Java Performance, 2nd Edition

Chapter 6. Garbage Collection Algorithms

Chapter 5 examined the general behavior of all garbage collectors, including JVM flags that apply universally to all GC algorithms: how to select heap sizes, generation sizes, logging, and so on. The basic tunings of garbage collection suffice for many circumstances. When they do not, it is time to examine the specific operation of the GC algorithm in use to determine how its parameters can be changed to minimize the impact of GC on the application.

The key information needed to tune an individual collector is the data from the GC log when that collector is enabled. This chapter starts, then, by looking at each algorithm from the perspective of its log output, which allows us to understand how the GC algorithm works and how it can be adjusted to work better. Each section then includes tuning information to achieve that better performance.

This chapter also covers the details of some new, experimental collectors. Those collectors may not be 100% solid at the time of this writing but will likely become full-fledged, production-worthy collectors by the time the next LTS version of Java is released (just as G1 GC began as an experimental collector and is now the default in JDK 11).

A few unusual cases impact the performance of all GC algorithms: allocation of very large objects, objects that are neither short- nor long-lived, and so on. Those cases are covered at the end of this chapter.

Understanding the Throughput Collector

We’ll start by ...