This discussion is archived
7 Replies Latest reply: Jun 30, 2011 2:04 AM by Mohan RSS

JMX performance tests

Mohan Newbie
Currently Being Moderated
We are in the initial stages of using JMX and I might be jumping the gun here. What kind of WebLogic cluster and performance impact tests do people use when introducing JMX ? Here I mean the performance of the MBean server on a cluster and also the overhead of JMX itself. How do you think the application performance will be impacted by JMX ?

Thanks
  • 1. Re: JMX performance tests
    gimbal2 Guru
    Currently Being Moderated
    Mohan wrote:
    We are in the initial stages of using JMX and I might be jumping the gun here. What kind of WebLogic cluster and performance impact tests do people use when introducing JMX ? Here I mean the performance of the MBean server on a cluster and also the overhead of JMX itself. How do you think the application performance will be impacted by JMX ?

    Thanks
    I can't speak for other people, but you are certainly jumping the gun. I wouldn't implement anything until I've seen it work; a working prototype is a good way to go to do an impact analysis on, plus you get some hands on experience that you can rely on when making decisions.

    It doesn't make much sense to ask open ended questions like "how will it impact performance" in a public forum filled with people you do not know. I can tell you "You'll be find dude, I did this a million times!" and you are none the wiser, because for all you know I'm just a poser.
  • 2. Re: JMX performance tests
    Mohan Newbie
    Currently Being Moderated
    Not sure how one can deduce so much from a open-ended question like this about the OP and the work. A prototype is already in the works. When I mentioned "tests" I actually meant test code and unit tests that I am writing.

    Thanks.
  • 3. Re: JMX performance tests
    gimbal2 Guru
    Currently Being Moderated
    Mohan wrote:
    Not sure how one can deduce so much from a open-ended question like this about the OP and the work. A prototype is already in the works.
    Well then you are not jumping the gun, you only provide incomplete information. Don't know what else can be said about the remaining gaps in your OP, its hard to tell what exactly you are asking for without stepping on your toes because you already did it but failed to mention it. Apparently you know how to create unit tests, so not much to add there without specifics. As for remaining tests, generally you don't code them but you perform them. I'm pretty sure that you know how to set up an environment for load testing and system integration tests, right?

    So what! What can anyone here help you with? I hope you can figure it out some time.
  • 4. Re: JMX performance tests
    jtahlborn Expert
    Currently Being Moderated
    A general answer to your general question:

    - the platform mbean server itself will have pretty much no impact, considering every jvm since jdk 1.5 has started the platform mbean server by default. (i.e. you are already running one in all of your jvms).

    - as for the "overhead" of JMX, that could mean a lot of things. for each "service" you will most likely have some extra metadata bits lying around in memory (similar to rmi). so, a small fixed memory overhead per-service. there should be 0 cpu impact unless you are using a notification mechanism, in which case the impact will be that of any other monitoring, pub-sub type code. as for the communication layer, jmx uses rmi by default, so the communication overhead will be similar to an rmi server.

    basically, if your services are doing nothing (no active monitoring, just exported interfaces), your overhead will be a small amount of memory per service and no cpu.
  • 5. Re: JMX performance tests
    Mohan Newbie
    Currently Being Moderated
    as for the communication layer, jmx uses rmi by default, so the communication overhead will be similar to an rmi server.
    I think here you mean the way clients like JConsole connect to the MBean Server ??

    I was thinking of custom Java code within the VM that connects to the server and exposes the MBeans over HTTP/JSON to a browser.
  • 6. Re: JMX performance tests
    jtahlborn Expert
    Currently Being Moderated
    Mohan wrote:
    as for the communication layer, jmx uses rmi by default, so the communication overhead will be similar to an rmi server.
    I think here you mean the way clients like JConsole connect to the MBean Server ??

    I was thinking of custom Java code within the VM that connects to the server and exposes the MBeans over HTTP/JSON to a browser.
    how would anyone else be able to give you the performance impact of your own custom code?
  • 7. Re: JMX performance tests
    Mohan Newbie
    Currently Being Moderated
    how would anyone else be able to give you the performance impact of your own custom code?
    I am not getting the flow of your question here because I was asking about the default RMI part because HTTP/JSON is not as heavy as RMI and that could have a performance impact. A browser is sufficient to view the details.

    I will pursue this thread later anyway after testing.

    Edited by: Mohan on Jun 30, 2011 2:04 AM

Legend

  • Correct Answers - 10 points
  • Helpful Answers - 5 points