Thursday, June 28, 2007

Jazoon 2007: Day three

The last thing we attended yesterday was Neil Gafter's Java Quiz BOF. Will have to take a look at FindBugs.sf.net.

Keynote with Erich Gamma. Eclipse Jazz looks mightily impressive. It's an Eclipse workbench addon that integrates *a lot* of groupware functionality in Eclipse. Think task-tracker, project members, svn commits and activities, the quick tour made it look really impressive (live demo). Unfortunately, Jazz is a commercial product. And would you really pay for anything that is based on Eclipse (being the obtrusively bug-filled platform it is)?

First session. SAP talk. Memory is split in heap space (objects) and perm gen. Heap dump content, objects, classes, classloaders, GC. Currently alive in memory objects.

Retained size of an object is not the same as the shallow size. Retained size includes the underlying chars for Strings f.ex. Retained size can be calculated by simulating a GC and see what objects remains.

By looking at dominator trees (parenting objects and their children), it becomes easy to calculate retained size. This technique (building dominator tree) is much cheaper in time use than to run a GC do discover what objects can be freed.

By grouping dominator tree by class it can be easier to the objects that are eating up your heap.

There's also a command line tool and an object heap qeuery language.

Critical problems encountered. Typical problems are innefficient data stcutures, XML, model/proxy driven classes.

All in all, this was a hidden product presentation for SAP's memory analyzer, but has some reasonably useful hints on debugging memory leaks/memory use.


...

Second session, JavaFX.

Lots of demos in this talk. More entertaining than it is useful, but nice to get some first hand input on JavaFX. Seems like it won't be useful performance-wise before after a year or so. Good, will check back on it then.

...

DTrace talk. Was split into two; a DTrace intro (fits me fine), and a case study done on Java across different OS'es (but the work/report there is not done).

Is like truss(1), but can inspect many other aspects It's scriptable. It's in java6 by default. Visual tool on the way for Leopard.

Is not intrusive/safe to the runtime. No impact on the probes that are enabled by default.

The awk-like language is called D.

Many performance bottlenecks can be found and analyzed with a couple of 1-liners of DTrace. Jaron Jenson has some good examples.

Over to the second part of the talk. They tried to reveal whether the Java behaviour/characteristics were different between OS'es.

Measured on different platforms, amd, intel and sparc.

The project isn't done because MacOSx does not have DTrace built into it. Won't be released until this fall. FreeBSD is there, but you have to get a snapshot. Results will be put on the developers' blogs.

For us not on java 1.6, we can use the DVM agent (prototype).

To properly make use of DTrace you should have good knowledge of OS, C, Java VM, etc. Have a long way to go.

...

After the technical sessions we were offered to see the Die Hard 4 premiere, but we preferred travelling to Rapperswil by boat for a bit of tourist activity. We stopped by HSR and met pattern-author Peter Sommerlad and got some interesting insight in research they're doing to reduce code complexity/increase code quality. First time I've seen academics deal with stuff so close to my actual field of work. Useful.

Rapperswil was an old village/castle like thing and we got a fine tour around the place with this excellent story teller guy who appearantly was keeper of the city records of something. Was a fine opportunity to socialize with other Jazooners.

After the tour we ate some nice pasta and headed back to Zurich for more free beer and karaoke. Great day.