Software Aging Metrics Analysis Tools

Warning: You may need an OPLUS phone to enable OneTrace collection.

Built on Python 3.8, the tool simulates and tests software aging on cell phones using the uiautomator2 library, collects metrics based on the OneTrace tool and the adb shell command, and supports rapid analysis and diagnosis of aging metrics based on scipy and pymannkendall. It collects metrics data based on the OneTrace tool and the adb shell command, and supports fast analysis and diagnosis of aging metrics based on the scipy`'](https://pypi.org/project/scipy/) and [pymannkendall`' libraries. The tool has been tested to function correctly on both Windows and Linux (MacOS is also theoretically supported).

Environment configuration

Python environment

• Install Python and its environment management tools. We recommend using miniconda, but you can also use venv if you prefer, or create a standalone environment in docker.

• Create a virtual environment based on Python 3.8. For example, in conda, you can create a virtual environment based on Python 3.8 via the

  conda create --name aging_test python=3.8

  to create a Python 3.8 environment named aging_test.

• Enter that environment and install the required third-party libraries. In conda, you need to first activate the new environment you just created:

  conda activate aging_test

  Go to the project directory:

  cd [path_to_AndroidAgingTest]

  and install the required pypi packages from requirement.txt:

  pip install --requirement requirements.txt

ADB

The Android Debug Bridge (adb) is a versatile command-line tool that allows a computer to communicate with a cell phone and run a series of shell commands. You can get the adb tool by downloading SDK Platform-Tools from Google, and it is recommended that you install We recommend installing version 34.0 and above. After installation, make sure adb is in the system environment variable.

FreakingAwesome-CLI (optional)

This project requires FreakingAwesome-CLI to convert the captured .frk files to database for further analysis.

Run the experiment

Initialize the device

• Make sure the device is ROOTed and the test app is installed.
• Reboot the phone.
• Make sure the device is connected to the computer via adb, i.e. the device can be queried via adb devices.

Parameter Configuration

The parameters required for the experiment are adjustable in config.yaml in the project directory, including:

The parameters required for the experiment can be adjusted in config.yaml in the project directory, including:

• app_list: the list of APP packages to be used for testing; to modify the APP list, append it in the following format:

  example_app_list: &example_app_list
      - com.example.app1
      - com.example.app2
      - ...

  And reference the above list as *example_app_list in test_app_list or stress_app_list.

• base: base parameters for the experiment run

  • serial: specifies the serial number of the selected device, which can be determined from the value in the first column of adb devices.
  • working_path: absolute path to the project.
  • trace_path: Path where the aging metrics data collected by the experiment will be placed.
  • test_app_list: list of APPs launched sequentially in a short test sequence
  • stress_app_list: list of APPs that cycle through aging operations during stress tests
  • stress_time_per_app: the time interval (in minutes) between runs of each app in a stress test round
  • trace_interval: interval (in seconds) between two adjacent runs of the tracing script to collect metrics
  • monkey_interval: Interval (in milliseconds) between each Monkey event generation in a stress test
  • jank_threshold: Sets OneTrace's threshold for determining jank events (in frames)

• run_t&s: experimental parameters for combining stress tests and short fixed sequences

  • duration: overall duration of the experiment (in hours)
  • t_xiaobai: duration (in minutes) of each fixed test sequence monitoring round
  • t_stress: duration of each stress test round (in minutes)

• run_stress: pure stress test experiment parameters

  • duration: overall duration of the experiment in hours
  • sample_interval: interval between two adjacent samples

Run the experiment

Make sure your phone is unlocked and on the desktop, and run run_t&s.py using your previously configured Python interpreter, depending on the type of experiment desired:

python run_t&s.py

or run_stress.py:

python run_stress.py

On first run, uiautomator2 may install some dependencies on your phone.

If it runs properly, you will be able to observe in:

• Fixed/random sequence runs on the phone, depending on the experimental parameters set.

  Tips: You may need to adjust the power plan during the run to ensure that the computer does not go to hibernation/sleep during the experiment. You can also use PowerToys Awake on the Windows side to quickly adjust the hibernation strategy without changing the power plan.

• (After ≥ one round) observe the file structure in trace_path as follows:

tracing_path
├── otrace0
│ └── OT_TRACE_xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx.frk
│ └── ...
├── otrace1

│ └── ...
├── otrace ...

│ └── ...
├── lt.csv
├── jank0.csv
├── jank1.csv
├── jank ...
├── tracing0.csv
tracing1.csv
├── tracing ...

Data Analysis

The files and scripts required for data analysis are located in the data_analysis folder in the project directory.

Data preprocessing

Since the collected indicators come from different formats and different data formats exist, it is necessary to first preprocess the indicator data. This function is performed by the . /data_analysis/impl_dataprocess.py.

TIPS: Before that, you may need to batch call the CLI to convert the FRK files in each folder of the trace_path to DB. this can be done by adding a new folder to the . /data_analysis/frk2db.py by calling convert_sql_impl().

The preprocessor will be implemented in the exp_dir output file: indicators_dir.

• indicators_data.pkl: packaged as pd.DataFrame containing the raw indicator data, with rows corresponding to the indicators, columns corresponding to the results of each experiment, and the elements of the DataFrame consisting of a list of the values of the indicators obtained during the experiment.
• indicators_intrested.csv: Estimate the trend of the indicator data based on pymannkendall, and output the results of the trend test of each indicator in each experiment to csv, the results may be: no trend, increasing, decreasing.

Indicator analysis

The path is . /data_analysis/indicator_analysis.ipynb Jupyter Notebook enables analysis and visualization of indicator data. See the Notebook internals for instructions on how to do this.