6 Replies Latest reply: Sep 3, 2014 12:18 PM by RandallT RSS

    Issue with Java Mission Control 5.3

    Subhash Ganesan

      I see the following issue when using Java Mission Control 5.3 - the text area that appears in the Method Profiler section of the Code => Call Tree tab accepts text for filtering purposes. However, the filtering is seen to be applicable only to the top level nodes. For e.g., I am unable to search for a certain Java package name (or its part) that I know exists deep down in the stack trace. Is there a way to search for text that appears in lower level nodes? I find that the search does not work even if I expand the tree using the Expand Subtree option. Thanks for any insight.

        • 1. Re: Issue with Java Mission Control 5.3
          Klara Ward, Java Mission Control Dev-Oracle

          I'm afraid this works as designed (which doesn't mean than the design is good)

           

          Workaround would be to expand all subtrees and then copy the text to some other tool

           

           

          We've redesigned this completely in the future version we're working on, not sure if we will fix this in the minor updates released before that though.

          • 2. Re: Issue with Java Mission Control 5.3
            Subhash Ganesan

            Thanks a lot Klara, for your reply.

             

            There is another observation with the Flight Recorder that I wanted to share here - again, this is regarding the Method Profiler section in the Code => Call Tree tab. The Sample Count values in the Stack Trace table are seen to vary considerably between consecutive recordings of the same use case. For e.g., I see a value of 97 against a certain Java method from my first recording, and a value of 77 (which is roughly 20% lower) for the same method from a second recording done soon afterwards; and the use cases for the recordings were exactly identical. Is this an expected behavior? This can sometimes mislead a developer into believing that a particular performance fix made to a certain method has worked (since the Sample Count is seen to be lower), while in reality the lower value is not caused by the fix at all. Or is there some configuration that I may be missing while recording the use case?

             

            Thanks again, for your help.

            • 3. Re: Issue with Java Mission Control 5.3
              Klara Ward, Java Mission Control Dev-Oracle

              Since the Method Profiler is a sampling profiler, I guess this could happen, depending on what settings you have, how cpu intensive your application is, and probably some other factors.

              I'm not sure we recommend Flight Recorder as a performance measuring tool, more for finding the cause of performance problems.

              Makes sense?

              • 4. Re: Issue with Java Mission Control 5.3
                Subhash Ganesan

                Thanks again, for your response. I was using the Flight Recorder only to see if there were performance issues with a certain Java application. And I was checking if a certain performance fix that I made had improved the sample counts. Since the counts had come down considerably after the fix (for the exact same use case and configuration), I initially assumed my fix had worked. However, I found later that even without my fix in place, the counts were sometimes seen to be lower than for the original recording. Hence thought I should ask you.

                • 5. Re: Issue with Java Mission Control 5.3
                  Subhash Ganesan

                  Hello - any thoughts on what I had mentioned in my last post (pasted below for clarity)?

                   

                  Thanks again, for your response. I was using the Flight Recorder only to see if there were performance issues with a certain Java application. And I was checking if a certain performance fix that I made had improved the sample counts. Since the counts had come down considerably after the fix (for the exact same use case and configuration), I initially assumed my fix had worked. However, I found later that even without my fix in place, the counts were sometimes seen to be lower than for the original recording. Hence thought I should ask you.


                  Thanks for any insight.

                  • 6. Re: Issue with Java Mission Control 5.3
                    RandallT

                    I also have this issue on 5.3. It is very difficult to do even a basic level of profiler analysis. The tool is ok at finding big watermelons (CPU cost or memory allocations), but when you want to drill in and see any more detail, it becomes quite painful. Best case is that I would be able to "swim" through the profile data, moving easily from method-to-method, and see the recorded sample counts in a base (samples in this method itself) and cumulative (samples in this method and all called methods) fashion. The indentation in the tree view can be painful. I would suggest a more simplified view as an option,

                     

                    PARENT (callers) methods, full names and cumulative sample counts (percentages would be nice as well)

                    ...

                     

                    SELF (called) method, full name and cumulative sample counts (percentages would be nice as well)

                     

                    CHILD (calles) methods, full names and cumulative sample counts (percentages would be nice as well)

                    ...

                     

                    Clicking on any row in the above immediately makes that method become the SELF in the view.

                     

                    Also, I should be able to enter the class/method name into some text box above this view and easily get into the data for the method. With the current tool, there is no way to quickly go to the data for a specific known method.