MemCapture

Memory capture and analysis tool for RDK

Build

Dependencies

MemCapture requires the following 3rd party libraries:

• Nlohmann JSON
• Inja

For desktop builds, these can be installed using vcpkg.

The incbin library is also included as a header in the 3rdparty directory.

VCPKG Usage

First, install VCPKG: https://vcpkg.io/en/getting-started.

Then build the project using the below commands:

$ mkdir build && cd ./build
$ cmake -DCMAKE_BUILD_TYPE=Release  -DCMAKE_TOOLCHAIN_FILE=[path to vcpkg]/scripts/buildsystems/vcpkg.cmake ..
$ make -j$(nproc)

Yocto

For Yocto builds, ensure the nlohmann/json and inja libraries are added as recipe dependencies.

Run

Usage: MemCapture <option(s)>
    Utility to capture memory statistics

    -h, --help          Print this help and exit
    -o, --output-dir    Directory to save results in
    -j, --json          Save data as JSON in addition to HTML report
    -d, --duration      Amount of time (in seconds) to capture data for. Default 30 seconds
    -p, --platform      Platform we're running on. Supported options = ['AMLOGIC', 'REALTEK', 'BROADCOM']. Defaults to Amlogic
    -g, --groups        Path to JSON file containing the group mappings (optional)

Example:

$ ./MemCapture --platform AMLOGIC --duration 30 --groups ./groups.json --output-dir /tmp/memcapture_results/

Averages are calculated over the specified duration.

Process Grouping

To ease analysis, MemCapture supports grouping processes into categories. This is done by providing MemCapture with a JSON file containing the groups and regexes defining which process(es) should belong to that group.

For example if the provided JSON file contained the below:

{
  "processes": [
    {
      "group": "Logging",
      "processes": [
        "syslog-ng",
        "systemd-journald"
      ]
    }
  ]
}

The syslog-ng and systemd-journald processes would belong to the Logging group and show up in the process list with that group name.

An example file (groups.example.json) is provided in the repo.

Results

By default, MemCapture will produce a file called report.html in the selected output directory. If no directory is provided, it will create a MemCaptureReport subdirectory in the current working directory to save the files.

This report can be opened in any web browser and contains a summary view of all the metrics collected.

If the -j argument is provided to MemCapture, then an additional results.json file will be created. This contains the raw data from MemCapture and is designed for importing into a backend system for analysis/reporting.

Notes

Tool currently supports three platforms - AMLOGIC (default), REALTEK and BROADCOM. Not all stats are available on all platforms