Generational algorithms make use of a property found in most Java programs. Several studies have revealed a pattern in the way objects are used. It has been found that objects that get created early in program execution usually live longer than objects that get created later. It is usually the youngest objects that are garbage collected first.
Look at the diagram below. The blue area in the diagram is a typical distribution for the lifetimes of objects. The X axis represents the bytes allocated by the JVM, and the Y axis represents th number of surviving bytes (live objects). The sharp peak at the left represents objects that can be reclaimed (i.e., have "died") shortly after being allocated. Iterator objects, for example, are often alive for the duration of a single loop.
Image source: The above image has been taken from the document "Tuning garbage collection with the 5.0 Java[tm] virtual machine"
If you notice, the distribution stretches out to the the right. This is because some objects live longer. Typically these are objects that have been created when the program started, and they live for the duration of the program. The lump observed after the first drop represents those objects that are created for some intermediate process. Some applications have very different looking distributions, but a surprisingly large number possess this general shape. The diagram above shows that most objects have a very short life span. Generational algorithms take advantage of this fact to optimize garbage collection in the JVM.
Generational algorithms divide the heap into several sub heaps. As objects get created, they are put in the sub heap that represents the youngest generation. When this area becomes full, the garbage collector is run and all unused objects in that sub heap get garbage collected. Objects that survive a few garbage collection attempts are promoted to a sub heap representing an older generation. The garbage collector runs most frequently in the younger heaps. Each progressively older generation of objects get garbage collected less often.
Generational algorithms are very efficient because they make use of certain well known properties of Java programs. Sun's Hotspot VM uses a modified form of generational algorithms. However, please note that this algorithm will not work efficiently with programs that make non-standard use of memory. We may configure the algorithm to work appropriately with such programs if we understand their memory usage well. If not, it is best to keep the default implementation.
Note: This text was originally posted on my earlier blog at http://www.adaptivelearningonline.net