Uploaded image for project: 'JBoss Enterprise Application Platform'
  1. JBoss Enterprise Application Platform
  2. JBEAP-20567

[GSS](7.3.z) Performance Falloff In Embedded Broker

    XMLWordPrintable

Details

    • Bug
    • Resolution: Obsolete
    • Major
    • None
    • 7.3.3.GA
    • A-MQ7
    • None
    • False
    • False
    • Undefined
    • Hide

      Load a broker with 100000 messages of varying size

      • I use the attached webapp and invoke it with:
        http://localhost:8080/producer-servlet/InboundProducer?msgs=50000&delay=0&maxSize=16536
        

      and again to add a few large messages to the mix:

      http://localhost:8080/producer-servlet/InboundProducer?msgs=5&delay=0&maxSize=16536000
      

      And again to produce 50k more:

      http://localhost:8080/producer-servlet/InboundProducer?msgs=50000&delay=0&maxSize=16536
      

      I then consume the messages with a single-threaded core consumer (attached). Look for delta times in the log hitting 40-50ms.

      Leave the client and restart the broker (it may take a couple restarts)

      Delta times drop to 0-2 ms and messages are drained quickly.

      Show
      Load a broker with 100000 messages of varying size I use the attached webapp and invoke it with: http: //localhost:8080/producer-servlet/InboundProducer?msgs=50000&delay=0&maxSize=16536 and again to add a few large messages to the mix: http: //localhost:8080/producer-servlet/InboundProducer?msgs=5&delay=0&maxSize=16536000 And again to produce 50k more: http: //localhost:8080/producer-servlet/InboundProducer?msgs=50000&delay=0&maxSize=16536 I then consume the messages with a single-threaded core consumer (attached). Look for delta times in the log hitting 40-50ms. Leave the client and restart the broker (it may take a couple restarts) Delta times drop to 0-2 ms and messages are drained quickly.

    Description

      When consuming messages using an external CORE client, performance seems to inexplicably fall off from 100s of msg/s throughput (with 0-2ms latency) to around 20 msg/s with around 50ms latency. The falloff does not seem to be related to paging or client acknowledgements. Running a tcp dump between the broker and client, it appears that the client is acking messages, but the time between acknowledgement and subsequent dispatch jumps from sub-ms to around 50ms.

      Restart of the broker restores performance, though sometimes multiple (2-3) restarts are required.

      Attachments

        1. MqReceiver.zip
          13 kB
        2. producer-servlet.war
          61 kB
        3. slowdown0-c.svg
          389 kB
        4. slowdown0-lock.svg
          13 kB
        5. slowdown0-malloc.svg
          1.46 MB
        6. standalone-full.xml
          34 kB

        Activity

          People

            jboss-set_jira JBoss SET
            rhn-support-dhawkins Duane Hawkins
            Votes:
            2 Vote for this issue
            Watchers:
            5 Start watching this issue

            Dates

              Created:
              Updated:
              Resolved: