3 Replies Latest reply on Nov 10, 2009 3:36 PM by clebert.suconic

    Tests over LargeMessageExample

    clebert.suconic

      I have been doing a few tests over the LargemessageExample.

      Memory:

      - I can't get it to fail. I have used several file sizes, and have also made to send it multiple files here locally, and FlowControl is protecting OMEs on both sides.
      One thing that I have seen is the Memory Warning from the MemoryManager when using 10G files. But if I force a garbage collection, the Memory always goes back to the usual sizes, so I haven't seen any issues here.


      Consumer performance:

      - The BufferedInputStream didn't make any difference on receiving a message.
      I have measured the time on receiving multiple messages with and without the BufferedInput, and I couldn't get any significant differences.

      - If I set FlowControl to a huge number on both ways, the performance on consuming and sending is the same. Maybe there is a possibility of optimizing consumer flow control, as the flow control on consumer is taking longer than producing.

        • 1. Re: Tests over LargeMessageExample
          timfox

          I cannot replicate the issue any more, the error no longer appears (it did so every time). I can only asssume it has been fixed in the latest refactorings.

          I have also disabled the memory manager by default. The memory manager is kind of bogus since it's memory calculations include non referenced objects which might be ready for GC. It's not practical to force GC every time before printing the memory manager statistics.

          • 2. Re: Tests over LargeMessageExample
            clebert.suconic

            As for producing being faster than consuming, I have created this JIRA, so we can investigate possible optimizations later.


            https://jira.jboss.org/jira/browse/HORNETQ-212

            • 3. Re: Tests over LargeMessageExample
              clebert.suconic

               

              "timfox" wrote:

              I have also disabled the memory manager by default. The memory manager is kind of bogus since it's memory calculations include non referenced objects which might be ready for GC. It's not practical to force GC every time before printing the memory manager statistics.



              Just a thought...

              maybe we should add -verbose:gc in our run.bat and run.sh.

              I haven't been on the production side for a while, but what I remember from previous jobs is that people will usually add that.

              That would maybe help educate users?