet the
needs of their users. Similarly, the users can log into their
designated portion of the cloud to access their data, operating
systems and other applications they may have saved. One piece
of software that allows for monitoring how resources are being
distributed is the OpenStack Horizon Dashboard web interface.
The physical system hardware of this proposed framework
can be observed in Fig. 6. The controller, compute and network
nodes can be observed as a stack of three-rack system, each of
which has at least one terabyte of available hard drive space.
Each of the nodes are connected to a switch, which is connected
to the MAAS controller. From the MAAS controller, the private
cloud network can access an Internet connection through NAT
between both network interface cards. The network
administrator can log into OpenStack and provision data.
Fig. 6. Front View of Hardware used for Mobile Cloud Network
To test and implement data and resource allocation within a
mobile network, it is possible to sanction portions of the clouds
resources to recreate a real mobile network of multiple users.
The mobile cloud framework will then be able to provision
resources to sets of mobile users when needed and the load
balancer can then be tested.
V. CONCLUSION
In this paper, we have presented the current needs of mobile
cloud networks for mobile users. It has been shown that mobile
devices will need to access the mobile cloud to reserve
processing power and battery life. A proposed testbed for the
smart load balancer is presented. A simulated framework was
then created to describe how the network will handle resources
and a physical system has been implemented. Future work will
then bring to use a smart load balancer to handle large data of
varying degrees of security and necessity to help mobile devices
operate more efficiently.
ACKNOWLEDGMENT
This research has been funded by the Georgia Southern
University College of Engineering and Information Technology
(CEIT) Faculty Research Seed Grant.
REFERENCES
[1] S. Khan, A. Gani, A. Wahid, et. Al. Towards an Applicability of Current
Network Forensics for Cloud Networks: A SWOT Analysis. IEEE
Access, vol 4, pp. 9800-9820, 2016.
[2] N. Chalaemwongwan and W. Kurutach, "Mobile Cloud Computing: A
Survey and Propose Solution Framework" in 13th International
Conference on Electrical Engineering/Electronics, Computer,
Telecommunications and Information Technology (ECTI-CON), Chiang
Mai, 2016.
[3] S. Thamarai Selvi, C. Valliyammai, G. P. Sindhu and S. Sameer Basha,
"Dynamic resource management in cloud," in Sixth International
Conference on Advanced Computing (ICoAC), Chennai, 2014.
[4] H. Hu, Y. Wen, H. Wang and A. Begen, "Cloud mobile media," in China
Communications, 2016.
[5] Yuan Zhang, Jinyao Yan and Xiaoming Fu, "Reservation-based resource
scheduling and code partition in mobile cloud computing," in IEEE
Conference on Computer Communications Workshops, San Francisco,
CA, 2016.
[6] "OpenStack Open Source Cloud Computing Software", OpenStack, 2016.
[Online]. Available: http://www.OpenStack.org/software/sample-
configs. [Accessed: 13-Sep- 2016].
[7] "Chapter 1. Architecture -OpenStack Installation Guide for Ubuntu 14.04
- juno", Docs.OpenStack.org, 2016. [Online]. Available:
ht t p://docs.OpenSt ack.org/juno/install-
guide/install/apt/content/ch_overview.html. [Accessed: 14-Sep-2016].
[8] [8] "OpenStack Docs: Overview", Docs.OpenStack.org, 2016. [Online].
Available: ht t p://docs.OpenSt ack.org/liberty/install-guide-
obs/overview.html. [Accessed: 14- Sep- 2016].
[9] Z. Su, Q. Xu, M. Fei and M. Dong, "Game Theoretic Resource Allocation
in Media Cloud With Mobile Social Users", IEEE Transactions on
Multimedia, vol. 18, no. 8, pp. 1650-1660, 2016.
[10] X. Wang, Y. Sui, C. Yuen, X. Chen and C. Wang, "Traffic-aware task
allocation for cooperative execution in mobile cloud computing",
IEEE/CIC International Conference on Communications in China, 2016.
[11] F. Liu, P. Shu, L. Ding, J. Yu, D. Niu and B. Li, "GEARING
RESOURCE-POOR MOBILE DEVICES WITH POWERFUL
CLOUDS: ARCHITECTURES, CHALLENGES, AND
APPLICATIONS", IEEE Wireless Communications, 2013.
[12] J. Tai, J. Zhang, J. Li, W. Meleis, and N. Mi, “ARA: Adaptive Resource
Allocation for Cloud Computing Environments under Bursty
Workloads”, IEEE International Performance Computing and
Communications Conference, 2011.
[13] S. Heinzl and C. Metz, “Toward a Cloud-ready Dynamic Load Balancer
based on the Apache Web Server” Workshop onEnabling Technologies:
Infrastructure for Collaborative Enterprises (WETICE), 2013.
[14] T. Fankhasuer, Q. Wang, A. Gerlicher, C. Grecos, and X. Wang, “Web
Scaling Frameworks: A novel class of frameworks for scalable
webservices in cloud environments” IEEE International Conference on
Communication, 2014.
[15] C. Madhumathi and G. Ganapathy, “An Effective Time Based Load
Balancer for an Academic Cloud Environment” International Conference
on Computer Communication and Informatics, 2015
210