GreenHub Farmer: Real-world data fo rAndroid Energy Mining

Green Hub Farmer: Real-world data for Android Energy Mining

INTRODUCTION

MOBILE devices have become one of our most usedgadgets, with their battery life becoming of a high con-cern for users. In fact, battery life is traditionally known to beone of the major factors influencing consumer satisfaction [1],and was just recently pointed out, on top of usability, storageand durability, as the most important feature for smartphoneowners [2]. Battery life is such a growing concern that ithas been shown that 9 of 10 users suffer from low batteryanxiety [3], and is under discussion as a potential clinicalcondition:nomophobia, the fear of being without your mobilephone, in the Diagnostic and Statistical Manual of MentalDisorders [4].

https://codeshoppy.com/php-projects-titles-topics.html

On the other end, developers are also very concerned withtheir application’s battery life, as excessive battery consump-tion is one of the most common causes for bad app reviewsin app stores [5], [6]. In fact, developers are aware of thebattery consumption problem, and many times seek help insolving this, even if they rarely receive adequate advice [7]–[9]. Mobile brands have actually caught sight of this issue andhave tried to offer help by publishing developer guides aimedat extending battery life123.Reducing the energy that is consumed by mobile devicesis also an important problem from a sustainability point ofview. Indeed, the billions of phones that are in use these dayshave a global massive environmental footprint, and our digitalconsumption (which includes but is not limited to mobiledevice usage) is bound to have a greater impact on globalwarming than the aviation industry [10].Despite its importance, optimizing, or even analyzing en-ergy consumption for mobile devices is a difficult and labor-intensive task for both users and/or developers.For once, developers are using different monitoringtools [11]–[13] according to specific needs which often re-sults in a non systematized procedure and context specificfindings [13]–[15]. 

Monitoring the energy consumed by anapplication often results in extensive tests under several differ-ent scenarios and devices [16]–[18], both very time consumingand potentially requiring large initial investments. Indeed, evenconsidering Android alone, this is already a heavily heteroge-neous environment, as there exists thousands of potential com-binations among manufacturers, devices, operating systems,features, hardware components and apps, for example.For users, understanding the energy consumption of theirdevices is an even harder exercise. For once, their knowledgeregarding the hardware behavior is limited to their own de-vices, and without the proper tools and skills they cannot com-pare the energy behavior of their apps with others. Moreover,

different usage contexts of the same app (e.g., within differentOS versions and with different hardware components switchedon) results in different energy behaviors, and this has to betaken into account when performing any comparison.In this paper, we present a large dataset which is repre-sentative of real-world day-to-day usage of Android devices.Our dataset entries include information such as active sen-sors, memory usage, battery voltage and temperature, runningapplications, model and manufacturer, network details, etc,.This raw data was obtained by continuous crowd-sourcingthrough a mobile application. It is worth noting that all ourdata is publicly available, while maintaining the anonymityand privacy of all its users. Indeed, it is impossible to associateany data with the user who originated it. Thus far, our datasetincludes unique 12 million+ samples, from 900+ differentbrands and 5,000+ models, across 160 countries.This dataset was gathered within the GreenHub initiative4,acollaborative approach to energy consumption analysis withinAndroid. 

Our vision is to use the gathered data on the usageof mobile devices and application execution to help analyzeand identify opportunities to optimize energy consumption inAndroid devices, both for developers and users. Indeed, weexpect that useful information can be mined from the datasetas to help influence users in adopting more energy efficientbehaviors and to provide developers with indications of howefficient their application is and how it compares to others.In the case of developers, this is expected to triggerfurther analyses which are beyond the dataset itself. Thesemay explore the potential energy gains that have, e.g., beenproposed in the context of location services [19], contrast [20],color scheme [20], [21], data structure [22]–[25], programminglanguage [26]–[29], network usage [15], and API [17] usage.The remainder of this paper will describe: the developedinfrastructure (Section II) for the data collection, dataset,and a data-query prototyper; possible research directions fordevelopers and users (Section III); and finally the conclusionsof this paper (Section IV).