When the CLR is attempting to locate unreachable objects, is does not literally examine each and every object placed on the managed heap. Doing so, obviously, would involve considerable time, especially in larger (i.e., real-world) applications.
To help optimize the process, each object on the heap is assigned to a specific “generation.” The idea behind generations is simple: the longer an object has existed on the heap, the more likely it is to stay there. For example, the class that defined the main window of a desktop application will be in memory until the program terminates. Conversely, objects that have only recently been placed on the heap (such as an object allocated within a method scope) are likely to be unreachable rather quickly. Given these assumptions, each object on the heap belongs to one of the following generations:
Generations 0 and 1 are termed ephemeral generations. As explained in the next section, you will see that the garbage collection process does treat ephemeral generations differently.
The garbage collector will investigate all generation 0 objects first. If marking and sweeping (or said more plainly, getting rid of) these objects results in the required amount of free memory, any surviving objects are promoted to generation 1. To see how an object’s generation affects the collection process, ponder Figure 8-5, which diagrams how a set of surviving generation 0 objects (A, B, and E) are promoted once the required memory has been reclaimed.
Figure 8-5. Generation 0 objects that survive a garbage collection are promoted to generation 1
If all generation 0 objects have been evaluated, but additional memory is still required, generation 1 objects are then investigated for reachability and collected accordingly. Surviving generation 1 objects are then promoted to generation 2. If the garbage collector still requires additional memory, generation 2 objects are evaluated. At this point, if a generation 2 object survives a garbage collection, it remains a generation 2 object given the predefined upper limit of object generations.
The bottom line is that by assigning a generational value to objects on the heap, newer objects (such as local variables) will be removed quickly, while older objects (such as a program’s application object) are not "bothered" as often.