docs/html/tools/debugging/systrace.jd

page.title=Analyzing Display and Performance with Systrace
parent.title=Debugging
parent.link=index.html
@jd:body

<div id="qv-wrapper">
  <div id="qv">
    <h2>In this document</h2>
    <ol>
      <li><a href="#overview">Overview</a>
      </li>
      <li><a href="#generate">Generating Traces</a>
        <ol>
          <li><a href="#limit-trace">Limiting trace data</a></li>
          <li><a href="#config-categories">Configuring trace data categories</a></li>
          <li><a href="#running">Running a trace</a></li>
        </ol>
      </li>
      <li><a href="#analysis">Analyzing Traces</a>
        <ol>
          <li><a href="#long-processes">Long running processes</a></li>
          <li><a href="#display-interupts">Interruptions in display execution</a></li>
        </ol>
      </li>
    </ol>
    <h2>See also</h2>
    <ol>
      <li><a href="{@docRoot}tools/help/systrace.html">Systrace</a>
      </li>
    </ol>
  </div>
</div>

<p>After building features, eliminating bugs and cleaning up your code, you should spend some
  time looking at the performance of your application. The speed and smoothness with which your
  application draws pixels and performs operations has an significant impact on your users'
  experience.</p>

<p>Android applications operate within a shared resource environment, and the performance of
  your application can be impacted by how efficiently it interacts with those resources in
  the larger system. Applications also operate in a multithreaded environment, competing with other
  threaded processes for resources, which can cause performance problems that are hard to diagnose.
</p>

<p>The {@code systrace} tool allows you to collect and review code execution data for your
  application and the Android system. You can use this data to diagnose execution problems and
  improve the performance of your application.</p>


<h2 id="overview">Overview</h2>

<p>{@code systrace} helps you analyze how the execution of your application fits into the larger
  Android environment, letting you see system and applications process execution on a common
  timeline. The tool allows you to generate highly detailed, interactive reports from devices
  running Android 4.1 and higher, such as the report in figure 1.</p>

<img src="{@docRoot}images/systrace/report.png" alt="Systrace example report" id="figure1" />
<p class="img-caption">
  <strong>Figure 1.</strong> An example {@code systrace} report on 5 seconds of process execution
  for a running application and related Android system processes.
</p>


<h2 id="generate">Generating Traces</h2>

<p>In order to create a trace of your application, you must perform a few setup steps. First, you
  must have a device running Android 4.1 or higher. Setup the device for
  <a href="{@docRoot}tools/device.html#setting-up">debugging</a>, connect it to your development
  system and install your application. Some types of trace information, specifically disk activity
  and kernel work queues, require root access to the device, but most {@code systrace} log data
  only requires that the device be enabled for developer debugging.</p>


<h3 id="limit-trace">Limiting trace data</h3>

<p>The {@code systrace} tool can generate a potentially huge amount of data from applications
  and system sources. To limit the amount of data the tool collects and make the data more relevant
  to your analysis, use the following options:</p>

<ul>
  <li>Limit the amount of time covered by the trace with the {@code -t, --time} option. The default
    length of a trace is 5 seconds.</li>
  <li>Limit the size of the data collected by the trace with the {@code -b, --buf-size} option.</li>
  <li>Specify what types of processes are traced using the {@code --set-tags} option and the
  {@code --disk}, {@code --cpu-freq}, {@code --cpu-idle}, {@code --cpu-load} options.</li>
</ul>


<h3 id="config-categories">Configuring trace data categories</h3>

<p>To use {@code systrace} effectively, you must specify the types of processes you want to trace.
  The tool can gather the following types of process information:</p>

<ul>
  <li>General system processes such as graphics, audio and input processes (selected using trace
    <a href="{@docRoot}tools/help/systrace.html#tags">Tags</a>).</li>
  <li>Low level system information such as CPU, kernel and disk activity (selected using
    <a href="{@docRoot}tools/help/systrace.html#options">Options</a>).</li>
</ul>

<p>To set trace tags for {@code systrace} using the command-line:</p>

<ol>
  <li>Use the {@code --set-tags} option:
<pre>
$> python systrace.py --set-tags=gfx,view,wm
</pre>
  </li>
  <li>Stop and restart the {@code adb} shell to enable tracing of these processes.
<pre>
$> adb shell stop
$> adb shell start
</pre></li>
</ol>

<p>To set trace tags for {@code systrace} using the device user interface:</p>

<ol>
  <li>On the device connected for tracing, navigate to: <strong>Settings &gt;
      Developer options &gt; Monitoring &gt; Enable traces</strong>.</li>
  <li>Select the categories of processes to be traced and click <strong>OK</strong>.</li>
</ol>

<p class="note">
  <strong>Note:</strong> The {@code adb} shell does not have to be stopped and restarted when
  selecting trace tags using this method.
</p>


<h3 id="running">Running a trace</h3>

<p>After you have configured the category tags for your trace, you can start collecting
  information for analysis.</p>

<p>To run a trace using the current trace tag settings:</p>

<ol>
  <li>Make sure the device is connected through a USB cable and is
  <a href="{@docRoot}tools/device.html#setting-up">enabled for debugging</a>.</li>
  <li>Run the trace with the low-level system trace options and limits you want, for example:
<pre>
$> python systrace.py --cpu-freq --cpu-load --time=10 -o mytracefile.html
</pre>
  </li>
  <li>On the device, execute any user actions you want be included in the trace.</li>
</ol>


<h2 id="analysis">Analyzing Traces</h2>

<p>After you have generated a trace using {@code systrace}, it lists the location of the output
  file and you can open the report using a web browser.
  How you use the trace data depends on the performance issues you are investigating. However,
  this section provides some general instructions on how to analyze a trace.</p>

<p>The reports generated by {@code systrace} are interactive, allowing you to zoom into and out of
  the process execution details. Use the <em>W</em> key to zoom in, the <em>S</em>
  key to zoom out, the <em>A</em> key to pan left and the <em>D</em> key to pan
  right. Select a task in timeline using your mouse to get more information about the task.
  For more information about the using the keyboard navigation shortcuts and navigation, see the
  <a href="{@docRoot}tools/help/systrace.html#viewing-options">Systrace</a> reference
  documentation.</p>

<h3 id="long-processes">Long running processes</h3>

<p>A well-behaved application executes many small operations quickly and with a regular rhythm,
  with individual operations completing within few milliseconds, depending on the device
  and the processes being performed, as shown in figure 2:</p>

<img src="{@docRoot}images/systrace/process-rhythm.png" alt="Systrace exerpt of app processing"
id="figure2" />
<p class="img-caption">
  <strong>Figure 2.</strong> Excerpt from a trace of a smoothly running application with a regular
  execution rhythm.
</p>

<p>The trace excerpt in figure 2 shows a well-behaved application with
  a regular process rhythm (1). The lower section of figure 2 shows a magnified section of
  the trace indicated by the dotted outline, which reveals some irregularity in the process
  execution. In particular, one of the wider task bars, indicated by (2), is taking slightly
  longer (14 milliseconds) than other, similar tasks on this thread, which are averaging between
  9 and 12 milliseconds to complete. This particular task execution length is likely not noticeable
  to a user, unless it impacts another process with specific timing, such as a screen update.</p>

<p>Long running processes show up as thicker than usual execution bars in a trace. These thicker
  bars can indicate a problem in your application performance. When they show up in your
  trace, zoom in on the process using the
  <a href="{@docRoot}tools/help/systrace.html#viewing-options">keyboard navigation</a> shortcuts to
  identify the task causing the problem, and click on the task to get more information. You should
  also look at other processes running at the same time, looking for a thread in one process that is
  being blocked by another process.</p>


<h3 id="display-interupts">Interruptions in display execution</h3>

<p>The {@code systrace} tool is particularly useful in analyzing application display slowness,
  or pauses in animations, because it shows you the execution of your application across multiple
  system processes. With display execution, drawing screen frames with a regular rhythm is essential
  for good performance. Having a regular rhythm for display ensures that animations and motion are
  smooth on screen. If an application drops out of this rhythm, the display can become jerky or slow
  from the users perspective.</p>

<p>If you are analyzing an application for this type of problem, examine the
  <strong>SurfaceFlinger</strong> process in the {@code systrace} report where your application is
  also executing to look for places where it drops out of its regular rhythm.</p>

<img src="{@docRoot}images/systrace/display-rhythm.png" alt="Systrace exerpt of display processing"
id="figure3" />
<p class="img-caption">
  <strong>Figure 3.</strong> Excerpt from a trace of an application showing interruptions in
  display processing.
</p>

<p>The trace excerpt in figure 3 shows an section of a trace that indicates an interruption in the
  device display. The section of the <strong>SurfaceFlinger</strong> process in top excerpt,
  indicated by (1), shows that display frames are being missed. These
  dropped frames are potentially causing the display to stutter or halt. Zooming into this problem
  area in the lower trace, shows that a memory operation (image buffer dequeuing and allocation) in
  the <strong>surfaceflinger</strong> secondary thread is taking a long time (2). This delay
  causes the application to miss the display update window, indicated by the dotted
  line. As the developer of this application, you should investigate other threads in your
  application that may also be trying to allocate memory at the same time or otherwise blocking
  memory allocation with another request or task.</p>

<p>Regular, rhythmic execution of the <strong>SurfaceFlinger</strong> process is essential to smooth
  display of screen content, particularly for animations and motion. Interruptions in the regular
  execution pattern of this thread is not always an indication of a display problem with your
  application. Further testing is required to determine if this is actually a performance problem
  from a user perspective. Being able to identify display execution patterns like the example above
  can help you detect display problems and build a smooth-running, high-performance application.
</p>

<p class="note">
  <strong>Note:</strong> When using {@code systrace} to analyze display problems, make sure
  you activate the tracing tags for <strong>Graphics</strong> and <strong>Views</strong>.
</p>

<p>For more information on the command line options and keyboard controls for {@code systrace},
see the <a href="{@docRoot}tools/help/systrace.html">Systrace</a> reference page.</p>