No description, website, or topics provided.
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Type Name Latest commit message Commit time
Failed to load latest commit information.

Android Sampling Profiler Facade

There is a hidden sampling profiler in Android going back at least as far as Android 2.3. To make it easy to use the sampling profiler we have created an small library, the Android Sampling Profiler Facade. It works with Android 2.3 and Android 4.x. To use just drop the JAR file into the libs/ directory of your Android project. If your project is an Android Studio project you also have to add the JAR dependency to your build.gradle file:

dependencies {
    compile files('libs/sampling-profiler-facade.jar')

Since the Android sampling profiler is in-process you need to add some code to your app. The public API of the facade is contained in com.appfour.samplingprofiler.SamplingProfilerfacade.

First initialize the profiler, e.g. in on the onCreate() method of an activity. The init() methods take some parameters which can not be changed for this profiling session:

  • stackDepth - specifies number of stackframes that are being captured during profiling. With a higher number more of the callchain to hotspots will be visible when analyzing the captured data later. But a higher stack depth also requires more memory on the VM heap during profiling. A stack depth of ten is a good number to start with.

  • intervalInMs - specifies the interval between samples taken in ms. The lower the number the higher the resolution. But a lower number also means that the sampling profiler requires more memory and affects the behavior of the program more. An interval of 10 ms (equivalent to about 100 samples per second) is a good value to start with.

  • The third parameter specifies what threads are being sampled. There are several overloads for specifying all threads, a fixed number of threads, or a dynamically computed ThreadSet

Sampling can be started and stopped using the startSampling() and stopSampling() methods. This way only the interesting parts of the app are being profiled.

Once all interesting parts have been profiled the profiler can be shut down and profiling data can be written out with the writeHprofDataAndShutdown() method.

The following example shows how take start samples when the activity is started and how to stop taking samples and write the profiling data once the activity is left.

    protected void onStart() {
        SamplingProfilerFacade.init(10, 10, Thread.currentThread());

    protected void onStop() {
        File outFile = new File(Environment.getExternalStorageDirectory(), "sampling.hprof");
        try {
            FileOutputStream outStream = new FileOutputStream(outFile);
            try {
            } finally {
        } catch (IOException e) {
            Log.e("Sampling", "I/O exception writing sampling profiler data", e);

Analyzing the profiling data

The profiling data is written out in HPROF format. The only tool I have found which can analyze those files is JPerfAnal. It looks a bit outdated and does not have a lot of features but it works well enough for finding hotspots. After copying the .hprof file onto your PC run JPerfAnal with java -jar jperfanal.jar sampling.hprof.

The line number information captured does not seem to be accurate so the 'Method times by Caller' and 'Method times by Callee' sections are the most interesting.