kjkoster
10-05-2009, 21:16
Dear All,
Well well, looks like monitoring is on everyone's agenda these days. Adam Pilkington was kind enough to mention Java-monitor as one of the live monitoring tools.
Post mortem analysis on Java servers is a speciality in itself. The tools are there, but you need an array of them to cover all the subtleties of different types of heap dumps. Some tools are better at handling one edge case, and some are better at others. And that is just the Sun JVM, things get more interesting when you start running into IBM's JVM and Oracle's JRockit.
From reading Adam's blog post introducing the Post mortem JVM Diagnostics API (JSR 326) and Apache Kato (http://www.ibm.com/developerworks/java/library/j-kato.html), that is going to get a lot easier. We'll be getting much cleaner tool support at the JVM level. I hope this will let vendors compete on features, rather than on edge cases supported. :-)
At some point in the future, interfaces such as the one described in JSR-326 are going to be mainstream. When that is the case, we can start answering our user's most-requested feature: on-line, no-impact heap histograms, with full per-request memory use analysis. That's still quite a way away, but we can dream, can't we? :)
Kees Jan
Well well, looks like monitoring is on everyone's agenda these days. Adam Pilkington was kind enough to mention Java-monitor as one of the live monitoring tools.
Post mortem analysis on Java servers is a speciality in itself. The tools are there, but you need an array of them to cover all the subtleties of different types of heap dumps. Some tools are better at handling one edge case, and some are better at others. And that is just the Sun JVM, things get more interesting when you start running into IBM's JVM and Oracle's JRockit.
From reading Adam's blog post introducing the Post mortem JVM Diagnostics API (JSR 326) and Apache Kato (http://www.ibm.com/developerworks/java/library/j-kato.html), that is going to get a lot easier. We'll be getting much cleaner tool support at the JVM level. I hope this will let vendors compete on features, rather than on edge cases supported. :-)
At some point in the future, interfaces such as the one described in JSR-326 are going to be mainstream. When that is the case, we can start answering our user's most-requested feature: on-line, no-impact heap histograms, with full per-request memory use analysis. That's still quite a way away, but we can dream, can't we? :)
Kees Jan