Interview with Robert Berry of IBM on JVM Performance


News: Interview with Robert Berry of IBM on JVM Performance

  1. IBM Distinguished Engineer Robert Berry of Hursley, England addresses question of JVM: specifically regarding the growing importance of the server environment, the need to integrate the JVM(scalable, 64-bit, and Core) with middleware, security requirements, and the combination effect of all of these factors.

    Read Spotlight on Java performance.
  2. Very interesting! It seems to me that IBM has spent more time and effort (and is more worried about...) server side JVM scalability than Sun is. We've recently been doing some JVM benchmarking to try and understand the relative capacity of different servers (dual and quad cpu), different JVMs, and different J2EE AppServers. Some of our benchmarks reveal significant problems with the Sun HotSpot 1.3.1 JVM on the quad (Intel) servers - particulary related to multiple threads and object creation. I plan on putting some new results on our web pages next week - check out:

    (This has VolanoMark results at present only).


  3. Performance Performance[ Go to top ]

    I hope that they also dont forget about reliability.

    Their recent run-for-an-hour-or-two-then-just-exit-for-no-reason-with-no-error-code JVM bug (on AIX) had a serious impact on a number of production systems and a couple of seriously large deployments. It affected Tomcat, Websphere... quite a few things

    Even for a huge IBM client it took over a month or so for it to trickle up the priority list and then get fixed (once some of the directors started getting involved, it sped up considerably).

  4. Performance Performance[ Go to top ]

    What release of the AIX JVM fixes this issue? We're running on AIX and have not encountered that problem, nor do we want to. :)


  5. Looks like there's a VM out there that outperforms IBM ...
  6. It's not an apples-to-apples comparison. The JRockit uses a processor that is 20% faster (900 MHz vs. 750 MHz) than the Sun VM and the Sun VM uses a processor that is 7% faster (750 MHz vs. 700 MHz) than the IBM VM.
  7. Arre o assami bhaiya! :-)

    A comparison using just the raw CPU #s will not cut it. Have you tried normalizing the performance using their SPECint2000 scores of the various CPUs?
  8. So, are you saying that the CPUs used by the other JVM vendors themselves don't scale? No, why would any vendor want to do that? Isn't it in the best interest of every JVM vendor to use the best CPU for the best results?

    Anyways, all the SPECjbb2000 hardware configurations were not necessary used for SPECint2000 benchmarks, so there is not data in

    Anyway, IBM still kicks everyone's butt in CPU scaling as well; check out the top SPECint2000 performers: