Mobile cloud analytics in Big data era

ZORAN BOJKOVIC, DRAGORAD MILOVANOVIC

University of Belgrade

Studentski Trg 1, 11000 Belgrade

SERBIA

Abstract: Voluminous data are generated from a variety of users and devices and are to be stored and processed

in powerful data center. As such, there is a strong demand for building a network infrastructure to gather

distributed and rapidly generated data and move them to data center for knowledge discovery. Big data has

received considerable attention, because it can mine new knowledge for economic growth and technical

innovation. Many research efforts have been directed to big data processing due to its high volume, velocity

and variety, referred to as 3V. This paper first describes challenges for big data together with basic Map Reduce

structure. Then it presents existing approaches for big data analytics including general architecture. The second

part establishes the relation between mobile cloud and big data and provides geo-dispersed big data application,

including big data in mobile cloud sensing. Some open research questions that need to be investigate conclude

this work.

Key-Words: Big data analytics, mobile cloud, spatial data, wireless sensing

1 Introduction

Composed of text, image, video, audio, mobile or

other forms of data collected from multiple data

sets, big data is growing in size and complexity in a

fast way. As a consequence, a huge volume of

multi-dimension of data within a very short time

period is created [1]. This is having a profound

impact society and social interaction, creating at the

same time opportunities for business. Big data needs

vision and dialog from various disciplines such as

engineers, computer scientists, statisticians,

sociologists, etc. With the rapid development on the

of Things (IoT), much more data is automatically

generated by millions of nodes with high mobility,

for example sensors carried by moving objects or

vehicles. To achieve relevant information, mobile

big data has to be carefully analyzed. It provides

opportunities to understand behaviors and

requirements of mobile users allowing the delivery

of decisions for real applications.

On the other hand, network operators, will also

benefit mobile big data. It should be noted that

mobile big data yields many challenges to the data

mining, mobile sensing and knowledge. The rapid

growth of data size created a need for multi-

disciplinary collaboration from industries and

academics to develop new methods that can

accommodate data networking, management,

computational and statistical sciences to the big data

analysis. A multidisciplinary approach to big data is

shown in Fig. 1.

Fig. 1 Multidisciplinary relations in Big data.

This paper is organized as follows. First, we present

challenges for big data followed by basic

MapReduce structure. Then, we briefly review

existing approach for big data analytics. The

corresponding general architecture is included, too.

Mobile cloud and big data together with providing

geo-dispersed applications are presented. Finally, by

big data in mobile cloud sensing are reported.

2 Challenges for big data

Big data is the next thing in computing. As this data

cannot be processed using traditional systems, it

poses numerous challenges to the researchers [2].

This raises several new challenges such as how to

classify big data for multiple data sets, how to

analyze big data for multiple data sets, how to

analyze big data for different forms of data, how to

visualize big data without loss of information. Also,

privacy is one of the important concerns with big

data. Many problems of big data, such as

measurement, representation, compression, analysis

are also faced. Big data applications introduce

WSEAS TRANSACTIONS on COMPUTER RESEARCH

DOI: 10.37394/232018.2022.10.3

Zoran Bojkovic, Dragorad Milovanovic

E-ISSN: 2415-1521

Volume 10, 2022

unprecedented challenges, while existing theories

and techniques have to be extended and upgraded to

serve the forthcoming real big data applications. Of

course, new tools for big data applications need to

be invented, for example tools for the processing of

super large graph or matrix and statistical studies in

low or extremely low probability events.

Technical challenges for big data include real-time

analysis requirement, new storage models, as well

as parallel/distributed operators for data with new n-

dimensional array-based data structures [2]. These

data need to be server-managed or cloud-managed,

compared and visualized with joint efforts from

hardware/software engineers, computer researchers

and statisticians. Another challenge and an urgent

need belongs to big data visualization. In order to

lower the cost of visualization generation, new data

base technologies have to be taken into account

together with emerging Web-based technologies.

2.1 Basic MapReduce structure

Proposed by Google, MapReduce is well known

tool for big data processing. The architecture is

shown in Fig. 2. The structure is composed of two

phases: map phase and reduce phase.

Fig. 2 Basic MapReduce processing structure.

Therefore, there exists mapper program function

and reduce program function. It can be seen that in

the first phase, the input data is split into blocks.

The function of the mappers is to scan the data

block in order to produce intermediate data.

Secondly, the reduce phase starts from a shuffle

sub-phase. It shuffles intermediate data, run by

reducers. They are moved to the corresponding

nodes, for reducing. As for the reducers, they

process the shuffled data generating the final results.

When dealing with more complicated problems,

multiple map and reduce processes can be used.

2.2 MapReduce implementations

There exists a number of practical MapReduce

implementations. Nowadays the open source

Hadoop is the most widely used. Generally, Hadoop

and other MapReduce-like tools run on dedicated

server cluster. Therefore, studies have mainly

focused on optimization using such infrastructures,

enabling the analysis of massive data. In order to be

in position to realize big data analytics, a set of

requirements have to be taken into consideration.

Here, expected results, budget, response time,

reliability, accuracy, are included [3]. To enable big

data analytics, necessary functions are grouped into

three layers: infrastructure layer, platform layer, and

software layer.

In the infrastructure layers a network of machines

will be set up for computing, storage, and

communications. A trend for big data analytics is to

use the cloud as the infrastructure because of

scalability and economics. A cloud consists of

usable and accessible virtualized resources. From

the point of view of cloud usage, a cloud is a public

cloud if the cloud service provider offers services to

the general public.

In the platform layer, Hadoop is the computing

paradigm. Currently, is is used in many major big

data analyzers including Internet companies Yahoo,

Amazon, Facebook, Twitter and many others data

analytics companies [3]. As a Java implementation

of MapReduce, Hadoop supports large-scale data

analytics applications. However, MapReduce does

not fit iterative algorithms that need to handle the

dependency between data over iterations. Spark [4],

as in-memory computing framework is designed to

overcome Hadoop’s shortages in iterative

operations. It introduces a real-only data structure-

resilient distributed data sets. During the whole

iterative process, intermediate results can be cached

in memory. GraphLab has evolved to introduce a

new approach to distributed graph placement. It

should be noted that each platform may target

different applications.

The big data analyzer decides which of algorithms

to use in the first step of data analysis. However,

even for a given algorithm, there exist different

implementation methods with different languages

and on different analytics platform. Thus for a big

data analyzer in order to have reliable and accurate

data, it is necessary for a big data analyzer to

understand the details of algorithms.

2.3 General architecture of analytics

Block scheme of general architecture consists of

multi-source big data collecting, distributed big data

storing and finally intra/inter big data processing.

Big data collection is characterized by high volume,

high velocity and high variety (3V). General

architecture of big data analytics is shown in Fig. 3.

WSEAS TRANSACTIONS on COMPUTER RESEARCH

DOI: 10.37394/232018.2022.10.3

Zoran Bojkovic, Dragorad Milovanovic

E-ISSN: 2415-1521

Volume 10, 2022

Fig. 3 General architecture of Big data analytics.

The challenges are:

• how to efficiently store and organize high-volume

data,

• how to quickly process streaming and near-real-time

data, and

• how to accurately analyze structured and unstructured

data in order to maximize the value of big data.

Since big data collecting is pervasive, it is not

efficient to move high volumes of collected data

around for centralized storing. Thus, distributed big

data is proposed [5]. It means that big data will be

stored and organized in their original location. Big

data should be processed in parallel such that new

knowledge and innovation can be mined in a

reasonable amount of time. Big data processing is

divided in two groups: intra processing, and inter

processing.

In inter big data processing, all data belong to the

same organization. On the other side, if big data are

part of different organizations, it will be inter big

data processing. It should be emphasized that inter

big data processing will be more challenging

because big data sharing will first be executed

before processing. Also, many new security and

privacy issues will arise during the duration of big

data sharing.

3 Mobile cloud and Big data

3.1 Spatial Big data application

Mobile services (Google Maps, Navigation service)

are big data applications. Because of the big data set

and the data update rate is fast, benefits and

convenience are provided [6]. Thus mobile cloud is

used for supporting geo-dispersed big data

applications. Framework based on MapReduce for

simple and complex operations is proposed to

provide better support and for geo-dispersed big

data [7]. Block scheme of mobile cloud architecture

is shown in Fig. 4. It consists of several cloudlets

and a central cloud. Advanced MapReduce

framework (AMF) employs cooperative processing

in the mobile cloud and MapReduce to process geo-

dispersed big data.

AMF extracts the required multiple inputs from geo-

dispersed big data in parallel. After that, required

extracted multiple inputs from different cloud nodes

are aggregated and processed after performing

complex operations. Then, creating the final results,

they are sent to the user. Different data aggregation

schemes are used for different application

requirements.

Fig. 4 Mobile cloud architecture for providing geo-

dispersed Big data applications.

As for the mobile cloud architecture, it consists of

several cloudlets and a central cloud which stores

part of the big data, while the cloudlets have large

amounts of thresh data as a part of big data which

are uploaded to the central cloud periodically to

update the data set. The central cloud has got

sufficient computational resources to process all of

the big data. Note, that we have large amounts of

fresh data from cloudlets to the central clod, while

communication delays to mobile users is short.

Thus, a cloudlet should be utilized to assist in

performing operations on geo-dispersed big data and

reduce response time.

3.2 Mobile wireless Big data application

Mobile wireless big data applications together with

advanced technologies in mobile networking, are

emerging today. For example, Nike+ provides

important services for users with wireless connected

sensing devices and smartphones [8]. Company

Nike enhances its products, such as shoes, with built

in sensors that continuously track the user’s

movement in the period their workouts. The

collected data provide users with their instant

information such as their pace, GPS position,

distance moved. Of course, Nike+ apps installed in

the users’ smartphone collect data from the sensors

using wireless connections [9]. Nike+ has become a

big data platform that collects, stores and processes

data generated from a huge number of users.

3.3 Big data in mobile cloud sensing

Mobile cloud sensing combines mobile sensing,

cloud computing and big data to obtain process and

predict mobile sensor data. One of the bottleneck for

data centric sensing such as video surveillance,

image sharing, etc is network bandwidth. As more

Multi-source

Big data

collecting (3V)

Distributed

Big data

storing

Intra/Inter

Big data

processing

WSEAS TRANSACTIONS on COMPUTER RESEARCH

DOI: 10.37394/232018.2022.10.3

Zoran Bojkovic, Dragorad Milovanovic

E-ISSN: 2415-1521

Volume 10, 2022

sensors are integrated into mobile devices, the

generated data by them cloud exceed the network

capacity. A network infrastructure has to be applied

to support the data capacity as well as connection

request. Mobile cloud sensing brings together cloud

computing services and mobile sensing features.

Also, mobile cloud sensing contributes mobile

devices access to resources of cloud computing. On

the other hand, cloud computing infrastructure is

enabled to obtain real world data from mobile

sensing devices. The mobile cloud sensing

architecture with the building components is shown

in Fig. 5 [9].

Fig. 5 Essential building blocks of a mobile cloud

sensing system.

The data sensing unit consists of physical sensing

probes and social sensing probes. Physical sensing

probes include smartphones, tablets, and wearable

devices (smartwatch, smart glasses, smartbracelets),

while social sensing probes are posts on social

networks (Facebook, Twitter). Physical sensing unit

are raw format such as accelerometer data, ambient

light strength, pulse rate, digital image, audio data.

Other part of sensing data is from the social sensing

probes. Data preprocessing unit examines the raw

data from sensors and social networks, extracting

corresponding features encrypted and compressed to

minimize the data bandwidth and protect the data.

Network management unit can be optimized to

make sensing data throughput larger as well as to

make the integration of 5G network interfaces easier

and faster. Cloud platforms have sufficient storage

for sensing data. Data from sensing sources

converge here, while features are fed to the specific

machine learning tasks to be interpreted. Results are

stored on the cloud for accessing. Data

authentication and service interfaces interact

directly with end users.

4 Conclusion

Big data presents a new era of information

generation and processing. However, research work

of big data processing in the mobile cloud remains

in its infancy in spite of the fact that cloud

computing is a popular infrastructure that has the

resources for big data processing. On the other hand

mobile cloud computing is becoming important part

of big data applications. In connection with this

mobile opportunistic networks can function as a

complementary alternative infrastructures for

supporting the emerging big data applications. As a

technology that makes an intelligent and smart

world possible through mobile devices mobile cloud

sensing is changing the way we live. New data

systems and technologies are required to handle

mobile big data in a highly scalable, cost effective

and fault tolerant fashion.

References:

[1] J.Zhang and M.L.Huang, Data behaviors model

for Big data signal analytics, International

Journal of Big data intelligence, vol. 3, no.1,

pp. 1-17, 2016.

[2] H.Fang, et al., A survey of Big data research,

IEEE Network, vol.29, no.5, pp.6-9, Sept/Oct.

2015.

[3] H.Li and Sh.Meng, Big provision: a provision

framework Big data analytics, IEEE Network,

vol.29, no.5, pp.50-56, Sept/Oct. 2015.

[4] Spark engine for large-scale data processing

(available online http://spark-project.org)

[5] R.Lu, et al., Toward efficient and privacy-

preserving computing in Big data era, IEEE

Network, vol.28, no.4, pp.46-59, July/Aug.

2014.

[6] S.Shekhar, et al., Spatial Big data challenges

intersecting mobility and cloud computing, in

Proc. ACM International Workshop Data

engineering for Wireless and mobile access,

Scottsdale, AZ, pp.1-6, 2012.

[7] H.Zhang, et al., Processing Geo-dispersed Big

data in an advanced Map Reduce framework,

IEEE Network, vol.29, no.5, pp.24-30,

Sept/Oct. 2015.

[8] X.Yi, et al., Building a network highway for

Big data: architecture and challenges, IEEE

Network, vol.28, no.4, pp.5-13, July/Aug. 2014.

[9] G.Han, Sh.Liang, H.Zhang, Mobile cloud

sensing, Big data and 5G networks make an

intelligent and smart world, IEEE Network,

vol.29, no.2, pp.40-45, March/April 2015.

Physical probes Social probes

Data sensing unit

Creative Commons Attribution License 4.0

(Attribution 4.0 International, CC BY 4.0)

This article is published under the terms of the Creative

Commons Attribution License 4.0

https://creativecommons.org/licenses/by/4.0/deed.en_US

WSEAS TRANSACTIONS on COMPUTER RESEARCH

DOI: 10.37394/232018.2022.10.3

Zoran Bojkovic, Dragorad Milovanovic

E-ISSN: 2415-1521

Volume 10, 2022