“Tour de France” data for the improvement of energy consumption in devices powered by limited energy sources

In this article, we propose a dataset about the energy consumption of mobile devices that was collected as part of a “Tour de France” with an electrical wheelchair. Part of these data was used to propose a mathematical model based on an experimental methodology of the energy consumed in mobile devices. Based on these precise measurements in a real environment, we have elaborated predictive models of energy consumption. The objective of this paper is to make accessible the data related to these publications to other researchers in several fields of science (computer science, telecommunications, meteorological science, artificial intelligence, statistics...). To our knowledge, this is the first publication of a dataset recording real world energy consumption data in mobile devices.


Value of the Data
• The presented experimental data are useful to measure several metrics related to the energy variation in a mobile device.• The data are beneficial for all the scientists who are exploring energy consumption in devices powered by limited energy sources as a knowledge resource.• The data can be used to analyze the load of the Central Processing Unit (CPU) and the Graphics Processing Unit (GPU) as well as to make a more detailed follow-up of the energy variation on other type of mobile environment.• The provided data can be used for further analysis in order to define models to improve energy handling in mobile devices, to define other types of scenarios given the large size of the data measured during the collection, or to optimize our methodology (based on part of the dataset [3,4] ).

Data Description
The presented research is part of the development of mathematical models, for modeling and evaluating the energy cost in mobile environments.
Our study focused on the Android system (version 4.4.4(KitKat)).The main objective of the data collection was to model the energy consumption of a particular application running on a mobile device.The data collection made possible to develop a specific mathematical model of energy consumption for monitoring the energy consumption of playing a local video, remote video streaming as well as in the case of navigating with a Global Positioning System (GPS) by acting on the following parameters: This analyse is focused on the development and measurement of energy costs.Our work focused on monitoring energy consumption based on several experimental scenarios.
Scientific data was collected on the basis of several scenarios in different locations.These measures were carried out in a real environment and confronted with the consumption measures made beforehand in a controlled environment.
Collecting and analyzing massive data in a real environment and diversifying it in a reasonable time creates a scientific challenge.The objective of the carried out research is to model the energy consumption of a particular application running on a mobile device.
In the rest of our paper, we will base ourselves on two types of case studies: • The disconnected mode in which all the measurements were carried out in the presence of only Global System for Mobile Communications (GSM).• Connected mode where all connection techniques have been activated (3G, 4G, Wi-Fi, GPS, etc.) depending on the chosen scenario.
The raw data are stored in text files (in CSV format).They are divided into four main groups depending on the chosen scenarios.Table 1 presents the different characteristics of the group of scenarios used for the evaluation.
The files are named as follows:

scenario_name_freq_GHz_[Option]_[departure_city]_[arrival_city]
Table 2 shows some examples of measurement files for all scenarios in disconnected mode.The list of all files relating to the disconnected mode is presented in Tables 3 and 4 .
For convenience, each measurement file is divided into two sub-files: The first one contains the list of measurements made during the experiment.The second one has the same name with the suffix "info" which gives information about active applications during the experiment.
The empty files "info files" (Zero line), means that the option which gives information was deactivated at the time of collection for these experiments.
The connected mode can be described as follows in Table 5 Tables 6 and 7 show the storage structure of the connected mode files.

Data preparation
The measurement file is represented in Tables 8 and 9 for the experiment that was carried out on 6 May 2016.The obtained data are stored in  • Screen Brightness: Screen brightness state, The average number of accesses per application and statistics on the sensors ( Table 10 ), the "info file" contains also information about the percentage of use of CPU resources, the average use of virtual memory per application, the maximum use of virtual memory per application, average real memory usage per application and maximum real memory usage per application, all these information are detailed in the Table 11 .The meaning of the applications not recognized are:

Meaning of the columns of
• GsmaService : The GSMA provides a range of services to assist the broader mobile industry.
• Launcher3 : Application launcher, it is the equivalent of the desktop on Android.
• Google Partner Configuration : Application that helps for run and configure applications in conjunction with Google products.• Play-Fi : Application for set up to streaming music.
• Cron Tasker Free : Application for Scheduling tasks simply by configuring the time.
• SmartcardService : The smart card service is a included system in the operating system in a root directory.it appears to the smart card provider [5] .• SVI Settings : A Switch Virtual Interfaces Represents a logical Layer 3 interface on a switch.

Methodology
This dataset was used to model the energy consumption of one or more specific applications in a mobile device [6,7] .On the one hand, it allows the monitoring of energy consumption according to the frequency and the total number of operations, on the other hand, to put into practice our proposed methodologies through several concrete case studies.
To identify the parameters of the model, our methodology must determine the type of correlation between the number of total operations and the elementary energy dissipated at a previously defined frequency.The total number of operations is calculated by summing the active clock cycles of each processor per measurement interval.This active clock cycle number is the load of the CPU multiplied by the measurement interval duration and by the CPU frequency.The dissipated energy during a measurement interval is the product between the battery power and the measurement interval duration.
The average consumption per clock cycle can be expressed as follows [3] : Load i , E dissipated E clock cycle shows the total energy per clock cycle, Load i the load of CPU i and E dissipated corresponds to the dissipated energy measured experimentally.Our methodology consists in establishing a regression of energy per measurement interval as a function of the total number of operations.The research presented proposes a methodology to evaluate and modeling the energy costs of mobile devices.
The proposed methodology is developed in three stages: • data collection, • data preparation, • energy consumption modelling.
The aim is to monitor energy consumption by acting on the following parameters: • frequency of the processors, • initial level of the battery, • energy dissipated by the clock cycle.

Tools data collection
The tools used to develop our methodology are [4] : 1. Trepn Profiler, a diagnostic tool for profiling performance and power consumption of Android applications.All tests of this experimentation were processed by version V6.2s.Trepn profiler provides information on system status, network status, graph performance, speed, processor frequency etc. 2. Cronoid, an automation tool that allows performing tasks on a regular basis (like KornShell).
It also enables automatic task running when the status of the terminal changes.The version used is Cronoid-3.5.1.3. CPU Frequency (V1.0.2), a tool that allows the user to change the CPU frequency setting to save energy or to achieve better performance.
The data collection stage of our methodology is based on the following steps: • Preparation of the test platform (CPU frequency management based on the governor) in order to have the rights to fix the frequencies of the CPU with the CPU Frequency tool.• The role of the Cronoid tool is to automate tasks in order to minimize interaction with the user.
In the case of variable frequencies, the same methodology can be applied by filtering the rows of the measurement table recorded by group of identical frequencies.To collect the data, a "Tour de France" was carried out using an electrical wheelchair to perform measurements in a diversified environment based on several scenarios.details related to the location of the measurements are detailed in Table 12 The tool used for all the experiments is a tablet (HP ProSlate 8) which was previously rooted.This choice was linked to the capacity of the battery which allows a large number of tests for We have presented a methodology to build a model of energy consumption of applications on mobile devices.This methodology begins with recording precise measurements of energy consumption when using the applications.The strong point of our methodology is in the technique of processing recorded data which leads to a precise model.The proposed solution can be used to define an optimal frequency for one or more applications in order to provide a better experience for users with reduced energy consumption.
For the implementation of our methodology, we selected the following parameters, obtained by Trepn Profiler: • Total load per CPU, • Memory usage, • CPU frequency, • Battery level, • Battery power.

Disconnected state
The measurements of the dissipated energy were carried out according to the number of total operations at a fixed frequency.Fig. 1 show example of experimentation by setting the processor frequencies at 1.5 GHz.

Connected state
The aim of this section is to follow behaviour the energy consumption variation by activating Wi-Fi.To stay in the same context, the remote video was identical to that used in disconnected mode.We determine the type of correlation between energy consumption and the number of total operations, then we compare the obtained result with those in offline mode.Fig. 2 shows an example of experimentation for a remote video in navigation mode with variable frequency (NFV).Raw data sources of this figure is stored in: NVF_default_freq_youtube_with_WiFi_Carcassonne_Narbonne.csv at Zenodo platform: https://zenodo.org/record/3739472 In this study, we have treated the different types of dependencies between energy dissipation and the frequency of several cases in disconnected mode (LVFF, LVVF) and connected mode (NFF, NVF).
The difference between the types of regressions can be explained in part by the variability of the signal strength received for GPS and Wi-Fi (for connected mode).On the other hand, the scenarios concerned in connected mode require the activation of other parameters, which increases the rate of inputs / outputs which are partly responsible for the additional energy cost.

Ethics Statement
Our work does not involve any use of human subjects nor animal experiments.The data are purely technological.

Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Table 1
Characteristics of the reference scenarios.

Table 2
Description of the disconnected scenario file names.

Table 3
List of the LVFF files in the dataset.

Table 4
list of the LVVF files in the dataset.

Table 5
Description of the connected scenario file names.

Table 9
: • Time [ms]: The interval of measurements in milliseconds (ms), • Memory Usage [Kb]: Memory used per measurement interval in Kilobyte [Kb],

Table 6
composition of NFF files in the dataset.

Table 8
Example of raw measured data (part 1).

Table 11
Example of system resource allocation.