Re: Cayenne performance testing

From: Andrus Adamchik (andru..bjectstyle.org)
Date: Wed Feb 01 2006 - 13:16:29 EST

  • Next message: Kevin Menard: "Re: Cayenne performance testing"

    Yeah, I am also not psyched about the web frontend. On the one hand
    it introduces extra uncertainty to the test results (TCP/IP lag,
    etc). It would be harder to separate signal from the noise, as the
    noise will increase significantly.

    On other other hand if JMeter is such a great tool to configure
    scenarios and generate reports, at least it is worth a try. We can
    combine both solutions by writing the cases to JUnit API like I
    suggested before and then create a thin servlet-based test invoker
    (with no frameworks or JSP - just plain servlet that starts a unit
    test).

    This way we will have the ability to run with Ant outside of the web
    container, or run with JMeter. BTW, if anyone wants to take a lead on
    setting this up, please speak up ;-)

    Andrus

    On Feb 1, 2006, at 9:51 AM, Gentry, Michael ((Contractor)) wrote:
    > You can still do diffs from one Cayenne version to another. I just
    > want
    > to eliminate things like Jetty/Tomcat performance (they do differ in
    > startup times/etc) and only focus on Cayenne. Also, it would be much
    > simpler to run if Cayenne-only. Wouldn't force someone to install a
    > bunch of extra software that arguably doesn't give them anything,
    > except
    > a bigger footprint on the HD. The more layers you add, the harder
    > it is
    > to setup and maintain.
    >
    > I've not looked at JMeter, but perhaps I can take a peek at it
    > later if
    > I have time.
    >
    > Thanks!
    >
    > /dev/mrg
    >
    >
    > -----Original Message-----
    > From: news [mailto:new..ea.gmane.org] On Behalf Of Ahmed Mohombe
    > Sent: Wednesday, February 01, 2006 9:44 AM
    > To: cayenne-deve..bjectstyle.org
    > Subject: Re: Cayenne performance testing
    >
    >
    >> I kind of disagree about the web application testing part. I
    >> think if
    >> we are going to profile code, it should be Cayenne code. Let the
    >> Tapestry/Struts/JSP/JSF/Tomcat/Jetty/Resin/WebSphere/JBoss/etc guys
    >> worry about profiling their code.
    > As Andrus said, it's not about the raw performance, but about the
    > "Diff"
    > from
    > one version to the other of Cayenne. In this case, since the
    > webapplication on
    > top of Cayenne remains the same, the diff would represent the pure
    > cayenne numbers.
    >
    > Considering how simple is to do this with JMeter, IMHO it
    > represents the
    > smallest
    > possible effort with the greatest results. Not to mention about the
    > nice
    > diagrams
    > and statistics JMeter offers, all this for free, with the lease
    > possible
    > work.
    >
    > just my 2 cents,
    >
    > Ahmed.
    >
    >



    This archive was generated by hypermail 2.0.0 : Wed Feb 01 2006 - 13:16:31 EST