An Evaluation of Smartphone

Resources Used by Web

Advertisements
by

Abdurhman Albasir

A thesis
presented to the University of Waterloo
in fulfillment of the
thesis requirement for the degree of
Master of Applied Science
in
Electrical and Computer Engineering

Waterloo, Ontario, Canada, 2013

c Abdurhman Albasir 2013
I hereby declare that I am the sole author of this thesis. This is a true copy of the thesis,
including any required final revisions, as accepted by my examiners.

I understand that my thesis may be made electronically available to the public.

ii
 Abstract

With the rapid advancement of mobile devices, people have become more attached to
them than ever. This rapid growth combined with millions of applications (apps) make
smartphones a favourite means of communication among users. In general, the available
contents on smartphones, apps and the web, come into two versions: (i) free contents that
are monetized via advertisements (ads), and (ii) paid ones that are monetized by user
subscription fees. However, the resources (energy, bandwidth, processing power) on-board
are limited, and the existence of ads in either websites or free apps can adversely impact
these resources. These issues brought the need for good understanding of the mobile
advertising eco-system and how such limited resources can be efficiently used.
This thesis focuses on mobile web browsing. Surfing web-pages on smatphones is one of
the most commonly used task among smartphone users. However, web-page complexity is
increasing, especially when designed for desktop computers. On one hand, the existence of
ads in web-pages is essential for publishers’ monetization strategy. On the other hand, their
existence in webpages leads to even higher complexity of the webpages. This complexity
in the smartphone environment, where the battery and bandwidth resources are limited,
is reflected in longer loading time, more energy consumed, and more bytes transferred.
With this view, quantifying the energy consumption due to web ads in smartphones is
essential for publishers to optimize their webpages, and for system designers to develop an
energy-aware applications (browsers) and protocols. Apart from their energy impact, ads
consume network bandwidth as well. Therefore, quantifying the bandwidth consumption
due to downloading web ads is crucial to creating more energy and bandwidth aware
applications. This thesis first classifies web content into: (i) core information, and (ii)
forced “unwanted” information, namely ads. Then, describes an approach that enables
the separation of web content in a number of a websites. Having done so, the energy cost
due to downloading, rendering, and displaying web ads over Wi-Fi and 3G networks is
evaluated. That is, how much energy web ads contribute to the total consumed energy
when a user accesses the web. Furthermore, the bandwidth consumed by web ads in a
number of well-known websites is also evaluated.
Motivated by our findings about ads’ impact on the energy and bandwidth, the thesis
proposes and implements a novel web-browsing technique that adapts the webpages deliv-
ered to smartphones, based on a smartphone’s current battery level and the network type.
Webpages are adapted by controlling the amount of ads to be displayed. Validation tests
confirm that the system, in some cases, can extend smartphone battery life by up to ∼
30% and save wireless bandwidth up to ∼ 44%.

iii
 Acknowledgements

All praise is to Allah for giving me the ability and knowledge to accomplish this thesis.
Several people deserve sincere recognition. I am grateful to my advisor, Prof. Kshi-
rasagar Naik, for his guidance, support, patience, and encouragement. I would like also to
extend my appreciation and thanks to my thesis committee members, Prof. Liang-Liang
Xie and Prof. Pin-Han Ho for sparing their time in reviewing my thesis.
I would like also to acknowledge the Ministry of Higher Education of Libya for finan-
cially supporting this research work.
Thanks go to my friends for their prayers and support. I am very fortunate to have so
many exceptional and genuine people in my life. I thank my friends and colleagues, Tarek
Alwazini, Maazen AlSabaan, Tareq Abdunabi, Abdulhakim Abogharaf, Majid Altamimi,
Mahdi Alghazali, Elfetori Ibrahim and Mohamed Elmassad for all our fruitful discussions
and for their friendship.
Last and by far not least, I am indebted to my parents, brothers, and sisters for their
continuous support, patience, and prayers. I owe a great deal of gratitude to my wife,
Khawla, and my daughter, Jori, for their continuous understanding, unlimited encourage-
ment, and unending love during the years of my study.

iv
 Dedication

to my beloved dad, Prof. Ali Elbasir

v
Table of Contents

List of Tables viii

List of Figures ix

1 Introduction 1
1.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Problem Statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3 Solution Strategy and Contributions . . . . . . . . . . . . . . . . . . . . . 5
1.4 Thesis Organization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

2 Background and Literature Review 8

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.2 Challenges in Handheld Development . . . . . . . . . . . . . . . . . . . . . 8
2.3 Mobile advertising . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.3.1 Mobile online advertising ecosystem . . . . . . . . . . . . . . . . . . 12
2.3.2 Advertisement format . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.3.3 Ad selection mechanisms . . . . . . . . . . . . . . . . . . . . . . . . 17
2.3.4 Ad pricing models . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.3.5 Advertisements in Smart Devices . . . . . . . . . . . . . . . . . . . 19
2.4 Literature Review of Existing Research . . . . . . . . . . . . . . . . . . . . 20
2.4.1 Energy Cost of Advertising in hand-held devices . . . . . . . . . . . 20
2.4.2 Bandwidth Cost of Advertising in hand-held devices . . . . . . . . . 24

vi
3 Measurements and Analysis 29
3.1 Performance Metrics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.2 Testing Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.2.1 The Available Ad-blocking Techniques . . . . . . . . . . . . . . . . 30
3.2.2 Our Ad-blocking Strategy . . . . . . . . . . . . . . . . . . . . . . . 31
3.2.3 Energy Test Set-up . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.2.4 Bandwidth and Network Test Set-up . . . . . . . . . . . . . . . . . 34
3.3 Energy and Bandwidth Measurements of Web Advertisements in Smartphones 35
3.3.1 Energy Measurements Test Cases . . . . . . . . . . . . . . . . . . . 35
3.3.2 Bandwidth Measurements Test Cases . . . . . . . . . . . . . . . . . 45

4 Smart Mobile Browsing Strategy 47

4.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
4.2 Design requirements and objectives of SWB . . . . . . . . . . . . . . . . . 48
4.2.1 System Architecture . . . . . . . . . . . . . . . . . . . . . . . . . . 49
4.3 System Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
4.3.1 Client Side Configuration . . . . . . . . . . . . . . . . . . . . . . . 52
4.3.2 Server Side Configuration . . . . . . . . . . . . . . . . . . . . . . . 54
4.3.3 System Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
4.4 Validation and Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
4.4.1 Energy and Bandwidth Results . . . . . . . . . . . . . . . . . . . . 56

5 Conclusion 63

References 64

vii
List of Tables

2.1 Summary of the work done on in-apps advertising energy cost . . . . . . . 21

2.2 Summary of the work done on in-apps ad bandwidth cost . . . . . . . . . . 25

3.1 Summary the traffic statistics while web browsing with and without ads . . 39
3.2 The cost of ads over a Wi-Fi interface . . . . . . . . . . . . . . . . . . . . . 42
3.3 The cost of ads over a 3G interface . . . . . . . . . . . . . . . . . . . . . . 44
3.4 Traffic breakdown per web-page . . . . . . . . . . . . . . . . . . . . . . . . 45

4.1 The functions of client’s components . . . . . . . . . . . . . . . . . . . . . 50

4.2 Summary of the estimated battery life under three different cases . . . . . 60

viii
List of Figures

1.1 Sample website showing the core information and some ads (marked in red) 4
1.2 Sequence diagram of downloading webpage . . . . . . . . . . . . . . . . . . 5

2.1 Mobile online advertising ecosystem. . . . . . . . . . . . . . . . . . . . . . 13

2.2 Sequence diagram of Mobile web browsing (DNS: Domain Name System.) . 15
2.3 Example of in-App Ad library as it appears in the source code file [23] . . 15
2.4 Examples of text ad formats . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.5 Examples of Display ads format. . . . . . . . . . . . . . . . . . . . . . . . . 18

3.1 Ad blocking strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32

3.2 The energy measurement set-up. . . . . . . . . . . . . . . . . . . . . . . . . 33
3.3 The battery bypass connection . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.4 The bandwidth measurement set-up . . . . . . . . . . . . . . . . . . . . . . 35
3.5 The real-time energy footprint of a number of websites over a Wi-Fi connection 36
3.6 A sample of Energy-Bandwidth Mapping in a Webpage . . . . . . . . . . . 38
3.7 The real-time energy footprint of a number of websites over a Wi-Fi con-
nection with and without ads . . . . . . . . . . . . . . . . . . . . . . . . . 40
3.8 Energy measurement of our browser . . . . . . . . . . . . . . . . . . . . . . 41
3.9 The energy consumption of different websites over 3G and Wi-Fi networks 43
3.10 Traffic statistics while web browsing with and without ads. . . . . . . . . . 46

4.1 System Architecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49

ix
4.2 The Graphical User Interface of the Smart Server and Smart Browser . . . 53
4.3 HTTP header modifications . . . . . . . . . . . . . . . . . . . . . . . . . . 55
4.4 Comparison of the energy consumption over a Wi-Fi connection . . . . . . 57
4.5 Comparison of the energy consumption over a 3G connection . . . . . . . . 58
4.6 Energy consumption comparison between SWB and normal browsing . . . 59
4.7 Bandwidth measurements . . . . . . . . . . . . . . . . . . . . . . . . . . . 61

x
Chapter 1

Introduction

The past decade has witnessed a huge growth in smartphones’ popularity. People all over
the world now tend to have smartphones and the number of mobile phones in the market
has been estimated to be almost equal to the number of people worldwide in 2013 [6].
A significant percentage of these mobile subscribers already have a smartphone or are
planning to get one. A report published by ComScore in September 2012 stated that
over 115 million people in the United States owned smartphones out of 234 million total
subscribers [16]. This statistic shows a penetration of close to 50%. Google Android is
currently the top ranked with 52.2%, and Apple comes after with 34.3%. Such growth and
popularity is driven by the vast variety of applications (apps) now available. This rapid
growth of smartphones combined with millions of apps make them a favourite means of
communication among users, but their functionalities and development are constrained by
a number of challenges (Battery life, Bandwidth, and Security and Privacy.)
The volume of wireless traffic continues to grow exponentially. Web browsing and
network related applications (NRAs) rely entirely on the Internet for their operation. They
are considered the main contributors to the ever-growing wireless traffic. Since the traffic
generated by smartphones forms a new addition to wireless networks, the demand for
studies that investigate the nature of smartphone traffic has emerged [21] [40]. These
research studies are answering interesting questions, such as (i) how much traffic a device
generates per day; (ii) which app contributes the most in the total smartphone-generated
traffic; and (iii) which app utilises the most network resources. The results can be used
by mobile operators to improve their wireless quality of services (QoS) [14] [32] [35]. Some
studies have broken down the traffic generated by smartphones to the app level, to explore
how much traffic is generated by each app. However, more questions remain unanswered.
For example, in mobile web browsing, what is the bandwidth cost of delivering certain

1
information to an end user? More specifically, how many bytes are required for downloading
informations that is not of user interest? Thus, more investigation is needed, and more
solutions that utilize this scarce resource efficiently are of great importance.
Smartphones use Lithium-ion batteries due to their high energy densities. However,
there is no tangible evidence that in the near future more energy can be packed into the
small space available on today’s modern hand-held devices [45]. In addition, more power-
ful processors, high resolution displays, huge numbers of applications, and the advanced
wireless technologies (LTE, WiMAX) are employed in today’s smartphones and tablets.
Such technologies and components drain energy at very high rates and are considered as
battery hungry. For instance, the authors of [65] found that smartphone components,
namely, CPU, Wi-Fi, LCD, GPS, and 3G links, consume up to, 28.6, 24.6, 85.9, 33.6, and
36.5 Joules, respectively. Moreover, the energy cost of sending a simple SMS message and
placing a call through 3G is more energy expensive than using GSM, according to [45].
Although much research has focused on their battery life and the energy consumption in
smartphones, solutions that target the hardware design as well as software have yet to
emerge. Since software solutions are easier to implement and flexible to customize, they
are more desirable and applicable to developers.

1.1 Motivation
The subject of this thesis was firstly motivated by two facts:

1. The high growing popularity of mobile web browsing. Recent statistics show that
users prefer to use web browsers for digital news delivery. According to a survey
conducted in U.S in 2012, 61% of smartphone users (and 60% of tablets users) use
mainly browsers for reading news [36]. Thus, browsers are considered to be one of
the most used smartphone apps.

2. The increasing complexity of webpages. A study done by [63] shows that over half of
the pages are not optimized for mobile browsers. Although many websites have mo-
bile versions of their webpages, people tend to view the full/desktop versions for the
richer offered content. And despite the fact that some well known website companies
have already created mobile versions of their websites, their sub-pages are actually
desktop versions (full versions). These webpages are getting more complex [10, 68],
in other words, more computationally intensive, and the existence of ads in webpages
leads to even higher complexity. This complexity in smartphone environments where

2
 the resources (e.g., battery and bandwidth) are limited is reflected in longer loading
times, more energy consumed, and more bytes transferred.

Therefore with these facts in mind, we were first motivated to study the energy and band-
width impacts of ads in webpages. Then, after studying the impact of web ads on smart-
phone battery life and bandwidth, we were motivated by the measurements and the findings
obtained in section 3.3 to design a system that is energy and bandwidth efficient. Be-
cause ads can drain considerable energy during mobile web browsing (in some web-pages
it reaches 18% of the total energy) and because bandwidth overhead caused by download-
ing ads in some webpages reaches 50%, we propose a novel mobile browsing technique
Smart Mobile Web Browsing (SWB) that is a function of the smartphone resource and the
number of ads allowed to be displayed on webpages. SWB mainly targets (i) extending
smartphone battery life; and (ii) preserving the bandwidth needed to download webpages.
As mentioned in the previous section, having ads displayed in webpages is essential for
having free web content; therefore, our approach aims to balance the satisfaction of end
users as well as the webpage owners (publishers.)

1.2 Problem Statement

Web browsing is becoming essential for everyday life, especially with the ever-increasing
popularity of smartphones. The web content delivered to end users has witnessed drastic
changes. Webpages used to be simple “traditional” static pages compromised of only text
and some images. However, modern webpages are dynamic and media-rich. The owners
of these pages (publishers), moreover, rely heavily on ads as a source of revenue. These
ads are very media-rich and can consume a lot of resources, like battery and bandwidth.
Entirely eliminating ads from webpages is not practical, in view of the big role they play
in the web eco-system and in the process of having free web content. These factors have
led to the demand for mobile browsing solutions that adapt pages in a way that is energy
and bandwidth efficient.
When a webpage is loaded on a user’s screen, what actually is downloaded are the
core information that a user is interested in (such as news, emails, stocks, etc.), and extra
enforced “unwanted” information that comes up in the form of web advertisements (ads).
Figure. 1.1 shows an example of the type of content that is downloaded and displayed on
a user’s screen. In a smartphone environment where resources are limited, displaying ads
on the screen is not an inexpensive task from the energy and bandwidth perspectives. To
be displayed, an ad needs to be fetched, rendered, and finally displayed. Each of these
actions requires specific resource: network access (radio interface and bandwidth), certain

3
Figure 1.1: Sample website showing the core information and some ads (marked in red)

CPU computations, and graphics. Figure. 1.2 illustrates the extra needed steps to display
the ads. These extra requests are reflected as an increase of battery consumption.
The main problem explored in this thesis is as follows. The complexity of webpages is
increasing, especially if the webpages are designed for desktop computers. The existence
of advertisements (ads) in webpage leads to even higher complexity. This complexity in
a smartphone environment, where resources are limited (e.g., battery and bandwidth)
is reflected in longer loading time, more energy consumed, and more bytes transferred.
Therefore, evaluating resources used by web advertising is very important to smartphone
users as well as to web designers. Thus, this thesis focuses on quantifying the energy and
bandwidth cost due to web ads in smartphones and on developing a way to limit this
cost. Based on the measurements and analysis performed, the question of how smartphone
battery life can be extended by controlling the number of ads to be displayed has emerged,
as well as how the bandwidth needed to download webpages can minimized.
None of the studies surveyed in our literature review focus specifically on measuring the
resources used in mobile devices by advertisements’ overhead during mobile web browsing.
Most address only the issue in mobile applications, and some study the cost of ads on PCs
or laptops but not on mobile devices. Therefore and to the best of our knowledge, this

4
 Figure 1.2: Sequence diagram of downloading webpage

study is the first that analyses mobile resources usage due to displaying ads on webpages
based on real measurements. Also, although a number of attempts have tried to enhance
mobile web browsing, no existing studies explore how web contents can be adapted just by
controlling the number of ads allowed in webpages, so as to maximize the battery life of
smartphone and minimize the required bandwidth.

1.3 Solution Strategy and Contributions

The strategies used to maintain the goals of this thesis are briefly explained in the following
points. More details about these solutions strategies are provided in Chapters 3 and 4.

5
 1. Measuring Ads’ Energy and Bandwidth Cost: To measure exactly the energy and
bandwidth cost of web ads, an approach to separate what we call core information
and overhead “unwanted” information (ads) was needed. We decided to download a
number of websites and have our own web server. These allowed us to enable and
disable the delivery of ads as needed. Then, a number of test cases were conducted
to measure the energy and bandwidth costs of web ads. The details of this strategy
can be found in Chapter 3.
2. Designing Energy and Bandwidth-aware Web Browsing: Developing a mobile brows-
ing technique, called Smart Mobile Web Browsing (SWB), that is a function of smart-
phone resources and the number of ads allowed to be displayed on webpages required
an application-level modification. We modified the HTTP protocol so as to allow
embedding a smartphone current resources (battery level and network type) in the
HTTP header of the request made by the browser. On the server side, a certain
policy is applied, resulting in an adapted webpage being sent back to the client.
The adaptation of the web content comes in how many ads can be displayed on the
webpage. The details of this strategy can be found in Chapter. 4.

This work makes the following contributions:

• quantifies the ad traffic generated during mobile web browsing (bytes/mo).

• estimates the additional cost of downloading ads to a user’s monthly bill ($/mo).
• quantifies the energy consumed to download and display ads on webpages while
mobile web browsing.
• investigates the impact of using different networks (3G and Wi-Fi) on the energy
consumed to download and display ads on webpages.
• proposes a smart mobile browsing technique that aims to extend smartphone battery
life and save the wireless bandwidth by controlling the amount of ads to be displayed.
This solution attempt to keep the ads and at the same time alleviate its impact on
smartphone limited resources.

1.4 Thesis Organization

The rest of the thesis is organized as follows. Related research and background are pre-
sented in Chapter 2; the main challenges of handheld development are discussed. The

6
nature of mobile web advertising is explored and relevant existing work is reported. Chap-
ter 3 describes the testing methodology used and the experiments carried out to achieve
the first two contributions of this thesis. It also presents and discusses the measurements
of energy and bandwidth used for ads. Chapter 4 contains the proposed smart mobile
web browsing framework and the validation results. Finally, Chapter 5 documents some
conclusions of this thesis and recommendations for future work.

7
Chapter 2

Background and Literature Review

2.1 Introduction
This chapter introduces the challenges being faced in the development of modern handheld
devices, smartphones and tablets, is given. An elaboration on the limited resources avail-
able on them and their implications on the user experience. The thesis framework requires
a good understanding of how online advertising works and how such advertisements are
embedded in both web pages and applications. Therefore, an overview of web advertising
systems with a taxonomy of mobile advertising in handhelds is provided.
The ultimate goal of this thesis is to highlight the cost of web advertising in terms
of bandwidth and energy. Several related studies are discussed in the literature review
section.

2.2 Challenges in Handheld Development

Services offered by smartphones range from communication, entertainment, social network,
to other Internet-based activities. As stated in the introduction, the growth of smartphone-
use is ever increasing, primarily because today’s smartphones have: powerful processors and
graphics processing units, significant amounts of flash memory, high-resolution screens with
multi-touch capability, and diverse applications, all thanks to the huge recent developments
in both software and hardware. In spite of the fact that those diverse apps and services
are available to everyone effortlessly, having such services and apps is faced by many

8
problems and challenges in development and design. The following subsections discuss
some important issues in smartphone development.
Services offered by smartphones range from voice and Internet usage to other enter-
tainment options. As mentioned in the introduction, the growth of smartphone popularity
has sky-rocketed. A report published by ComScore in September 2012 stated that over
115 million people in the United States owned a smartphones out of 234 million total sub-
scribers [16]. This fact shows a penetration of over 50%. The report also remarks that
Google Android is currently in the top ranked with 52.2%, and Apple comes next with
34.3%. The trend worldwide is almost the same, according to mobiThinking [38]. The
International Telecommunication Union ITU [4], states that there were 6 billion mobile
subscriptions at the end of 2011. This figure was to reach 6.5 billion by the end of 2012,
and by the end of 2016, it is expected to reach 8 billion.
The primary reasons behind the increased popularity of smartphones are the power-
ful processors and graphics processing units, significant amounts of flash memory, high-
resolution screens with multi-touch capability and diverse applications, all driven by the
rapid evolution in both software and hardware worlds. It is implied that such advanced
services and technologies could not have been accomplished without advanced research.
Overcoming many challenges has led to the success of smartphones and the great function-
alities offered. This section mentions some of the most important problems that have been
highlighted by researchers.

• Battery life
In smartphones, Lithium-ion batteries are used due to their high energy densities.
However, there is no tangible evidence that in the near future more energy can be
packed into the small space available on today’s modern hand-held devices [45]. In
addition, more powerful processors, high resolution displays, huge number of appli-
cations, and advanced wireless technologies (LTE, WiMAX) are employed in today’s
smartphones and tablets. Such technologies and components drain energy at a very
high rates and are considered to be battery hungry. For instance, the authors in
[65] found that smartphone’s components, namely, CPU, Wi-Fi, LCD, GPS, and 3G
links, consume up to, 28.6, 24.6, 85.9, 33.6, and 36.5 Joules, respectively. Moreover,
the energy cost of sending a simple SMS message and placing a call through 3G is
more energy expensive than using GSM, according to [45].
Much research efforts have focused on the battery life and the energy consumption in
smartphones. Those efforts range from: testing [7],[41] and measuring certain activ-
ities, estimating and modelling the energy consumption of smartphone components
(CPU, radio interfaces, GPS, and camera, etc.), measuring the energy cost of some

9
 popular apps, to studying the impact of ads on the battery life. Solutions that could
alleviate the battery life problem were proposed, as shown in section 2.4.1. Such solu-
tions have targeted the hardware design as well as software. Since software solutions
are easier to implement and flexible to customize, they are seen as more desirable
and applicable to developers. Some software solutions include: using the cloud to
minimize the energy cost for apps that use the Internet, developing optimization
tools, and designing apps with energy efficiency in mind, as shown in section 2.4.1.

• Bandwidth
A very recent addition to Internet traffic is the data traffic generated by smartphones
and tablets. Along with increased number of smartphone subscribers, smartphones
Internet traffic is increasing exponentially. Nokia Siemens published a report stating
that cellular traffic is forecast to grow 10 times faster than fixed Internet traffic [64].
Ericsson also confirms that fact and shows that most of the cellular traffic is gen-
erated from smartphones, and network traffic is actually surpassing voice traffic [3].
However, due to the growing traffic generated by smartphones and tablets, wireless
networks are edging to near capacity. As a result, serious pricing and performance
issues may come to picture [9]. Thus, more research is needed to know better the
characteristics of the traffic generated from hand-held devices [14][59].
In fact, the bandwidth issue in the wireless environment can be broken down into two
main categories: (i) from a network operator’s perspective; with the smart devices
available today and the vast number of apps that essentially function via Internet
connection, and the fact that radio resources are scarce, many performance issues
have emerged. Such an issue is considered to be expensive to tackle. For instance,
Vasona Networks, in 2013, has put up a budget of $12 million to help operators
optimize their network and solve the problems of mobile data congestion [34]. (ii)
From an end user’s perspective, given the facts mentioned in the first point, mobile
operators tend to offer metered pricing plans that are more affordable to customers
than the unlimited ones. Consequently, another issue arises and impacts user’s expe-
rience: bandwidth hungry apps. According to [30] and [66], such apps generate extra
traffic that is not needed for the actual operation of these apps. The extra traffic
“overhead” refers to the advertisements and analytics traffic that was found to be a
significant contributor in monthly data usage and in a user’s monthly bills.

• Security
The development of both devices and services have been market-driven. Most of
them focus on new features but neglect security [19]. A recent study conducted
by security software provider McAfee found that the amount of malicious software,

10
 also known as malware, targeting Android, had jumped by 76% since the previous
quarter. At the same time, Android had surpassed Symbian as the most-attacked
mobile platform. Security issues due to ads are beyond the scope of this thesis.
Smartphone platforms, in some cases, need to give access to various resources, such as
GPS, Network interfaces, and system tools, thus preventing the phone from sleeping.
Applications that require many privileged-to-smartphone resources can create a high
chance for security and privacy hazards. Such apps can perform some unauthorised
actions by users or send some private information to remote servers without the
awareness of users.

In a summary, the three above mentioned issues are very serious challenges to smart-
phone development. They require more research efforts to: (i) find out the main contribu-
tors of such issues (ii) come up with techniques and solutions that alleviate their severity.
It is important to mention that the involvement of third-party advertisements, as a mon-
etization technique, in apps and in other services offered on smartphones increases the
seriousness of these issues. A lot of studies have found that ads contribute significantly to
the severity of such issues as will be shown in detail in sections 2.4.1 and 2.4.2.

2.3 Mobile advertising

The field of advertising has evolved rapidly with today’s advanced technologies. The variety
of telecommunication network technologies and the increasing penetration rate of end-user
devices createan ideal environment to reach billions of people effortlessly. Newspapers,
magazines, TVs, radios, and street posters were the only means of advertising about 10
years back. With the advances in technology today, consumers are not only receiving ads
wherever they are, but rather they can interact with it immediately. Moreover, ads are no
longer distributed in a random way, they are selected and delivered in a smart way, using
a user’s location for example, so that consumers are exposed only to ads that are relevant
to their interests. Therefore, mobile advertising with such potentials is one of the largest
sources of revenue. According to [27], the mobile advertising market is forecast to grow
from $3.4 billion in 2010 to $22 billion in 2016, with a 37% annual growth rate.
To further clarify the concept of mobile advertising, the Mobile Marketing Association
in [8] defines it as “any paid form of marketing, advertising or sales promotion activity
aimed at consumers and conducted over a mobile channel.” The definition includes all
forms of mobile advertising, in other words, any advertising mechanism that delivers ads
to consumers over a mobile channel [22]. To narrow the field of mobile advertising, this

11
thesis focuses on advertising using online resources on smartphones and tablets, which is
one of the most effective ways of advertising. Guha et al. in [24] consider it as a key
economic force in the Internet economy. Mobile online advertising can be defined as any
form of delivering ad content over the Internet to a mobile hand-set, e.g. smartphones,
tablets.
It is important to mention that we are not particularly interested in the philosophy of
the mobile advertising market. Rather, we want to give a general background of how those
systems work and what entities contribute to the make-up of the system from a technical
point of view. This chapter provides a brief background on the architecture of mobile
advertising; however [23][33][62][53][13] provide more details on the architecture of mobile
advertising eco-systems and the related measures of effectiveness.

2.3.1 Mobile online advertising ecosystem

The mobile advertising ecosystem is composed of four main entities; Users, Publishers,
Ad networks, and Advertisers [53][13][17][31]. These actors are directly related to today’s
modern technologies. Therefore, the ecosystem of mobile advertising has grown exponen-
tially, as illustrated next. This section begins with explaining each entity and the role that
each one takes, and ends with answering the question of how the system actually works.

• Advertisers are companies that want to persuade people (users) to buy their prod-
ucts. They advertise their products by placing their ads on applications or websites
through an advertising network or broker. Advertisers pay both publishers and ad
networks based on specific pricing models [52], [13]. Gucci, and Toyota are example
of advertisers.

• Ad networks, sometimes called ad brokers, are organizations who link advertisers

with publishers as shown in Fig. 2.1. Their main function is to select ads that
are then put on publisher’s application or web-page. Based on either certain data
gathered about the end user or the content that a publisher provides, appropriate
ads are selected and delivered to the end user. Examples of advertising networks are
AdMob, AdSense, and iAds.

• Publishers can be either the owner of a web-page or an application. In the case of

web-pages, the owners of such websites allow ads to be displayed on their web-pages
in order to make profit and keep providing end users with free content. Application’s
developers who offer free versions of their apps allow displaying ads on them so
that they can keep those apps available to end users free of charge. Both cases are

12
 Figure 2.1: Mobile online advertising ecosystem.

basically interested in monetizing their contents and increasing user loyalty. For
large publishers, a common practice is that they themselves take the role of ad
networks. They usually have their own departments to take care of the work done by
ad networks, i.e., making the negotiation with the advertisers regarding the allowed
ad contents and the pricing model they offer as well as performing the ad matching
process based on the end user interest (ad selection).

• Users the last players in this system are the owners of the smart devices. They are
interested in either using certain free apps or the contents of certain websites. Users
interact with ads that are displayed on their screens. Consequently, they complete
the circle that keeps the mobile ad ecosystem going by clicking on the ads they have
an interest in.

The mobile advertising systems works as follows: all the actors above contribute to the
generation of the revenues that keep the system going and all the players satisfied. As
illustrated in Fig. 2.1, there are two scenarios to implement mobile online advertising:

• Mobile Web Advertising

13
 In this scenario, the needed requests and messages to complete the task of download-
ing and displaying ads are depicted in the sequence diagram in Fig. 2.2. It starts
with an end user fetching a certain website (publisher, e.g. www.mydictionary.com);
the publisher’s server then sends an HTML file that contains the pages contents plus
other elements (Cascade Style Sheets CSS, Images, and JavaScribt codes/files.) The
browser at the end user side starts rendering the HTML file. It will find tags for the
other elements that require fetching. It sends a GET request to retrieve these ele-
ments. Ads come to the picture in the form of HTML or JavaScribt code, once such
code is triggered while rendering the HTML file, another GET response is generated
and sent to a specific ad server (Ad network, e.g. AdSense) to download ad content.
The ad sever figures out what sort of ads should be sent to the users based on a
matching technique that is done with the publisher’s webpage content (Contextual
targeting.) Once a user finds an interesting ad and clicks on it, he/she will be redi-
rected to the advertiser’s web-page. Behind the scene; first, advertisers make special
sort of agreements with the ad network that specifies the pricing method and provide
the ad network with the ads of products they want to promote. Publishers as well
make other agreements with the ad networks that specify the financial conditions and
the some regulations on what kind of ads are allowed to be published on their pages.
The ad network, in turn, provides the publisher (website) with a piece of JavaScribt
code or HTML tags that the publisher embeds in its web-page.

• Mobile in-Application Advertising

In this scenario, things are different. Developers of mobile applications embed some
ad libraries, as illustrated in Fig. 2.3, that are supplied to them by ad networks,
within their applications’ code. These ad libraries are pieces of code that send re-
quests to ad networks in order to download ads and display them on the apps while
they are run by users. The same role is played by the ad network (e.g., AdMob,)
which connects both advertisers and publisher (apps developers) together. In this
scenario, though, ad libraries are set to send certain information regarding the users
so that ad networks deliver ads that best match the user interest, which is called
targeted advertising. Moreover, ad requests are periodically sent by the embedded
ad libraries to ad networks, in order to show a variety of products and maximize the
chances that a user clicks on them. As a result, developers maximize their profit by
having more clicks on the ads they display on their apps. In some particular cases,
namely games, another form of in-app ads can be found, where some brands are
displayed in the background graphics of the game itself.

14
Figure 2.2: Sequence diagram of Mobile web browsing (DNS: Domain Name System.)

Figure 2.3: Example of in-App Ad library as it appears in the source code file [23]

15
2.3.2 Advertisement format
Different ad formats can support different ad types. The most common formats used in
mobile advertising can be categorized as follows:

• Text ads

Figure 2.4: Examples of text ad formats

Text ads on mobile devices contain mainly three lines of text: (i) a title; (ii) a short
description; and (iii) a URL that links to the advertiser’s website. Fig. 2.4 shows an

16
 example of such ad formats. This type of ad is very common in search advertising
and considered to be the most dominant one as well as the less annoying in terms of
user experience. For example, when a user searches for certain information using a
search engine, the first few results in the top of the list are usually ads. The ads in
such advertising are usually selected based on: (i) the keywords that the user used,
(Keyword-targeted advertising), or (ii) based on the content of a webpage, (Contextual
targeting).

• Display ads
Display ads are one of the oldest forms of advertising. They are basically images
designed with high graphics content and with a link to the advertiser’s website, and
occupy a fixed area of a webpage or an application. Fig. 2.5 shows an example of this
kind of ads. The most familiar type of display ads are banners which are normally
located on the top or/and the bottom of web-pages or apps. In some cases these ads
include rich media within the banner ad itself. Those ads come in a form that consist
of multimedia elements such as sound and animation. In addition, those ads require
browsers capable of executing JavaScript codes.

2.3.3 Ad selection mechanisms

There are several methods used to best select ads to be displayed on the user screen. Those
methods are based on either: (i) the information gathered about the end user Behavioural
Targeting, or (ii) the web-pages’ content that a user is surfing (Contextual Targeting).

• Behavioural Targeting: In this method, ad networks make use of the information

gathered about the end user. Then they apply complex algorithms to accurately
match the user’s interests and needs with appropriate ads. Such information can be
found in the form of the history of a user’s browsing, the user’s current location, and
the apps that a user has on his/her mobile device. Based on this information, ad
networks profile users and hence are able to provide them with the most relevant
ads. For example, ad networks would send ads that are about sports tools to a user
who is known to have searched or visited specific sports tools section of a general
sport-store website).

• Contextual Targeting: This method involves no user information in the process. In-
stead, the ad network selects ads based on scanning the keywords of the page that
a user is surfing. Such a process is very complex and requires tools that uses data
mining algorithms to map the web-page content with the relevant ad content. For

17
Figure 2.5: Examples of Display ads format.

18
 example, if a user is surfing a website that provides information about learning the
English language, the ad network would send ads for some English Language Schools
or electronic dictionaries.

2.3.4 Ad pricing models

As mentioned in subsection 2.3.1, an ad network makes a number of agreements with
advertisers and publishers to get the system functioning appropriately; one of them is the
pricing agreement. Based on the following pricing models, ad networks and publishers get
their revenue.

• CPC (Cost Per Clicks): In this model, the cost of advertising is based on the number
of clicks received. Advertisers pay the publisher and the ad network that delivered
the ad to the end user when the user clicks on the ad. However, a user does not
necessarily buy something. From the advertisers side, once a set amount of clicks has
been reached, the advertising campaign is made inactive.

• CPM (Cost Per Mille): In this model, the cost per mille is media term describing the
cost of 1,000 impressions- views. The advertiser pays for the exposure of their ads.
In such a case, advertisers buy a package with a certain number of views; this can
be any positive number. Once that fixed number is reached, the campaign remains
in a finished status, e.g., a $70 CPM means the advertiser pays $70 per every 1,000
times his/her ad appears on any website.

• CPA (Cost Per Action): Advertising impressions are free in this model, and adver-
tisers pay only for the number of users who complete some action that advertisers
feel will lead to a sale. Publishers run the ad at their own risk and do not get paid
until users click the ad and make a purchase, sign up for a service, or register for a
newsletter.

2.3.5 Advertisements in Smart Devices

A fundamental question that can be asked is: Why are smart devices a suitable environment
for advertising? Basically, the features that mobile environment offer make them a suitable
advertising platform. In addition, the almost 6.8 billion mobile-cellular subscriptions world-
wide [6] is an indication that there is a huge opportunity that advertisers can reach their
audience almost effortlessly. Compared to web-based advertising, apps and other mobile

19
services create an attractive and unique channel to the advertisers to present their ads to
the potential consumers. Moreover, advertisers have the capability to instantly target the
consumers individually by leveraging the Global Positioning System (GPS) location and
the search history of web-browsers and map apps.

2.4 Literature Review of Existing Research

2.4.1 Energy Cost of Advertising in hand-held devices

The energy consumption and the battery life in smartphones have been the topic of numer-
ous publications [42][39][47][59]. Energy management solutions in a smartphone environ-
ment were discussed in details in [58]. However, fewer publications have reported on the
issue of energy consumed by the advertisements embedded in both apps and web pages.
In this section, the current existing work is categorized into applications and web pages,
as described below.

2.4.1.1 Applications

Prior studies have deeply analyzed the usage of mobile apps from an energy perspective,
since battery life is a very important aspect of smartphones. In particular, they focused
on the free apps, because free apps rely extensively on ads as a source of revenue. Such
ads are actually battery hungry and adversely affect battery life. The energy studies on
smartphone apps were either done by measuring the energy using hardware tools [47][59]
or estimating it using software tools [42][39], such tools are based on modeling energy-
consumption behavior. Table 2.1 shows a summary of the work done on the impact of
in-apps advertising on the battery life of smartphones.

In [42] for instance, a huge effort has been made to profile the energy consumed by
hardware components of smartphones (GPS, Camera, CPU, Wi-Fi, 3G, etc.) and link
the amount of energy drain to the originating apps. The authors implemented a software
tool called eprof that is used to estimate the energy consumed by apps. The tool is a
fine-grained energy profiler that can model the power consumption online on Android and
Window smartphones. The modelling of the consumed power is done based on Finite State
Machine, FSM [43]. Such a tool is able to track program calls as well as track the power

20
 Table 2.1: Summary of the work done on in-apps advertising energy cost

Ref. Platform Kind of Network Evaluation Findings Proposed Solu- Improvements/

# Apps Access Method/ Mea- tion Observations
Mode surement tools
[42] Android, Games, Wi-Fi, Eprof: an 65%-75% of the en- Certain bun- 20% to 65% re-
Win- news, 3G energy ac- ergy consumed by dles strategies duction of the
dows Social net- counting tool; apps is due to ads. total energy
working Energy Mod- consumption of
apps, Web elling Software 4 selected apps
Browser based on FSM.
[47] Android Free games Wi-Fi Experimental in 5 popular games, N/A The power
apps work using ads consume up to consumed by
Monsoon 4%-15% of apps to- ads depends on
Power monitor tal energy how frequently
tool [56] ads are fetched
[59] Android A purpose- Wi-Fi, Experimental Parameters such Adcache, a The power con-
built app 3G work using as; refresh inter- caching mech- sumption due
Monsoon val and retrieval anism that ads is reduced
Power monitor mechanisms can reduces the by 50% using
tool [56] significantly im- impact of Adcache
pact the power ad traffic on
cost of ads. How- battery life
ever, ads power
overhead due to
3G connectivity is
more than Wi-Fi.
[39] Widows Games, 3G WattsOn [37], Ads contribute by Ad prefetching a reduction
news, an energy esti- ∼65% of the to- with app usage of more than
weather, mating tool for tal app’s communi- prediction 50% in energy
productiv- Windows mo- cation energy consumption
ity, social bile phones. due to serving
networking ads
apps
[39] Windows Browsers Wired Experimental The measurements N/A Based on their
(on a connec- work show that the addi- results and
PC) tion tional energy con- some national
sumption due web statistics, they
advertising is 2.5 found that the
W. total energy
used by ads
equalize the
annual elec-
tricity usage
of 1891 Dutch
households.

21
consumed by hardware components of the smartphone, and finally map the power con-
sumption to the apps that initiated such calls. This solution is meant to help developers
design energy-aware apps. The findings of Pathak et al. [42] revealed a number of interest-
ing points, most importantly the amount of energy that is spent by third-party advertising
libraries.The authors claim that 65% - 75% of the energy consumed by free apps is spent
on fetching advertisements and uploading users’ data to remote servers. Such data is used
by advertising networks to serve up ads that are most relevant to users.

Another study that addresses the cost of ads in smartphones was published by Prockova
et al. [47]. The authors investigated the influence of ads on mobile games apps running on
Android platform. They found that the existence of ads in such apps increases the energy
consumed by the device while using a Wi-Fi confection. For instance, in five mobile games,
4-15% of the total energy consumption was due to requesting ads. The reason behind this
increase is that each time an app publishes an ad, it first fetches it and then displays it.
In other words, the radio interface has to be turned on to transfer the ad content from the
ad server to the smartphone each time a new request is made by the embedded ad library.
Hence, more energy would be consumed during such tasks. The process takes place peri-
odically, depending on how frequently a new ad is fetched. The experimental work in [47]
focuses on measuring power consumption due to network usage, i.e., the cost of fetching
advertisements. The authors relied on comparing free apps that include ads and the paid
ones that are ad free. The difference in the measured consumed energy between the two
types is the pure energy cost of fetching and displaying ads on free games on smartphones.

Two months later, the energy consumption of ad networks was addressed in [59].
Vallina-Rodrigues et al. started by characterizing mobile ad traffic and then discussing the
benefits of some techniques that are used to mitigate the energy and network signalling
overhead caused by the current ad delivery systems. They proposed a system called Ad-
cache that minimizes the cost of ad delivery mechanism by friendly caching ad content and
serving it locally when needed. Adcache allows the retrieval of ad contents under certain
optimal network conditions. The advantage is to cache the ad content whenever the the
optimal network condition is satisfied and displayed it on used apps later. Such a tech-
nique enables publishing of ads with less energy wastage as well as less network signalling
overhead. Their results shows saving of up to 50% of the energy consumed by the process
of delivering ads using Adcache. The evaluation was done on a purpose-built app where
measurements were taken over both radio interfaces, namely, Wi-Fi and 3G, with different
ad retrieval mechanisms and different refresh intervals. They, moreover, compared the

22
energy consumption of requesting ads from most popular ad networks: AdMob, Millennial
Media, and InMobi.
Mohan et al. in [39] have addressed the in-app mobile advertising from two important
aspects. First, they have studied how much ads in Widows phone apps contribute to the
total energy consumption. They found that ads, on average, account for 65% of the total
app’s communication energy. Then they proposed a solution based on ads prefetching and
apps usage behaviour predictability. The evaluation of the proposed approach shows a
reduction of 50% in energy consumption due to serving ads.

2.4.1.2 Webpages

In the context of studying the energy cost of web ads while mobile browsing, there are
not much studies that talked specifically about this issue [55]. However, there are several
research articles that investigated mobile browsing from energy perspective [67, 57, 42, 68]
as well as other performance aspects, namely, Speed, simplicity, and usage and usability
[63, 26, 50].

Studying the energy cost of web advertising on smartphones or hand-helds in general

has not been performed so far, to the best of our knowledge. However, there is a study done
by Simons and Pras [55] at targeted the hidden energy cost of web advertising on desktops
(PCs). They investigated the amount of CPU and display energy consumption due to web
advertising and tracking. They found that displaying ads that are rich in animations and
graphics is an energy-expensive task. It puts the CPU under computational stress that
causes higher power consumption and leads to harming the environment as well as costing
more money. For instance, their results reveal that the cost to render and display web
ads is 2.5 W. Using some national statistics, and based on their results, they found that
the total energy used by ads equalize the annual electricity usage of 1891 Dutch households.

The browser’s choice impact on battery life of laptops was investigated by Walton in
[61]. He made an effort to answer the question of: Which web browsers is the most energy
efficient on laptops? Webpages energy performance modelling was the issue in several stud-
ies. In [42], Pathak et al. profiled the Android browser for 30 seconds over a 3G link using
the proposed eprof. The scenario was: a user opens the browser, does a Google search and
then closes the browser. they found where exactly the energy spent in the browser. HTTP,
GUI, user tracking and TCP conditioning consume 38%, 5%, 16% and 25% of the total
energy consumed by the browser, respectively. In [67], a similar tool called PowerTutor was

23
described. PowerTutor is a tool, available on the Android Application Market, that esti-
mates and profiles the power drained from apps and informs developers about the resultant
energy implications of their software designs on the battery life. Understanding webpage
characteristics from energy perspective was considered in [68][69]. The authors exploited
such understanding and applied it to achieve more energy-efficient mobile web browsing.
Their methodology focuses on effectively making full use of processor frequency scaling to
balance its performance with the resultant energy consumption of loading webpages.

In [57], Thiagarajan et al. analysed the energy consumed by mobile browsers. They
measured the total energy cost of rendering web page elements as well as the 3G radio
energy needed to fetch such a web page. Then, they went further by breaking down
such measurements and answering the question of how much energy each and every web
component consumes: cascade style sheets CSS, Javascript, images, and plug-ins. Based on
their measurements and analysis, a number of recommendations were provided to develop
and build more energy efficient site. However, the ads were not in the focus in [57]. Some
other studies looked at web browser and web pages from other perspectives. For instance,
in [63], Wang et al. critically analysed the slowness of web browsers on smartphones. They
found that the main reason the slow web browsing is the process of fetching different web
contents. They concluded that the key point of improving mobile browsers is to speed up
such process. In [26], Hoehl et al. investigated the possibility of benefiting from presenting
mobile version websites on desktop computers so that simpler web contents can be provided
to people with cognitive disabilities.

2.4.2 Bandwidth Cost of Advertising in hand-held devices

Due to the scarcity of resources on mobile devices, the topic of characterizing the traffic for
mobile devices has been addressed in several studies in the recent years. Such studies range
from understanding the nature of data traffic to analysing the cost of particular contents
(i.e. ads.)

24
 Table 2.2: Summary of the work done on in-apps ad bandwidth cost

Ref. # Platform Kind of Network Evaluation Method/ Findings Proposed Solution Improvements/ Observations
Apps Access Measurement tools
Mode
[66] Android, Games, Wi-Fi Experimental work us- The cost of 4 free apps per N/A Given a limited monthly data plan, the
iOS Enter- ing tcpdump [28]. tcp- month range from $2 (∼40 cost of some of the free apps is 48%-
tainment, dump a network mon- MBytes) to $13 (∼210 1299% higher than the same paid apps
Education, itoring tool that cap- MBytes), depending on the
Life style, tures all the 802.11 platform, data plan price,
free app packets and usage time. (Note:
tools. The cost analysis is based
on empirical calculations)
[30, 29] Android Free apps Wi-Fi, Experimental work us- Some free apps that are CAMEO, a middleware frame CAMEO can mitigate the cost (Energy
3G ing Shark for Root App played for 30 minutes a work that can mitigate the cost and Bandwidth) of ad delivery mecha-
[1] day, can generate up to 40 of ad delivery. It’s based on nism significantly
MBytes a month caching ads locally and pre-
dictively download them when-
ever un-metered or inexpensive
wireless connectivity is avail-
able. And then sever them lo-
cally when a user runs an app
[59] Android, N/A Cellular Off-line analysis of a Ad traffic contribute with N/A N/A
25

iOS network data set comprises of 1% of the total 22 TBytes

22 TBytes of traffic, (∼220 GByte).
that belongs to one day
of European mobile op-
erator with 3 million
users.
[51] Any de- Any apps Wi-Fi, N/A N/A MASTAds, an ad delivery ar- MASTAds provide means to deliver
vices 3G chitecture that can be inte- ads to mobile phone apps with mini-
grated to the existence ad eco- mum resource usage (Energy & Band-
systems with minimal changes. width) without compromising user’s
privacy
[60] Windows Browser Wired Experimental work us- Ads take up 7-9% of the to- N/A The additional cost would be $4.5-
(PCs) connec- ing Wireshark [15] and tal data traffic 7, $50-80 while browsing domestically
tion Proximodo [46]. The and internationally, respectively
first tool was used to
monitor the traffic, and
the second to enable
and disable the ads
 Since the generated mobile data traffic is a new addition to the traditional traffic, the
first study that noticed the emergence of the added mobile traffic was by Maier et al. [35].
They have found in their data set that comprise more than 20,000 residential DSL lines,
Digital Subscriber Lines, that the contribution of the traffic due to mobile hand-held de-
vices to the overall traffic volume is still small. However it is growing very rapidly compared
to the traditional traffic. The work in [21] is similar to the one done by Maier. However,
Falaki et al. in [21] broke it down to the application level and analysed a small data set
that consisted of 43 smartphone users to study the nature of smartphone traffic. They
found that the largest contributor of the traffic is web browsing. Based on their analysis,
moreover, they proposed several mechanisms that can improve the power consumption of
smartphones as well as its overall performance. In [32, 14], attempts have been made to
analyse mobile traffic. Such researches were conducted to provide network operators with
a clearer picture of the nature of mobile traffic and the originating applications, so the
network operators would optimize their quality of service accordingly. However, Chung et
al. [14] were the first to show the contribution of the traffic due to advertisements. They
found that ads contribute with 0.007% of the total mobile traffic that was gathered from
a university with 4000 users. These studies have led to deeper studies that focus on the
traffic generated only from mobile advertising as we will see next.

2.4.2.1 Applications

Several previous studies have focused on analysing the overhead traffic generated by net-
work related apps. Specifically, they focused on the free apps that embed ad library to
generate revenue to the developers. In [66], for instance, Zhang et al. answered the ques-
tion of how expensive Free Apps are. In their study, they consider the traffic that is not
necessary to the app’s operation as overhead, specifically, ads and analytic data traffic.
Their measurements were done on a Wi-Fi access point, and they found that some free
apps generate up to 1 MB of overhead traffic while being in use for 10 minutes. Moreover,
they estimate the cost of such traffic in terms of the monthly payment a user makes for
cellular data plan. A comparison between a number of free and paid apps were made. In-
terestingly enough, their detailed measurements show that, given a limited monthly data
plan, the cost of some of the free apps is 48%-1299% higher than the same paid apps.
In [30, 29], an attempt was made to mitigate the real cost of free mobile apps. Khan et al.
[30, 29] first analysed the network overhead due to the delivery of mobile ads. They found
that, for 30 minutes of active use of free apps, such overhead can contribute with a good
fraction of the total data usage. They also found a free app namely Fruit Ninja Game,

26
played for 30 minutes a day, can generate 40 MB of ad traffic in a month. Such app usage
can cost ∼56 cents a month which is significantly expensive compared to the paid version
of the same app (without ads) that cost only 99 cents. Then, based on the observations
gotten from the above analysis, they proposed a new advertisement delivery mechanism
called CAMEO that can reduce the bandwidth cost of in-application advertising. Their
method aims basically to: (i) predict the kinds of ads that should be displayed to the user
based on utilizing his/her context; (ii) cache the ad content (prefetching) to a local cache
when the user is connected to un-metered connections (such as Wi-Fi); and (iii) display it
when the app is being used.
In another major study, Vallina-Rodriguez et al. [59] used data sets, from European mo-
bile operator corresponding to one day of traffic to characterize mobile ad traffic. They
developed a methodology that classifies the gathered traffic into three main components:
ad network traffic, analytic traffic, and mediation services traffic. Their analysis shows
that ad traffic contributes with 1% of all mobile traffic in the data set, with 31 kB per day
for 50% of the devices. They found that ad traffic generated by 50% of Android devices
accounts for more than 5% of the total daily traffic. The case for Apple devices is almost
the same, the volume of ad traffic generated by 20% of iPad and iPhone devices account
for more than 4.7% and 7.6% of their traffic, respectively.
In a very recent study, Seneviratne et al. [51] focus on in-app advertising. They
studied the existing in-app mobile advertising eco-system, which enabled them to precisely
address the issue of maximizing system efficiency, i.e. minimizing the resources used to
deliver ads to smarphones without compromising users’ privacy. Moreover, they proposed
a new mobile advertising system, called MASTAds (Mobile Anonymous but Still Targeted
Ads), that delivers targeted ads efficiently without having the traditional conflicts, which
are saving energy and bandwidth and preserving users’ privacy.

2.4.2.2 Webpages

There have been some studies in literature that quantified the traffic generated while web
browsing. An attempt was done by Erman et al. [20] to classify the HTTP traffic in homes.
Their study covered 17,000 broadband DSL subscribers’ traffic for a month in Texas. They
answered some questions that characterize the nature of HTTP traffic in homes. For in-
stance, they found that not all the HTTP requests are made by PCs or laptops. Also,
they found that 11% of all HTTP requests are being made to communicate to advertising
servers, which is equivalent to 0.2% (that is ∼312 GB) of the total bytes downloaded.
However, their study did not particularly target the amount of ad traffic generated only
from mobile web browsing. Their results show the ad traffic generated by PCs, laptops,

27
mobile application, and mobile web browsing.

Brande et al. [60] considered ads as an unwanted traffic while mobile browsing. They
investigated ads traffic contribution in the total web traffic. They built a testing system
in a way that allows them to block the ads when needed, and automated the browsing
process to mimic the browsing behavior. Their results reveal that web advertisement take
up 7-9% of the total data traffic. Based on the percentage of the generated ad traffic and
the average domestic and roaming charges, the authors estimated the cost of web adver-
tisement for mobile users. They found that the additional cost would be $4.5-7 and $50-80
while browsing domestically and internationally, respectively.

28
Chapter 3

Measurements and Analysis

This chapter describes a number of experiments carried out to achieve the first two contri-
butions of this thesis. Section 3.1 describes the performance metrics used throughout this
thesis. Section 3.2 details our testing methodology and test benches; finally, the results
and measurements of the energy and bandwidth are presented and analysed in Section 3.3.

3.1 Performance Metrics

The measurements and analysis of smartphone resources are conducted based on specific
chosen performance metrics. The main metrics used to evaluate the impact of web adver-
tising during mobile web browsing are the Bandwidth (B.W) and Energy Consumption.

• Bandwidth is the bandwidth needed to download web-pages, and has been chosen to
be an evaluation metric for the following reasons: (i) downloading web ads consume
B.W, and in the wireless environment, the available data plans tend to be limited and
costly, especially if the monthly cap is exceeded, and (ii) not all of the downloaded
information is of interest to end users. Accordingly, this “scarce” resource is evaluated
to study the impact of web advertising during mobile web browsing. The term B.W
cost in this thesis means how many bytes are transferred over the network to download
a web-page.

• Energy Consumption: as mentioned in section 2.2, the battery life is known to be

the bottleneck in the advancement of smartphones; therefore, and since ads are con-
sidered overhead data that consumes extra energy, the energy consumed while web

29
 browsing is chosen as a performance metric. The term “energy cost” in this thesis
means how is much energy drained from the battery when a web-page is accessed.

3.2 Testing Methodology

To perform the testing, the first challenge was how to separate the web content. A method
was sought to block or insulate the ads from the web-pages. The available blocking tech-
niques are as follows.

3.2.1 The Available Ad-blocking Techniques

• Ad-blocking Applications: A number of apps available online can block the ads,
namely Ad-Vanish Lite1 , NoRoot Ad-Remover Lite2 , and AdFree3 . These apps work
by turning off the Internet connection completely. They are designed to block ads
from Android games and works is as follows: a user is requested to prepare an Ad-
block list, so if he/she wants to stop displaying ads on a game, all he/she needs to do
is put this game on the Ad-block list. These apps in turn make sure that the games
will have no access to the Internet any more, and hence the ads are blocked. These
Ad-blocking applications work fine for the apps that do not require Internet con-
nection. In our case, web browsing, establishing an Internet connection is essential.
Therefore, this Ad blocking option does not satisfy our testing methodology.

• Browser Plug-in Ad-blockers: mobile web browser such as Mozilla Firefox and An-
droid can use plug-in ad-blockers to block ads. Some plug-in apps are available online
(Adblock Plus 4 ,) their functionality is based on filtering out the ad URLs. These
plug-ins have a black list that contains advertising companies’ URLs; thus, when-
ever an ad request made by the browser is found on the black-list, that request is
aborted or cancelled. Our comments on these ad-blocking methods are as follows:
(i) installing these plug-ins will change the way browsers work, and hence the mea-
surements will be distorted. In other words, having such plug-ins working in the
background of the browser requires extra activities and computational power. As a
result, extra energy is consumed or wasted. For our case, where we are looking to
measure the real energy and bandwidth required to download and display ads, using
1
https://play.google.com/store/apps/details?id=com.atejapps.advanishlite
2
https://play.google.com/store/apps/details?id=com.atejapps.litenorootadremover
3
https://play.google.com/store/apps/details?id=com.moshedavidson.browser
4
https://adblockplus.org/en/android-install

30
 this option will give us distorted measurements that are not accurate. (ii) Not all
of the ads will be blocked since the operation of these plug-ins relies entirely on the
specified list, which may not contain all the advertising companies (Ad networks);
see section 2.3.1 for further details. As a result, some ads will still be displayed even
with using this option. (iii) The ad links (ad tags) that a web-page has are already
delivered and parsed by the browser; as a result, some bandwidth and energy would
have been consumed. For these reasons, we had eliminated this option as well.

• Proxies: Proxies, namely, Privoxy [46], can be installed at a middle point (e.g., a
Wi-Fi access point or a network router) to block ads. The mechanism used to block
the ads is similar to the one explained in Ad-blocking Applications. Therefore, this
option has also been eliminated.

The last two options suffer from the same problems; moreover, they require human effort
and time to keep their ad-blocking lists up-to-date. Therefore, another approach was need
to achieve the goal of blocking ads without engaging the smartphone in this process.

3.2.2 Our Ad-blocking Strategy

The approach used to tackle the ad-blocking problem and take full control of the web
content crystallizes in having our own web hosting server. By doing so, we can either enable
the ads or remove them completely from the web-pages before they are even requested by
the end user. The process is as follows; (1) choose certain web-sites, (2) download them and
made two copies of every one, (3) modify and remove ad codes found in one of them and
keep the other one as is, and (4) upload one at a time to perform the testing. We modify the
the web-pages manually by removing all the ad-related content from the requested HTML
document. Figure 3.1 illustrates the sequence of the ad-blocking process and the final view
that is displayed on the end user screen after the ad-blocking technique is applied.

3.2.3 Energy Test Set-up

The test bench used to measure the energy consumption in smartphones is described in this
section. Our experimental setup is shown in Fig. 3.2. We use a Monsoon power monitor
[56] to power the smartphone and accurately measure the real-time consumed current. We
partially disconnect the phone’s battery to allow the communication work between the
phone and the battery. We do so to prevent the phone from going into sleep mode or
power saving modes. As shown in Fig. 3.3, the battery has four pins, namely, the positive

31
Figure 3.1: Ad blocking strategy

32
 Figure 3.2: The energy measurement set-up.

(pin # 1) and ground (pin # 3) pins, communication (pin # 2) and temp line (pin # 4)
pins. We insulate the battery’s pin # 1 and connect the one in the phone directly to the
power meter. We keep the other three pins connected; however, we also connect the power
monitor ground to the phone and battery ground as illustrated in Fig. 3.3. By doing so,
we can measure the actual current drained by the smartphone without any hidden cost.
The power monitor is connected to a desktop where the measurements are received
and stored for off-line manipulation and analysis. The PC runs the Monsoon Power Tool
software. This power monitor provides very accurate measurements; at a sampling rate
of 5000 sample per second, and also has adjustable DC voltage output (from 2.1 to 4.5
V.) Moreover, Matlab is installed in the PC, and is used to process that measurement and
generate the result graphics. We performed our experiments on a Galaxy Nexus smartphone
running Android V4.2.1. The phone settings during the testing were kept constant to
provide consistence measurements.
To conduct the experiments, we chose Android Browser that is the built-in web browser
in Galaxy Nexus smartphone and another browsing application “semi web browser” that
we developed. The energy cost experiments were done over a 3G cellular link and a Wi-Fi
private access point (AP) installed in our lab. The AP is Cisco Linksys, and it supports
the IEEE802.11g interface. For the 3G cellular connection, an HSPA+ Bell prepaid SIM
card was used. This connection provides a typical speed up to 7 to 14 Mbit/s [2]. The
experiment was done under good network conditions. For the phone configurations, we
followed the methodology proposed by [7], in order to have consistence measurements.
To ensure that the measured energy is the real cost of requesting web-pages, we make

33
 Figure 3.3: The battery bypass connection

sure no data is cached and all the web elements will be requested entirely fresh from the web
server in each web request. Therefore, we clear the browser’s cache before every experiment
and also disable all the other network-related applications (e.g., Android update libraries.)
The web server used for testing, is an HTTP server implemented in Java, located in our
lab. The server is deployed on a Dell desktop machine running Windows 7.

3.2.4 Bandwidth and Network Test Set-up

The test bench used to monitor the traffic going from and to the smartphone is shown in Fig.
3.4. We installed Wireshark on a Mac-book laptop to leverage the Monitor Mode feature
that is available built-in on Mac-book laptops. This feature allows the packet sniffing
task to performed passively; that is, we can capture all the traffic exchanged between the
AP and the smartphone without involving any activities on the phone. Wireshark is a

34
 Figure 3.4: The bandwidth measurement set-up

powerful monitoring and analysis tool, and enables us to monitor packet information from
the radio level up to the application level packet information. The collected packet traces
are analysed to classify how much bandwidth web ads consume and how much bandwidth
the non-ad related web contents consume.

3.3 Energy and Bandwidth Measurements of Web Ad-

vertisements in Smartphones
3.3.1 Energy Measurements Test Cases
Using the described testing infrastructure for measuring the energy consumption we were
able to measure the instantaneous drained current when a user is accessing the web. We
started our energy experiments by testing how much energy is consumed when browsing
a number of websites. The terms “drained current” and “consumed energy” are used
interchangeably throughout this thesis due to the direct relationship that connects both of
them, as shown by the equation E = I ∗ V ∗ t, where E is the energy, I is the current, V is
the voltage, and t is the duration time.
We chose a number of websites: (i) the most-accessed local news websites 5 , (ii) a highly
5
www.thereocrd.com

35
 1000 1000

900 900

800 800
Current (mAmp)

Current (mAmp)
700 700

600 600

500 500

400 400

300 300
0 10 20 30 40 50 60 70 0 10 20 30 40 50 60
Time (Sec) Time (Sec)

(a) www.macleans.com (b) www.therecord.com

1000 1000

900 900

800 800
Current (mAmp)

Current (mAmp)

700 700

600 600

500 500

400 400

300 300
0 10 20 30 40 50 60 0 5 10 15 20 25 30 35 40 45
Time (Sec) Time (Sec)

(c) www.thestar.com (d) www.canoe.com

1000

900

800
Current (mAmp)

700

600

500

400

300

200
0 5 10 15 20 25 30 35 40 45
Time (Sec)

(e) www.nytimes.com

Figure 3.5: The real-time energy footprint of a number of websites over a Wi-Fi connection

36
ranked international news website6 , and (iii) three samples of highly ranked Canadian
magazine websites 7 , according to Alexa [5]. We surfed the full version of each website,
using the built-in Android web browser over Wi-Fi and 3G connections. Figure. 3.5
illustrates the energy behaviour of a number of news websites.
First of all, we noticed a common energy-consumption footprint for mobile web brows-
ing. As can be seen in Fig. 3.5, the energy behaviour starts with the phone’s idling energy,
which is the state before a user launches the web browser. This state is marked on the
figures and found to be approximately 300 mAmp, and is followed by high-energy activities
that last for some period of time. The high-energy-activities period represents the energy
needed to; (i) launch the browser app, (ii) download the web content (the network/radio
interface energy cost), (iii) render the received HTML file (the computational energy cost),
and (iv) display what is being rendered and is ready to be displayed on the user screen.
The next energy state represents the energy cost only of displaying the entire webpage
content. In general, we found that browsing a full version of the website over Wi-Fi for 30
seconds consumes ∼60 to 73.9 Joules.
To correlate the energy activities with the displayed web content, we started from an
end-user point of view and closely examined Fig. 3.5. We noticed, for instance, periodic
energy spikes that occur approximately every 5 seconds in Fig. 3.5(b). To investigate
the real cause of such an energy activity, we looked into the contents displayed on the
phone’s screen and found that these spikes are due to displaying an ad content. This ad
content is basically a dynamic ad that displays different ads every 5 seconds. We then went
further, to the application layer, and investigated the HTML document that contains this
ad content. We found out that this ad content is made of JavaScript piece of code, such
a content requires high computational power to render and display on the screen. Thus,
such activities cause extra energy consumption.

3.3.1.1 Energy-Bandwidth Mapping in Webpages

In an attempt to understand the energy behaviour of web browsing, we correlated the en-
ergy consumption footprint to the network activities needed to download the web content.
Doing so, we can identify how long fetching a web component takes, and consequently,
how much energy doing so consumes. Using the test bench described in section 2.4.2, we
conducted an offline analysis of the traffic being transferred from and to the smartphone
while a user is accessing the web. Figure. 3.6 maps the network activities (that is, the
loading time for each TCP connection) onto the consumed energy. As was expected, we
6
www.nytimes.com
7
www.macleans.com, www,thestar.com, and www.canoe.com

37
 Figure 3.6: A sample of Energy-Bandwidth Mapping in a Webpage

notice that there are multiple TCP connections are established in parallel. Multiple par-
allel TCP connections are typically needed to speed up the retrieval of web content. The
first content that the browser downloads is typically the index HTML documents, as shown
in Fig. 3.6. Once this document is downloaded, the browser starts to parse and script its
content. By doing so, the browser finds other web resources (images, CSS, JavaScript files,
ads, etc) to be fetched. Consequently, more TCP connections are established. Now, since
the core web content and the ad content come from different servers, in some cases 3 to
13 TCP connections were found to be dedicated to ad servers and to analytic servers that
gather statistic and other information about users. Table. 3.1 lists the number of TCP
connections (# of servers) needed for advertising purposes.
As explained in Section. 2.3, for the browser to display webpages, all web content
referenced/tagged in the index HTML document, certain processes are required. The
browser has to (i) resolve the DNS of the URL that is referencing that web object, (ii)
handle the HTTP request that is to be made, (iii) establish the TCP connection with the
server containing that object, and (iv) render that object based on the rules specified in the
CSS. These time-consuming processes are shown in our figure to be under the time period
of a high energy state, which is not very surprising considering the high computations
required by the browser to accomplish its task, i.e. displaying the requested webpages. We
noticed also that the overall time needed to download all the webpage content is around

38
10 seconds, in some cases. Approximately 2 seconds beyond those 10 seconds show high
energy activity; we believe that is due to the computation needed to render the remaining
fetched objects (objects that are downloaded with no network activities are involved).
We found it very difficult to separate the energy consumed by the rendering process and
the radio/network interface activity. Moreover, it is important to mention here that the
objective of this thesis is to quantify the web advertisement energy and bandwidth impact,
not to focus on analysing the energy consumption of web browsers.

Table 3.1: Summary the traffic statistics while web browsing with and without ads
Website # of # of Time download Time download Energy
servers servers the page with the page with- difference
with Ads without ads(Sec) out ads (J)
Ads
www.therecord.com 17 9 10.5 8 15
www.nytimes.com 23 19 10 6 6.2
www.macleans.com 35 22 7 11 6.1
www.thestar.com 29 17 7.5 5 10.6
www.canoe.com 19 16 15.5 15 4.5

To identify the relationship between the number of TCP connections and the corre-
sponding energy consumption, we modified some webpages using our Ad-blocking strat-
egy. Applying this strategy allowed us to change the number of TCP connections that
take place. Then, we compared the energy consumption in both cases for each website,
and found, as expected, that the more TCP connections we have, the more is energy con-
sumed, as shown in Table 3.1. Moreover, as the number of TCP connections increases, the
fetching loading time increases as well, and so does the energy consumption. Downloading
the ad content prolongs the active period of the radio interface, and hence increases the
energy consumption.

3.3.1.2 Ads Energy Impact over Wi-Fi

To study the impact of web advertising on energy, we started by browsing the full version
of a number of websites over a Wi-Fi network, using the phone’s built-in Android Browser.
Then, we applied our Ad-blocking strategy, described in section 3.2.2, and repeated the
same experiments that were conducted for the full versions of the websites. Next, we
compared the energy needed to download web-pages with and without ads. Figure 3.7
shows the energy consumption for a number of websites. The dashed line refers to the web-
pages that are ad-free and the solid line refers to the full version of web-pages (web-pages

39
 1000 1000

900
Web−page without ads 900
Web−page without ads
Web−page with ads Web−page with ads
800 800
Current (mAmp)

Current (mA)
700 700

600 600

500 500

400 400

300 300
0 10 20 30 40 50 60 70 0 5 10 15 20 25 30 35 40
Time (Sec) Time (Sec)

(a) www.macleans.com (b) www.therecord.com

1000
1000

900 Web-page without ads 900 Web-page without ads

Web-page with ads Web-page with ads
800
800

Current (mAmp)
Current (mAmp)

700 700

600 600

500 500

400 400

300 300
0 5 10 15 20 25 30 35 40 45 0 5 10 15 20 25 30 35 40 45
Time (Sec) Time (Sec)

(c) www.thestar.com (d) www.canoe.com

1000

Web-page without ads

900 Web-page with ads

800
Current (mAmp)

700

600

500

400

300

200
0 5 10 15 20 25 30 35 40 45
Time (Sec)

(e) www.nytimes.com

Figure 3.7: The real-time energy footprint of a number of websites over a Wi-Fi connection
with and without ads

40
 1200

Without Ads With Ads

1000

800
Current (mAmp)

600

400

200

0
1 31 61 91 121 151 181
Time (Sec)

Figure 3.8: Energy measurement of our browser

with ads.) It is worth mentioning here that during our energy measurements we performed
passive network sniffing over the Wi-Fi connection as well. Our offline analysis of the
gathered network traces confirmed our previous argument that the more TCP connections
there are, the longer the loading time is, and consequently, the more energy is consumed.
As shown in Table 3.1, the energy overhead due to ads ranges from 4.5 to 15 Joules.

Reliability of Measurements
To ensure the reliability of our measurements, we developed a semi-web browser on Android.
The architecture of the browser is explained in detail in section 4.2.1. To ensure more
reliable measurements, the experiments were repeated multiple of times. We wanted to
keep the same settings throughout all of the experiments, and most importantly, minimise
end-user interaction with a phone. Therefore, we developed our own web browser. Doing
so (1) enabled us to repeat the same experiments a number (7 times) of times at consistent
time intervals (30 seconds); that is, every 30 seconds a request was made. (2) meant that
no information was allowed to be cached. (3) minimized end-user interaction with the
phone; that is, a single tap on the app’s icon was enough to start the browsing. Figure
3.8 shows the energy footprint while using our browser app. Moreover, we conducted a
number of tests just to ensure that our app energy consumption did not differ a lot from
the built-in Android Browser. We found in many cases that both measurements were very
close, and within some cases a 100 % match was noticed, as the measurements show in
Table 3.2 and Table. 3.1.
Table 3.2 summarizes the energy consumed by a number of websites, with and without
ads. The difference in energy consumption ranges from 6 to 17.5 %. Again, these figures are
due what we call ad-overhead, that is, the net energy cost of the processes of downloading,
rendering, and displaying web ads over a Wi-Fi network. These numbers are average
values. We noticed that the ad contents vary each time a request is made, and observed

41
 Table 3.2: The cost of ads over a Wi-Fi interface

Difference
Energy in the
Website Content Type Energy (J) Overhead Estimated
due to Ads Battery Life
(min)
browsing with ads 68.5 16% 27
www.therecord.com browsing without ads 59
browsing with ads 60 9% 23
www.nytimes.com browsing without ads 55
browsing with ads 65 12% 17
www.macleans.com browsing without ads 58
browsing with ads 67 17.5% 35
www.thestar.com browsing without ads 57
browsing with ads 61.5 6% 8
www.canoe.com browsing without ads 58

42
that webpages with ads take longer to load. As we shown in Fig. 3.7, around 4 to 10
seconds extra are needed to download and display ads. These times are reflected in the
form of extra energy, as the measurements in Table 3.2 show. We compared the extra
times introduced by the existence of ads just for downloading, (Table 3.1), and the overall
time added due to downloading and rendering ads (Fig. 3.7.) In some cases, most ads
energy consumption was due to the process of rendering. As the test case of browsing
www.canoe.com shows, downloading the ads requires only half a second extra time, while
the energy footprint in Fig. 3.7-(d) shows an extra 10 seconds of high activity energy state.
Moreover, Table 3.2 shows an estimation of battery life, i.e., how long the battery would
last if we were to browse each version of the websites repeatedly until the battery died.
These numbers illustrate how much energy is wasted by the existence of ads in web-pages.

3.3.1.3 Ads Energy Cost over 3G

100 Website without Ads (Wi-Fi)

Website With Ads (Wi-Fi)
Website without Ads (3G)
80
Website With Ads (3G)

60

40

20

Figure 3.9: The energy consumption of different websites over 3G and Wi-Fi networks

We repeated the same experiments over a 3G connection. Table. 3.3 summarizes

the energy consumed by a number of websites, with and without ads. We noticed that
browsing over a 3G connection consumes more energy in general. The energy overhead due
to web ads ranges from 6 to 17.5%. However, the energy difference due to ads is less than
the difference noticed over Wi-Fi. Moreover, we observed that browsing a full version of
a website, with ads, over Wi-Fi, is less expensive than browsing an ad-free version of the
same website over 3G connection, as shown in Fig. 3.9. With 3G usage, batteries die faster

43
than with Wi-Fi, which was certainly expected taking into consideration the observations
mentioned in this paragraph. The higher energy consumption over the 3G network is due to
the energy tails and the capacity dissimilarity, according to [48]. “Energy tail” refers to the
high energy state that the 3G radio interface has after the network activity is terminated,
while the capacity dissimilarity refers to the 3G bandwidth capacity limit. Compared to
Wi-Fi, 3G is slower, and consequently, downloading webpages over 3G will take a longer,
resulting in higher energy consumption.

Table 3.3: The cost of ads over a 3G interface

Difference
Energy in the
Website Content Type Energy (J) Overhead Estimated
due to Ads Battery Life
(min)
browsing with ads 87.75 14% 22
www.therecord.com browsing without ads 76.74
browsing with ads 72.8 7% 15
www.nytimes.com browsing without ads 68.25
browsing with ads 80.55 8.23% 14
www.macleans.com browsing without ads 74.44
browsing with ads 82.65 17.5% 28
www.thestar.com browsing without ads 70.29
browsing with ads 73.05 6% 5
www.canoe.com browsing without ads 69.06

When we considered browsing over the two network interfaces, 3G and Wi-Fi, we found
a difference in battery life of, in some cases, one hour of operation. In other words, we
could browse the web for an hour longer when using a Wi-Fi connection instead of 3G.
Based on this observation, we came up with the idea of “Smart Mobile Web Browsing”
that we explain in Chapter. 4.

44
3.3.2 Bandwidth Measurements Test Cases
Using the previously described testing infrastructure for monitoring network activities while
mobile web browsing, we were able to capture the network traces. We browsed a number
of news websites and conducted an offline analysis to measure how much bandwidth ad
downloading requires.

3.3.2.1 Ads Bandwidth Cost

Table 3.4: Traffic breakdown per web-page

Total page
Total page Total ads
traffic
Web site traffic (with traffic in Bytes
(without
ads) in Bytes
ads)in Bytes
www.therecord.com 3,025,082 1,449,882 1,575,200 (52%)
www.nytimes.com 975,646 634,018 341,628 (35%)
www.thestar.com 714,961 351,732 363,229 (50%)
www.macleans.com 987,682 617,857 369,825 (37%)
www.canoe.com 6,005,729 5,861,272 144,457 (2.5%)

Using Wireshark traces, we investigated the traffic needed to download ad components

in webpages, and found that in one case, surprisingly enough, ad traffic comprises almost 50
% of the traffic needed to download some news webpages. More detailed traffic statistics are
shown in Fig. 3.10, where we see that when ads are enabled, 8 extra servers are contacted.
We can also see that the number of HTTP requests made by the client (smartphone) is
almost three times more when ads are enabled, and the total number of packets exchanged
between the client and the access point is doubled. Table 3.4 shows the overhead bytes
needed to download ads. These numbers are average values since the ad contents vary each
time a request is made.
In summary, we found that ads consume a considerable amount of energy and band-
width. To have a better sense of the effective cost of ads, we, next, give an empirical
estimation of (i) how much energy a smartphone would consume to download and display

45
 Figure 3.10: Traffic statistics while web browsing with and without ads.

ads, and (ii) how much bandwidth it would cost a user to download these ads for a certain
period of time. We assume that a user spends around one hour on web browsing a day, not
necessary continuously. In this hour, he/she would spend 2 minutes per website; therefore,
on average, he/she would access at least 30 websites. Based on Tables 3.3 and 3.4, we get
the following:

• the energy that a smartphone would consume, to download and display ads in one
hour of browsing is 360 Jules/day; and

• the amount of bandwidth needed to download ads would be equal to 2.6 Mbyte/day,
that is 78 Mbyte/month. Based on some US metered data plans (usage-based pricing
plans) [60], $ 12.48 would be spent just on downloading ads.

46
Chapter 4

Smart Mobile Browsing Strategy

Web browsing is becoming very essential for every day life, especially with the ever in-
creasing popularity of smartphones. The web content delivered to end users has witnessed
drastic change. Webpages used to be simple “traditional” static pages compromised of
only text and some images. However, modern webpages are dynamic and media-rich. The
owners of these pages (publishers), moreover, rely largely on ads as a source of revenue.
These ads are very media-rich and consume a lot of resources, like battery and bandwidth
as we show in Chapter 3. Eliminating the existence of ads entirely from webpages is not
practical knowing the big role they play in the web eco-system and in contributing to the
availability of free web content. These factors have led to the demand for mobile brows-
ing solutions that adapt webpages in a way that is energy and bandwidth efficient. In
this chapter, we propose and implement a novel web browsing technique that adapts the
webpages delivered to smartphone based on the smartphone’s current battery level and
network type. Webpages are adapted by controlling the number of ads to be displayed.

4.1 Motivation
The Smart Mobile Web Browsing (SWB) design is motivated by the measurements and
the findings obtained in section 3.3. As mentioned there, the more ads a webpage contains
and the more the complex ads get, the more resources are consumed. As ads can consume
considerable energy while a user is mobile web browsing, in some webpages 18% of the
total energy, and that the bandwidth overhead caused by downloading ads in some web-
pages reaches 50%, we propose the above mobile browsing technique, whose pages delivery

47
mechanism is function of a smartphone’s resources and the amount of ads allowed to be
displayed in the webpages.
Our approach mainly targets (i) extending smartphone battery life; and (ii) preserving
the bandwidth needed to download the webpages. As mentioned in the previous section,
having ads displayed in web-pages is essential for having free web content; therefore, our
approach aims to balance the satisfaction of end users as well as that of webpage publishers.

4.2 Design requirements and objectives of SWB

This section presents the high-level concept and features of our Smart Mobile Web Browsing
(SWB) strategy. The key design decisions are as follows.

• Minimize the number of ads when needed. As shown in section 3.3, ads can be
very costly in terms of energy and bandwidth. Therefore, so that publishers do not
exhaust their audience’s smartphone’s resources, the amount of ads to be delivered
to the end user should be based on a device’s available resources. SWB adapts the
web-pages that are sent back to the smartphone based on a smartphone’s current
battery level and network type. The adaptation of the web content comes in the
form of how many ads are allowed to be displayed on the webpage.

• Minimum adjustment to the existing browsing systems. The modifications required

to be made on existing browsers and web servers are not complex. SWB implemen-
tation maintain this condition. For the server side, some plug-in modules can do
the necessary job. And for the client side, some components can simply be added to
existing browsers, as we explain in detail in section 4.2.1. The main challenge was in
what layer should we embed the required information in the client side. We decided
to choose the application due to the simplicity of adding optional fields in the HTTP
protocol. This approach facilitates the need to make minimum changes to existing
web browsers.

To achieve the above design goals, our solution strategy crystallizes in making the
server intelligent enough to sense a smartphone’s operational environment and provide it
with customized webpages. Having done that, SWB can extend a smartphone’s battery
life while users still enjoy browsing the full versions of their favourite websites (full pages).
This system also takes into account the type of network that the mobile client is connected
to and accordingly provides it with full pages that require less bandwidth to download.

48
 Figure 4.1: System Architecture

In this system, unlike other web adaptation solutions, e.g. [12], no user involvement is
required, the client application (browser) sends the current available resources (Battery
Level and Network Type) to the server, along with the HTTP request. Based on that
information, the server sends web content back to the client.

4.2.1 System Architecture

In this subsection, we introduce the overall system architecture, and the decision policy
that the server applies to prepare the response. Our approach is a client-server based
system, as depicted in Fig. 4.1, similar to other conventional web-browsing techniques.
The details of the system as follows:

• Smart Browser

49
 As shown in Fig. 4.1, the client application browser consists of six components,
namely User Interface, WebView engine, HTTP Header Handler, Network Inter-
face Handler, Battery Level Listener, and Connection Type Listener. The first four
components exist in all browsers; we present them separately to show exactly the
internally basic components of conventional browsers, and how the our system can
be implemented in general. The function of each component is described in Table
4.1.

Table 4.1: The functions of client’s components

Component Description and Function

User Interface This displays what user sees and interacts with when
running the browser.
WebView engine This is an Android Java class that provides browser-like
functionalities.
HTTP Header Handler This handles the modification that we do to the HTTP
request’s header.
Network Interface Handler This handles the HTTP requests that are made by the
browser.
Battery Level Listener This obtains smartphone battery level every time a user
makes a web request, and passes it to HTTP Header
Handler.
Connection Type Listener This gets the type of network the smartphone is con-
nected to and passes it to HTTP Header Handler.

• Smart Server
The Server in our system can be any HTTP server. We added a HTTP Header Parser
to extract the information that the browser includes with the HTTP request. Then
the information is passed to the Decision and Response Preparation unit, where the
policies are applied and decisions are made based on the current available resources
on the smartphone. A proper webpage is next prepared and sent back to the client.

Note that when a user wants to access certain website, the browser gets the Bat-
tery Level and Connection Type, modifies the HTTP header, and sends the HTTP “GET”

50
Algorithm 1: Smart Web Browsing
Input:
Modified HTTP Request
Output:
Customised Web Content
Description:
1: Server ⇐ (i) receives the HTTP request, (ii) extracts Battery Level and
Connection Type from the header
2: if Battery Level ≥ 80 ∧ Connection T ype = W i − F i then
3: return full version of the website
4: else if Battery Level ≥ 60 ∧ Connection T ype = W i − F i then
5: return Web-pages with only textual ads
6: else if Battery Level ≤ 40 ∧ Connection T ype = W i − F i then
7: return ads free web-page
8: else if Battery Level ≥ 80 ∧ Connection T ype = Cellular then
9: return Web-pages with only textual ads
10: else
11: return ads free web-page
12: end if

51
request. On the server side, The pseudo-code shown in Algorithm 1 explains how the server
works. The server receives the request, extracts the information needed to make the deci-
sion, applies certain policies, prepares the proper web-page, and then sends the web-page
to the smartphone. Lines 2-12 of Algorithm 1 shows how the decision policy parameters
can be set, .

4.3 System Implementation

We implemented a prototype system, in which the client side was implemented on Android
and the server side was implemented in Java. Figure. 4.2 shows the GUI interfaces of the
client side and the server side.

4.3.1 Client Side Configuration

The configurations that must be made on the client side (Browser) are described next:.

• Battery Level Listener. This listener’s main duty is to obtain the smartphone battery
level every time a user makes a web request. Code 4.3.1 shows the simplicity of the
code to be added. The obtained battery level is then passed to the HTTP Header
Handler.

public float getBatteryLevel() {

Intent batteryIntent = registerReceiver(null, new
IntentFilter(Intent.ACTION_BATTERY_CHANGED));
int level = batteryIntent.getIntExtra(BatteryManager.EXTRA_LEVEL,
-1);
int scale = batteryIntent.getIntExtra(BatteryManager.EXTRA_SCALE,
-1);

if(level == -1 || scale == -1) {

return 50.0f;
}
return ((float)level / (float)scale) * 100.0f;
}

52
 (a) Server GUI (b) Client GUI

Figure 4.2: The Graphical User Interface of the Smart Server and Smart Browser

53
 • Connection Type Listener. This listener’s main duty is to identify the type of network
the smartphone is connected to and pass it to the HTTP Header Handler. Code 4.3.1
shows how we implemented this component. The obtained network type is also passed
to the HTTP Header Handler to be included in the HTTP header.

ConnectivityManager conMan = (ConnectivityManager)

getSystemService(Context.CONNECTIVITY_SERVICE);
//mobile
State mobile = conMan.getNetworkInfo(0).getState();
//wifi
State wifi = conMan.getNetworkInfo(1).getState();
String connection = new String();
if (mobile == NetworkInfo.State.CONNECTED || mobile ==
NetworkInfo.State.CONNECTING) {
//mobile
connection = "mobile";
} else if (wifi == NetworkInfo.State.CONNECTED ||
wifi == NetworkInfo.State.CONNECTING) {
//wifi
connection = "wifi";
}

• HTTP Header Handler. This handler modifies the HTTP request’s header. It inserts
the battery level and the type of network into the optional fields of HTTP protocol.
Figure. 4.3 (a) shows how the header looks after the new fields are added to the
conventional HTTP header (Fig. 4.3 (b).) These snapshots are for our system, taken
from Wireshark to illustrate the modifications the system makes into the HTTP
packets. Figures 4.3 (a) and 4.3 (b) show the headers appearance before and after
the modifications are made, respectively.

4.3.2 Server Side Configuration

We implemented an HTTP server in Java. The server, located in our lab, is deployed
on a Dell desktop machine running Windows 7. For our configurations we added the (i)
HTTP Header Parser that extracts the information that the browser includes with the
HTTP request; (ii) Decision and Response Preparation unit, where the server applies the

54
 Figure 4.3: HTTP header modifications

specified policies. Accordingly, a proper web-page is prepared and sent back to the client
based on its current available resources.

4.3.3 System Parameters

Certain parameters are associated with this mobile browsing technique, and are either
end-user controlled ones or the smartphone’s:

1. Battery Level This parameter changes according to the amount of energy drained
from the battery.

2. Connection Type This parameter is user controlled. The user connects to a cellular
or Wi-Fi network based on availability. The assumptions in this case are that (1)
the cellular network is a metered plan, based on usage; and that (2) Wi-Fi network
is completely free and requires no charge whatsoever.

These two parameters are used to evaluate how effective our system is, as shown in sec-
tion.4.4.

55
4.4 Validation and Results
In this section, we validate our system by means of measurements. We compare its per-
formance with the conventional web-browsing techniques and measure the energy and
bandwidth consumption under different scenarios.
All the energy and bandwidth measurements were conducted using the test benches
described in Chapter 3. Our Smart Browser was installed on a Galaxy Nexus smartphone
running Android V4.2.1.. The results are presented and discussed in the following subsec-
tions.

4.4.1 Energy and Bandwidth Results

The first evaluation scenario is as follows. We ran the Smart Browser for 10 minutes over
a Wi-Fi network. During this period, a page request was being made every 30 seconds, for
a total of 20 page requests. We set our server to deliver three kinds of web content: (1) the
full version of the web site, (2) a webpage with only textual ads, (3) a webpage with no ads
at all. In this scenario, the server sends back the full version for the first three minutes,
then switches to send back the webpage with only textual ads for the next 3 minutes, and
finally, it sends back the ad-free webpages for the last 4 minutes.
To compare our system with current browsing techniques, we disabled our system’s
features and browsed the same website again. We started by browsing the website’s full
version and redid the same browsing experience; that is, we requested the webpage repeat-
edly for 10 minutes. Using conventional browsing, however, the only type of content to
be sent back was the full version of a website. Then, we applied the other extreme case,
which is ad-free webpages, and redid the same browsing experience. Figure 4.4. shows the
energy drained from the battery as a function of time for the four test cases (websites);
by considering only the available Battery Level, 2 to 12 % of the consumed energy can be
saved using SWB.
Our second scenario was repeating the first experiment over a 3G network. Hence, the
webpage adaptation decisions were made based only on the current Battery Level. Figure.
4.5 shows the results obtained for the test cases.
The last scenario considers both parameters, Battery Level and Connection Type. This
scenario shows the overall effectiveness of our system. Figure. 4.6 shows the overall per-
formance of the system. In this scenario, the first case used only normal browsing (i.e.,
browsing only the full version of a website, with no adaptation made) for 20 minutes in
total, (10 minutes over Wi-Fi and the other 10 minutes over a 3G network.) The second

56
 120 120

100 100
Ad-free version Ad-free version
Our system
Our sytem
Full version
Full version
80
80

Energy (mAmp-H)
Energy (mAmp-H)

60
60

40
40

20
20

0
0 0 100 200 300 400 500 600 700
0 100 200 300 400 500 600 700 Time (Sec)
Time (Sec)

(a) www.macleans.com (b) www.therecord.com

90 90

Ad-free version Ad-free version

80 80
Our system Our sytem
Full version Full version
70 70

60 60
Energy (mAmp-H)

Energy (mAmp-H)

50 50

40 40

30 30

20 20

10 10

0 0
0 100 200 300 400 500 600 0700 100 200 300 400 500 600 700
Time (Sec) Time (Sec)

(c) www.thestar.com (d) www.nytimes.com

Figure 4.4: Comparison of the energy consumption over a Wi-Fi connection

57
 120
120

100
100
Ad-free version
Ad-free version
Our sytem
Our sytem
Full version
Full version 80
80

Energy (mAmp-H)
Energy (mAmp-H)

60
60

40 40

20 20

0 0
0 100 200 300 400 500 600 700 0 100 200 300 400 500 600 700
Time (Sec) Time (Sec)

120 120
(a) www.macleans.com (b) www.therecord.com

100 100
Ad-free version Ad-free version
Our sytem Our sytem
Full version Full version
80 80
Energy (mAmp-H)

Energy (mAmp-H)

60 60

40 40

20 20

0 0
0 100 200 300 400 500 600 0700 100 200 300 400 500 600 700
Time (Sec) Time (Sec)

(c) www.thestar.com (d) www.nytimes.com

Figure 4.5: Comparison of the energy consumption over a 3G connection

58
 2500
3500
Energy (Joules)

2000

3000
1500

1000
2500
500

0
Energy (J)

2000 Normal Browsing (WiFi & 3G) SWB

Browsing technique

1500

1000

500

0
www.nytimes.com www.thestar.com www.therecord.com www.macleans.com
Normal Browsing (10 min WiFi & 10 min 3G) 2656.3572 2680.416 3051 3045
SWB (20 min) 2540.196 2318.796 2755 2754
Normal Browsing (20 min 3G) 2730.6 3223.584 3217 3188

Figure 4.6: Energy consumption comparison between SWB and normal browsing

59
case was browsing using SWB, also 20 minutes in total. We found that SWB outperforms
in this test case. The energy results in Fig. 4.6 can be used to estimate a complete dis-
charging cycle of the battery; that is, consecutive webpage requests are being made until
the battery dies. The battery of the Galaxy Nexus smartphone can supply energy up to
1,750 mAmp-H, or approximately 28,830 Joules. Based on this number and the obtained
results in Fig. 4.6, Table. 4.2 presents a summary of our results under the three browsing
cases: Case # 1, normal browsing, that is, browsing is done half of the time over a 3G
network and the other half it is done over a Wi-Fi network; Case # 2, browsing using
SWB for the whole time; Case # 3, normal browsing over 3G network only. Applying such
scenarios, our results show that SWB can extend battery life up to an hour, compared to
Case # 3.

Table 4.2: Summary of the estimated battery life under three different cases
Battery Life Estimation in Hours
Test Cases Case # 1 SWB Case # 3 ∆ (SWB - Case # 3) ∆ (SWB - Case # 1)
nytimes 2.91 3.05 2.83 0.21 0.13
thestar 2.89 3.34 2.40 0.93 0.45
therecord 2.53 2.81 2.40 0.40 0.27
macleans 2.54 2.81 2.43 0.38 0.26

The bandwidth required to download webpages was measured in the test cases that
required a Wi-Fi connection. We could not measure the bandwidth when a 3G network
was used because Wireshark cannot be used for this purpose without extra tools. Also, we
did not want to use the apps that capture the traffic of the phone, due to the fact that their
operation will cause extra energy consumption, and thus, distort the energy measurements.
Therefore, we only measured the consumed bandwidth over a Wi-Fi network. Figure. 4.7
shows the results obtained form running our system for ten minutes. In this period of
time, a total of 20 webpage requests were made. The results confirms that SWB can
save wireless bandwidth up to ∼ 44% compared to normal web browsing. Regarding the
bandwidth consumed when a 3G network is used, we believe it will be the same as the case
of Wi-Fi network since the content to be downloaded is the same.
We made empirical calculations to estimate how much money a user saves using SWB.
According to [18], users spend approximately 53 minutes a day on mobile browsing. If
we assume that he/ she spends a minute on each webpage, the total number of requested
webpages would be 53. Based on our results in Fig. 4.7, ∼ 7 to 70 MBytes per day can be
saved using SWB under this scenario. That number accumulates to ∼ 210 to 2100 Mbyte
per month. Based on some metered data plans in US (usage-based pricing plans) reported

60
 70

60

50
B.W (Mbytes)

40

30

20

10

0
www.therecord.com www.nytimes.com www.macleans.com www.thestar.com
Full version 60.5 19.51 19.75 16.69
Our system 33.8 15.8 14.36 13.87

Figure 4.7: Bandwidth measurements

61
in [60], ∼ $ 11.2 to 336 would be saved using our system.

62
Chapter 5

Conclusion

This thesis has reported work done on the impact of the existence of ads in mobile appli-
cations and in webpage from energy and bandwidth “monthly data usage” prospectives.
Considerable amount of literature for the researchers and developers regarding the above
issues are provided in Chapter. 2. It is concluded that different measurement studies
have proven the adverse effect of ads on smartphone battery life and monthly data usage.
However, having free contents or free apps is no way possible without the deployment of
ads. Therefore, we surveyed the current solutions that attempt to keep the ads and at the
same time alleviate their impact, either by considering only the energy issue or only the
bandwidth issue, or both issues together.
In Chapter. 3, we focused on the energy and bandwidth impact of the existence of
ads in webpages and noticed many interesting observations. We investigated the energy
consumption due to ads in five well-known websites, and found that ads can consume ∼
3.5 to 12 Joules over Wi-Fi and 3G networks, that is ∼ 6 to 18 % of the total energy of
web browsing.
Then, we proposed a framework for mobile browsing that adapts the web-pages deliv-
ered to a smartphone, based on the smartphone’s current battery level and the network
type. The adaptation of the web content comes in the form of controlling the amount of
ads to be displayed on the web-page. Moreover, our system targets: (i) extending smart-
phone battery life; and (ii) preserving the bandwidth needed to download the webpages,
while balancing the satisfaction of the publishers of webpages as well as the end users.
The architecture and implementation of the system is explained in detail, in Chapter. 4.
By means of experimental validation, we showed that our system can extend smartphone
battery life by up to ∼ 25% and save wireless bandwidth up to ∼ 45%.

63
 In the future, the test bench of the evaluated websites can be further extended. Instead
of evaluating only five websites, more can be covered. We will also consider fully automating
the operation of the server side, in particular, the webpage adaptation process. Currently,
the amount of ads embedded in webpages are manually controlled. For this thesis, we
manually prepared a number of versions of the same webpage, in each webpage allowing
certain amount of ads. In future, we will work on automating this process, which will
require artificial intelligence. Machine learning can be applied to detect the ads on the
full-version webpages, and accordingly, the amount of ads can be controlled based on the
received current resources of the smartphone.

64
References

[1] Shark for root at google play. https://play.google.com/store/apps/details?id=

lv.n3o.shark&hl=en. [Online; accessed 30-Mar-2013].
[2] Bell hspa/umts. http://mobilesyrup.com/2010/12/12/
review-bell-novatel-wireless-u547-turbo-stick-42-mbps/, 2009.
[3] Mobile data traffic surpasses voice. http://www.ericsson.com/thecompany/press/
releases/2010/03/1396928, 2010.
[4] http://www.itu.org/, 2012.
[5] Alexa, the web information company. http://www.alexa.com/, 2013.
[6] The world in 2013: Ict facts and figures. http://www.itu.int/ITU-D/ict/facts/
material/ICTFactsFigures2013.pdf, 2013.
[7] A. Abogharaf, R. Palit, K. Naik, and A. Singh. A methodology for energy performance
testing of smartphone applications. In Automation of Software Test (AST), 2012 7th
International Workshop on, pages 110–116, 2012.
[8] Mobile Marketing Association et al. Mma code for re-
sponsible mobile marketing. http://www.barnesgraham.com/
resource-centre/industry-news-and-articles/article-89/
mma-code-for-responsible-moble-marketing, 2003.
[9] Howard Baldwin. Wireless bandwidth: Are we running out of room?
http://www.computerworld.com/s/article/9223670/Wireless_bandwidth_
Are_we_running_out_of_room_, 2012.
[10] Michael Butkiewicz, Harsha V. Madhyastha, and Vyas Sekar. Understanding website
complexity: measurements, metrics, and implications. In Proceedings of the 2011
ACM SIGCOMM conference on Internet measurement conference. ACM, 2011.

65
[11] Aaron Carroll and Gernot Heiser. An analysis of power consumption in a smart-
phone. In Proceedings of the 2010 USENIX conference on USENIX annual technical
conference, page 21, Berkeley, CA, USA, 2010. USENIX Association.

[12] Sindhura Chava, Rachid Ennaji, Jay Chen, and Lakshminarayanan Subramanian.
Cost-aware mobile web browsing. Pervasive Computing, IEEE, 11(3):34–42, 2012.

[13] N Chowdhury. A survey of search advertising, 2007.

[14] J.Y. Chung, Y. Choi, B. Park, and J.W.K. Hong. Measurement analysis of mobile
traffic in enterprise networks. In Network Operations and Management Symposium
(APNOMS), 2011 13th Asia-Pacific, pages 1–4. IEEE, 2011.

[15] Gerald Combs. Tshark-the wireshark network analyser. URL http://www. wireshark.
org.

[16] Comscore. comScore Reports September 2012 U.S. Mobile Subscriber Market
Share. http://www.comscore.com/Insights/Press_Releases/2012/11/comScore_
Reports_September_2012_U.S._Mobile_Subscriber_Market_Share/, 2012. [On-
line; accessed 23-Seb-2012].

[17] Marco Cristo, Berthier Ribeiro-Neto, Paulo B Golgher, and Edleno Silva de Moura.
Search advertising. In Soft Computing in Web Information Retrieval, pages 259–285.
Springer, 2006.

[18] Mian Dong and Lin Zhong. Chameleon: a color-adaptive web browser for mobile oled
displays. Mobile Computing, IEEE Transactions on, 11(5):724–738, 2012.

[19] Marwa Elfattah, Aliaa AA Youssif, and Ebada Sarhan. Handsets malware threats and
facing techniques. CoRR, 2012.

[20] J. Erman, A. Gerber, and S. Sen. Http in the home: It is not just about pcs. ACM
SIGCOMM Computer Communication Review, 41(1):90–95, 2011.

[21] H. Falaki, D. Lymberopoulos, R. Mahajan, S. Kandula, and D. Estrin. A first look

at traffic on smartphones. In Proceedings of the 10th annual conference on Internet
measurement, pages 281–287. ACM, 2010.

[22] Z. Fang, Y. Yang, F.M. Deng, and J. Cai. Research advances in mobile advertising
areas. Applied Mechanics and Materials, 248:555–558, 2013.

66
[23] Michael C Grace, Wu Zhou, Xuxian Jiang, and Ahmad-Reza Sadeghi. Unsafe exposure
analysis of mobile in-app advertisements. In Proceedings of the fifth ACM conference
on Security and Privacy in Wireless and Mobile Networks, pages 101–112. ACM, 2012.

[24] S. Guha, A. Reznichenko, K. Tang, H. Haddadi, and P. Francis. Serving ads from
localhost for performance, privacy, and profit. In Proceedings of the 8th Workshop on
Hot Topics in Networks (HotNets 09), New York, NY, 2009.

[25] Kelly Hodgkins. Angry birds android to top $1 million per

month in ad revenue. http://www.intomobile.com/2010/12/03/
angry-birds-android-1-million-ad-revenue/, 2010.

[26] J. Hoehl and C. Lewis. Mobile web on the desktop: simpler web browsing. In The
proceedings of the 13th international ACM SIGACCESS conference on Computers and
accessibility, pages 263–264. ACM, 2011.

[27] Breg Insight. Mobile advertising and marketing. http://www.berginsight.com/

News.aspx, 2011.

[28] V Jacobsen, Craig Leres, and Steven McCanne. Tcpdump/libpcap, 2005.

[29] Azeem J. Khan, V. Subbaraju, Archan Misra, and Srinivasan Seshan. Demo: context
driven advertisement optimizer. In Proceedings of the 10th international conference
on Mobile systems, applications, and services, MobiSys ’12, pages 487–488, New York,
NY, USA, 2012. ACM.

[30] Azeem J. Khan, V. Subbaraju, Archan Misra, and Srinivasan Seshan. Mitigating the
true cost of advertisement-supported ”free” mobile applications. In Proceedings of the
Twelfth Workshop on Mobile Computing Systems & Applications, HotMobile ’12,
pages 1:1–1:6, New York, NY, USA, 2012. ACM.

[31] Marc Langheinrich, Atsuyoshi Nakamura, Naoki Abe, Tomonari Kamba, and
Yoshiyuki Koseki. Unintrusive customization techniques for web advertising. Com-
puter Networks, pages 1259–1272, 1999.

[32] S.W. Lee, J.S. Park, H.S. Lee, and M.S. Kim. A study on smart-phone traffic analysis.
In Network Operations and Management Symposium (APNOMS), 2011 13th Asia-
Pacific, pages 1–7. IEEE, 2011.

[33] Ilias Leontiadis, Christos Efstratiou, Marco Picone, and Cecilia Mascolo. Don’t kill my
ads!: balancing privacy in an ad-supported mobile application market. In Proceedings

67
 of the Twelfth Workshop on Mobile Computing Systems & Applications, page 2. ACM,
2012.

[34] Sean Ludwig. Vasona networks raises 12m to fix our mobile data congestion issues.
http://venturebeat.com/2013/03/18/vasona-networks-funding/, 2013.

[35] G. Maier, F. Schneider, and A. Feldmann. A first look at mobile hand-held device
traffic. In Passive and Active Measurement, pages 161–170. Springer, 2010.

[36] Amy Mitchell. THE EXPLOSION IN MOBILE AUDIENCES AND A CLOSE LOOK
AT WHAT IT MEANS FOR NEWS. http://www.journalism.org/analysis_
report/future_mobile_news, 2012. [Online; accessed 23-Mar-2012].

[37] Radhika Mittal, Aman Kansal, and Ranveer Chandra. Empowering developers to
estimate app energy consumption. In Proceedings of the 18th annual international
conference on Mobile computing and networking, pages 317–328. ACM, 2012.

[38] mobiThinking. Global mobile statistics 2012 Part A: Mobile subscribers; handset mar-
ket share; mobile operators. http://mobithinking.com/mobile-marketing-tools/
latest-mobile-stats/a#mobilephonepenetration/, 2012. [Online; accessed 23-
Seb-2012].

[39] Prashanth Mohan, Suman Nath, and Oriana Riva. Prefetching mobile ads: Can
advertising systems afford it? In Proceedings of the 8th European Conference on
Computer Systems (EuroSys’ 13), 2013.

[40] Veelasha Moonsamy, Moutaz Alazab, and Lynn Batten. Towards an understanding
of the impact of advertising on data leaks. Int. J. Secur. Netw., pages 181–193, 2012.

[41] Rajesh Palit, Renuka Arya, Kshirasagar Naik, and Ajit Singh. Selection and execution
of user level test cases for energy cost evaluation of smartphones. In Proceedings of
the 6th International Workshop on Automation of Software Test, pages 84–90, 2011.

[42] A. Pathak, Y.C. Hu, and M. Zhang. Where is the energy spent inside my app?: fine
grained energy accounting on smartphones with eprof. In Proceedings of the 7th ACM
european conference on Computer Systems, pages 29–42. ACM, 2012.

[43] A. Pathak, Y.C. Hu, M. Zhang, P. Bahl, and Y.M. Wang. Fine-grained power modeling
for smartphones using system call tracing. In Proceedings of the sixth conference on
Computer systems, pages 153–168. ACM, 2011.

68
[44] Paul Pearce, Adrienne Porter Felt, Gabriel Nunez, and David Wagner. Addroid:
privilege separation for applications and advertisers in android. In Proceedings of
the 7th ACM Symposium on Information, Computer and Communications Security.
ACM, 2012.

[45] G.P. Perrucci, F.H.P. Fitzek, G. Sasso, W. Kellerer, and J. Widmer. On the impact
of 2g and 3g network usage for mobile phones’ battery life. In Wireless Conference,
2009. EW 2009. European, pages 255 –259, 2009.

[46] Privoxy. Privoxy. URL http://www.privoxy.org/.

[47] I. Prochkova, V. Singh, and J.K. Nurminen. Energy cost of advertisements in mobile
games on the android platform. In Next Generation Mobile Applications, Services and
Technologies (NGMAST), 2012 6th International Conference on, pages 147 –152, Sep.
2012.

[48] Feng Qian, Zhaoguang Wang, Alexandre Gerber, Zhuoqing Morley Mao, Subhabrata
Sen, and Oliver Spatscheck. Characterizing radio resource allocation for 3g networks.
In Proceedings of the 10th ACM SIGCOMM conference on Internet measurement,
pages 137–150. ACM, 2010.

[49] Ashwin Rao, Justine Sherry, Arnaud Legout, Arvind Krishnamurthy, Walid Dabbous,
and David Choffnes. Meddle: middleboxes for increased transparency and control
of mobile traffic. In Proceedings of the 2012 ACM conference on CoNEXT student
workshop, pages 65–66. ACM, 2012.

[50] Grischa Schmiedl, Markus Seidl, and Klaus Temper. Mobile phone web browsing: a
study on usage and usability of the mobile web. In Proceedings of the 11th Interna-
tional Conference on Human-Computer Interaction with Mobile Devices and Services,
page 70. ACM, 2009.

[51] Aruna Seneviratne, Kanchana Thilakarathna, Suranga Seneviratne, Mohamed Ali

Kaafar, and Prasant Mohapatra. Reconciling bitter rivals: Towards privacy-aware
and bandwidth efficient mobile ads delivery networks. In Communication Systems
and Networks (COMSNETS), 2013 Fifth International Conference on, pages 1–10.
IEEE, 2013.

[52] Venkatesh Shankar and Marie Hollinger. Online and mobile advertising: current sce-
nario, emerging trends, and future directions. Marketing Science Institute, pages
07–206, 2007.

69
[53] Maad Shatnawi and Nader Mohamed. Statistical techniques for online personalized
advertising: a survey. In Proceedings of the 27th Annual ACM Symposium on Applied
Computing, pages 680–687. ACM, 2012.

[54] Shashi Shekhar, Michael Dietz, and Dan S Wallach. Adsplit: Separating smartphone
advertising from applications. CoRR, abs/1202.4030, 2012.

[55] RJG Simons and A. Pras. The hidden energy cost of web advertising. 2010.

[56] Monsoon Solutions. Power monitor. http://www.msoon.com/LabEquipment/

PowerMonitor/.

[57] N. Thiagarajan, G. Aggarwal, A. Nicoara, D. Boneh, and J.P. Singh. Who killed my
battery: analyzing mobile browser energy consumption. In Proceedings of the 21st
international conference on World Wide Web. ACM, 2012.

[58] N. Vallina-Rodriguez and J. Crowcroft. Energy management techniques in modern

mobile handsets. Communications Surveys Tutorials, IEEE, 15(1):179–198, 2013.

[59] N. Vallina-Rodriguez, J. Shah, A. Finamore, Y. Grunenberger, H. Haddadi, and

J. Crowcroft. Breaking for commercials: Characterizing mobile advertising. pages
343–356, 2012.

[60] J. van den Brande and A. Pras. The costs of web advertisements while mobile brows-
ing. In Information and Communication Technologies, pages 412–422. Springer, 2012.

[61] Jarred Walton. Browser face-off: Battery life explored. http://www.anandtech.com/

show/2834/, 2009. [Online; accessed 19-Seb-2012].

[62] Yong Wang, Daniel Burgener, Aleksandar Kuzmanovic, and Gabriel Maciá-Fernández.
Understanding the network and user-targeting properties of web advertising networks.
In Distributed Computing Systems (ICDCS), 2011 31st International Conference on,
pages 613–622. IEEE, 2011.

[63] Z. Wang, F.X. Lin, L. Zhong, and M. Chishtie. Why are web browsers slow on
smartphones? In Proceedings of the 12th Workshop on Mobile Computing Systems
and Applications, pages 91–96. ACM, 2011.

[64] Nick Wood. Mobile data traffic growth 10 times faster than fixed over next five years.
http://www.totaltele.com/view.aspx?ID=448681, 2009. [Online; accessed 19-Seb-
2012].

70
[65] Chanmin Yoon, Dongwon Kim, Wonwoo Jung, Chulkoo Kang, and Hojung Cha. App-
scope: Application energy metering framework for android smartphone using kernel
activity monitoring. 2012.

[66] L. Zhang, D. Gupta, and P. Mohapatra. How expensive are free smartphone apps?
ACM SIGMOBILE Mobile Computing and Communications Review, 16(3):21–32,
2012.

[67] L. Zhang, B. Tiwana, Z. Qian, Z. Wang, R.P. Dick, Z.M. Mao, and L. Yang. Ac-
curate online power estimation and automatic battery behavior based power model
generation for smartphones. In Proceedings of the eighth IEEE/ACM/IFIP interna-
tional conference on Hardware/software codesign and system synthesis, pages 105–114.
ACM, 2010.

[68] Y. Zhu, A. Srikanth, J. Leng, and V. Reddi. Exploiting webpage characteristics for
energy-efficient mobile web browsing. Computer Architecture Letters, 2012.

[69] Yuhao Zhu and Vijay Janapa Reddi. High-performance and energy-efficient mobile
web browsing on big/little systems. Network, 2012.

71