PerfSets

It's a toolset to help dealing with perfetto tools:

• Automate executing & managing perfetto tracing sessions: perf;
• Enhanced post process functions of tracing files: draw_cpu;
• Extract logcat from a perfetto trace file that recorded with logcat data source enabled: extract_logcat.

perf

• Run perfetto tracing on an Android device.
• Collect tracing results from the device.
• Support detach mode and stop tracing by specific events

Usage

• Regular mode: record a trace file in a specific duration. Run perf -h to see the options.

• Detach mode: record a trace file until a specific event happens. Run perf detach -h to see the options.

draw_cpu

Calculate and visualize CPU load from Perfetto trace files.

Installation

Ensure you have Python 3.6 or higher installed. Install the required dependencies using pip:

python3 -m pip install pandas tqdm matplotlib perfetto-trace-processor

Install trace_processor_shell from the github release page: https://github.com/google/perfetto/releases.

Download the latest release package according to your platform, extract all files and path of the trace_process_shell will be passed to the script later.

Usage

Run the draw_cpu.py script with the necessary arguments:

Options:

• -f, --file: Absolute path to the trace file. (Required)
• -b, --binary: Absolute path to the trace processor binary. (Required)
• --window_size_ms: Window size in milliseconds. If not provided, it is automatically calculated to have approximately 200 points.
• --window_move_ms: Window move (step) in milliseconds. If not provided, it is calculated based on the window size.
• --output: Base path to save the CPU load DataFrames as CSV files. (Optional)
• --plot: Flag to enable plotting the CPU load curves. (Optional) Example:

python3 draw_cpu.py -f example/example.pb -b /path/to/trace_processor_shell

Alternatively, run the script with accleration of the c++ plugin powered by PyBind11.

cd draw_cpu_cpp
python3 -m pip install .    # install the plugin to the current python environment

python3 draw_cpu.py -f example/example.pb -b /path/to/trace_processor_shell

This plugin brings a significant speedup to the script, especially when processing large trace files. In my computer, calculating CPU load for a 100MB trace file took 0.06 seconds with the plugin, compared to 20+ seconds without it.

Example output:

Caveats

• Trace File: Ensure that the trace file is recorded with sched/sched_switch and sched/sched_wakeup categories enabled.
• Trace Processor Binary: Ensure that the trace processor binary path provided with the -b option is correct and executable.
• Performance: Processing very large trace files may consume significant memory and processing time.
• Visualization: Installing matplotlib is optional but required if you want to visualize the CPU load curves. If not installed, the script will skip the plotting step.
• CPU Load Capping: The script caps CPU load percentage at 100% to avoid unrealistic values.

extract_logcat

Extract logcat from a perfetto trace file that recorded with logcat data source enabled.

Installation

See the installation instructions for draw_cpu.

Usage

Run the extract_logcat.py script with the necessary arguments:

python3 extract_logcat.py -f example/example.pb -b /path/to/trace_processor_shell -o output.log

Todo

• draw_cpu.py: accelerate calculations by employing multi-threading.