Java profiling - how can I get a method by method analysis of my application?

Question

I want to run my Java app and for a given workload be able to see:

• how many times a given function was called
• how expensive each function call is in relative terms (i.e. how long each takes to execute)

I know broadly where the bottle neck is in my application but I need a much more fine grained view in order to narrow it down.

Thanks

Edit jvisualvm looks like the tool - it identified the problem in about 30 seconds. I just need to know what 'selftime' means in the context of a method profile. Thanks

Answer 1

Simplest approach for a program running in java 6 from Sun is to use the jvisualvm program in the jdk. Allows you to attach and profile without any special setup.

Answer 2

Easiest way is to use -prof, e.g: java -prof -jar yourjar.jar

That will print a file called java.prof after the program has finished running.

See the HPROF documentation page

In my application I use: -Xrunhprof:cpu=samples,thread=y,doe=y

This prints a report that contains, amongst other things, this:

CPU SAMPLES BEGIN (total = 55110) Sun Feb  7 17:02:51 2010 rank   self   accum   count  trace  method 1      69.68% 69.68%   38399 300361 java.net.SocketInputStream.socketRead0 2      24.40% 94.08%   13448 300386 java.net.SocketInputStream.socketRead0 3      0.20%  94.28%     108 300425 java.io.FileOutputStream.writeBytes 4      0.19%  94.47%     107 300976 java.net.PlainDatagramSocketImpl.receive0 5      0.19%  94.65%     102 300414 package.BlockingSampleBuffer.addSample 6      0.16%  94.82%      90 300365 java.net.SocketOutputStream.socketWrite0 7      0.16%  94.98%      89 300412 package.BlockingSampleBuffer.addSample 8      0.15%  95.13%      84 300430 java.lang.Object.wait 9      0.14%  95.27%      77 300592 java.io.FileOutputStream.writeBytes 10     0.14%  95.41%      76 300566 java.lang.AbstractStringBuilder.<init> 

So you can see the total time (in seconds) spent in various methods. In mine the app spends most of its time waiting for data from a remote host (not unlikely over an internet connection).