Mini Project Report On Android Battery Saver System: Department of Information Technology

DEPARTMENT OF INFORMATION TECHNOLOGY
Mini Project Report
On
Android Battery Saver System
INTERNAL GUIDE
DR.L. SRIDHARA RAO
ASSOCIATE PROFESSOR
SUBMITTED BY
CH. SAI VENKAT
(18671A1209)
INTERNAL GUIDE PROJECT CO-ORDINATOR
HOD-IT
Literature Survey:
The idea of using prediction with battery saving applications originates from examining how
prediction was used in other applications. In many cases, prediction was used to reduce wait
times and preserve battery life. However, these were part of larger projects, where battery life
was not the primary objective. The repeated mention of battery life in many articles was a
clear indicator of its importance to mobile applications, leading to the combination of both
concepts.
2.1. Battery Life

One of the most prominent issues with smartphones is battery life, with 37% of user stating it
is their biggest problem [2]. As more powerful smartphones are developed, concerns with
battery life increase. Users should be able to utilize their device as they wish for a full day
before a recharge is required. However, this is often not the case, leading to a change in our
activities to preserve battery life. This concern was not evident on desktop computers, as they
have a constant source of power. With the rise of smartphone services, it is one of the biggest
challenges faced by developers. While research into more efficient batteries is possible,
another area of research focuses on improving the efficiency of applications.
The solution proposed by Pathak et al. [3] is to create a compile-time dataflow analysis
solution that can detect no-sleep energy bugs. Dataflow analysis is defined as a set of
techniques that analyze the effects of program properties throughout a given program,
managed within a control flow graph. Their solution focuses on the sections where
smartphone component power is managed. If all of those sections have end points that turn
off the components, the program is free of no-sleep bugs. To test their application, they ran
their analysis on 86 different android applications. In addition to the 12 known energy bugs
detected, 30 new types of bugs were discovered. Pathak et al. note that this area of research is
relatively new, and they are making the first advances towards understanding and detecting
no sleep bugs.
Focusing on a specific type of application, Xu et al. [4] examined the built-in email clients of
Windows Phone and Android to determine areas of improvement. Windows phone uses
Microsoft Exchange, while Android uses Gmail. Gmail is one of the most popular 12
applications on a mobile device, with over 50 million users per month [5]. With such a large
user base, it is important that Gmail and other email applications are optimized for
functionality, accessibility, and power consumption. Unfortunately, functionality and power
consumption can be contradicting. Functionality requires the application to be constantly
checking for new messages, but continually syncing is extremely resource intensive. Finding
a balance between these two concerns is not only limited to email, and can be practical for
other applications.
The energy cost of receiving an email increases when the inbox is larger, and can be
attributed to the time is takes to update the metadata. The proposed solution is to partition the
inbox into two parts: a small inbox for recently received emails, and a large one for the
remainder. As most new messages will interact with recent messages, there is no need to
search through old emails. Xu et al. implemented these changes and proceeded to observe the
change in energy consumption. Their findings indicated an average energy reduction of
49.9%
While users generally dislike advertisements distracting them from a webpage, the study of
Albasir et al. [6] gives users another reason to detest them. The energy consumption of
advertisements was measured by examining a number of news websites under two
conditions. The first condition used the built-in web browser on the device to access websites,
measuring the amount of energy and bandwidth consumed. In the second scenario, the same
websites were revisited on a different browser designed to display the webpage without ad
traffic. The results indicated that advertisements can take up to 50% of the traffic required to
load the page. In addition, the energy consumption of ad generation represented
approximately 6 – 18% of the total energy from web browsing. While this study only
examined a small number of news websites, it highlights an opportunity to improve battery
life for mobile users.
2.2. Prediction
Many people may be familiar with smartphone prediction due to its use on their keyboard.
However, the applications of prediction extend well beyond such a simple use. The primary
benefit of prediction is speed, and a reduced wait time is always welcomed by users. Higgins
et al. [14] examine prediction on smartphones and illustrate how and when it can be used.
They have designed an API that leverages the uncertainty level in their prediction before
making a decision. The API can use three different methods when determining the predictive
error rate, each with a different drawback. Their API is used and tested on two applications: a
network selection, and a speech recognition application. The network selection application is
used to determine if the smartphone should transmit data over cellular, WiFi or both
mediums, 20 based on latency, bandwidth, dwell time, and energy usage. In the speech
recognition system, the API is used to determine if the recognition should be performed on
the device, the remote server, or both, based on latency, bandwidth, dwell time, application
compute time, and energy usage. For both applications, a scenario to use both options exists
because Higgins et al. consider the benefits of redundant strategies, as they understand the
uncertainty of predictive approaches. Their results for the network selection application
resulted in a 21% reduced wait time over cellular-only strategies, and 44% for Wi-Fi
preferred and adaptive strategies. For speech recognition, there were varying results
depending on the energy usage. Redundant strategies are still beneficial for low to mid-
energy cost scenarios, but prove to be too energy consuming for high-cost scenarios. In
addition, their API reduced recognition delay in the no-cost energy scenario by 23%. While
their study showcases the benefits of prediction, it fails to illustrate where and how it can be
used. Other research into the topic provides better examples of its practicality.
One notable example of prediction use is in mobile exercise applications. Kotsev et al. [15]
have begun using prediction to determine when users will exercise. . By predicting a pattern,
researchers can develop a better understand of what motivates users to exercise, allowing
them to create better tools to increase motivation. Their work begins with analyzing a dataset
generated from over 10000 users, with the goal of identifying as many different factors as
possible. The dataset provided information such as the type of activity performed, the country
the user is from, their social connectivity, when they exercise, and how long their exercise
for. While their research is inconclusive, they identify the top 10 features that can be used to
predict future behaviour, which are runs per week, mean runs per week, max runs per week,
min runs per week, average runs per week, 2 elapsed hours, min distance, min elapsed hours,
mean speed, and max speed.
Bulut et al. [16] have also utilized prediction in a unique way, creating a crowdsourced line
wait-time monitoring system with smartphones. Its implementation at grocery stores, DMV’s
and banks would allow users to make informed choices in time-sensitive environments.
Known as LineKing, it has been tested at a coffee shop at the State University of New York
at Buffalo. Customers who enter the shop will establish a connection with the service, where
any connection lasting longer than 2 minutes but less than 20 is deemed a customer. The
wait-time calculation is completed on the server side of the application, taking the time of the
day, the day of the week, and seasonality into account. The estimated wait times are accurate
within 2-3 minutes.
LineKing is comprised of two components, a client-side and a server-side. The client side is
comprised of three subcomponents: phone-state-receiver, wait-time-detection, and
datauploader. The phone-state-receiver is comprised of a variety of receivers registered to
monitor various events for the application. The most notable event is the device entering and
exiting the shop. The wait-time-detection component can use either location sensing or WiFi
sensing to detect the user’s presence at the shop. In order to preserve battery life, the
component begins monitoring the device under two conditions: if the user opens the
application to check the wait time or if the user is physically close to the shop. Once a
condition is triggered, the application begins to monitor the user’s location. For location
sensing, if the user is within a specific range of the shop, the application will set a proximity
alert to register the timestamp of entering the shop. If they are outside of the specified range,
the application will estimate the arrival time of the user, and recheck their location at that
time. If the user does not travel towards the shop after a certain amount of time, the
monitoring will cease. In the Wi-Fi sensing approach, the application will monitor Wi-Fi
beacons periodically to determine when the user enters and leaves the shop. Their calculation
in this approach takes into account the time delay of the scanning period. Lastly, the data-
uploader component is responsible for uploading the wait-times to the estimation system.
Rattagan et al. [19] examine prediction and battery life together, evaluating online power
estimations from battery monitoring units. They discuss the current methods of online and
offline monitors, indicating the pros and cons of each. While online methods are more
feasible and scalable, their results have a higher error rate due to three problems that are not
taken into consideration: battery capacity degradation, asynchronous power consumption
behaviour, and the effect of state of charge difference in hardware training. The battery
capacity of a device will decrease after usage, while online methods use the original battery
capacity value without taking this into account.
Peltonen et al. [20] attempt to construct energy models of smartphone usage through
crowdsourcing, whereas most research focused on a singular device or system. They use a
subset of data collected from Carat [11], a collaborative energy diagnostic system. The data
contains 11.2 millions samples from approximately 150 000 Android devices. The dataset
also provides energy rates that can be used to calculate battery consumption. Within this
dataset, they identify 13 different context factors, 5 of which are user-changeable settings and
8 are subsystem state information.
Anguita et al. [21] attempt to use machine learning to overcome battery limitations. They
propose sensors can be used to predict the actions of the user. They use an existing machine
learning framework and modify it to meet the resource constraints of a smartphone. Their
implementation is then validated in a series of test cases where the framework predicts
whether the user is walking, walking upstairs, walking downstairs, sitting, standing, or
laying. While the test cases are outside the scope of traditional application monitoring, they
illustrate the usage of machine learning is feasible for mobile devices, and the resource
requirements can be reduced.
2.3. Battery Application Review

The purpose of reviewing the current battery management applications was to determine the
functionality that is currently offered, and to check that prediction is not an established
approach. DU Battery Saver [22], Battery Doctor [23], and Battery Saver applications were
retrieved from the Google Play Store, using the search tag battery saver and battery life. They
were the highest rated applications, top results from searches, and had a minimum rating of
4.5/5.0. In addition, DU Battery Saver and Battery Doctor have over 8 million downloads
respectively as of May 2018. Battery Saver is no longer available in the Google Play Store as
of May 2018, and the number of downloads was not recorded.

Mini Project Report On Android Battery Saver System: Department of Information Technology

Uploaded by

Copyright:

Available Formats

Mini Project Report On Android Battery Saver System: Department of Information Technology

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Mini Project Report On Android Battery Saver System: Department of Information Technology

Uploaded by

Copyright:

Available Formats

DEPARTMENT OF INFORMATION TECHNOLOGY

Mini Project Report

Android Battery Saver System

INTERNAL GUIDE PROJECT CO-ORDINATOR

2.1. Battery Life

2.3. Battery Application Review

You might also like