As menti
oned yesterday, I’ve been playing with jvmstat. It
allows you to visually watch garbage collection going on in your
application. It’s quite fascinating to watch the difference as you try
different garbage collection algorithms. For instance, using the [-XX:+AggressiveHeap](http://java.sun.com/docs/hotspot/gc1.4.2/#4.2.2.%20AggressiveHeap|outl
ine) algorithm makes a huge difference to the amount
of GC Time (or at least it did in my benchmark).
It’s also very helpful in understanding how the various garbage collection algorithms work - something I’ve always glossed over in the past.
Below are some traces from JBoss being hit pretty hard by an increasing number of clients. The first image shows it just starting up - note that there has been little memory reclaimed by the old generation garbage collector. In the second image you can see the saw tooth pattern in the old generation memory usage. It is also interesting to see the copying of object between the two survivor generations.
The third image shows the use of the AggressiveHeap garbage collector. Note that most of the objects don’t even make it out of the Eden pool. My hypothesis is that this shows that most of my objects are short lived. A lot less time is spent doing GC, too - compare the occasional spikes in the GC time graph to the almost continual GC in the previous example.
Some useful links:
Tuning Garbage
Collection with the 1.4.2 JavaTM Virtual Machine
Diagnosing a
Garbage Collection problem
Frequently
Asked Questions about Garbage Collection in the HotspotTM JavaTM Virtual
Machine