Explanation of Java -Xprof Output

Question

I am trying to interpret profiling output produced by specifying the -Xprof flag.

For my application I get profiling output like this:

 Flat profile of 8.34 secs (775 total ticks): main
  Interpreted + native   Method
 10.4%    66  +    12    java.lang.ClassLoader.defineClass1
  3.6%    27  +     0    java.nio.HeapCharBuffer.<init>
  2.5%     1  +    18    java.io.UnixFileSystem.getBooleanAttributes0
...
 74.4%   380  +   179    Total interpreted (including elided)
     Compiled + native   Method
  0.3%     0  +     2    java.util.jar.JarFile$1.hasMoreElements
  0.3%     0  +     2    org.reflections.vfs.ZipDir$1$1.computeNext
  0.1%     0  +     1    java.lang.Object.<init>
...
  1.7%     3  +    10    Total compiled
         Stub + native   Method
  7.5%     0  +    56    java.util.zip.ZipFile.getEntry
  4.7%     0  +    35    java.lang.Object.getClass
  3.2%     0  +    24    java.lang.System.arraycopy
...
 23.2%     0  +   174    Total stub
  Thread-local ticks:
  3.1%    24             Blocked (of total)
  0.7%     5             Class loader

For each thread. My question is what is the difference between Interpreted, Compiled, and Stub methods, what are Thread-local ticks, and what is the meaning of the + native column? Is there any canonical documentation for the -Xprof profiler? Googling Xprof has yielded little.

Answer 1

The only place where I have ever found information on tools like XProf is some (slightly older) paper-based Java books.

• Interpreted + native: This figure shows the ticks used by the JVM while executing interpreted methods. The extra (+native) columns shows the native C methods that were called by these interpreted methods.
• Compiled + native: This figure shows the ticks used by the methods that were already parsed by the JIT compiler. After running your program a while, most of your major consumers from the interpreted section should appear as "Compiled" as JIT will compile them.
• Stubs + native: This figure is for JNI calls. This will likely to use the "native" column only as JNI is of course executed as a series of native calls.
• Thread-local ticks: This is listed as "miscellaneous" other entries and was written somewhere that "should not raise concerns from performance optimization perspective". I am not sure how much we want to trust that, but XProf is really not a documented tool just as you stated above.