Time Matters: Minimizing Garbage Collection Overhead with Minimal Effort

Parameterization of garbage collectors can help improving the overall run time of programs, but finding the best parameter combination is a tedious task. We used a simple brute-force optimization algorithm for the Java Parallel GC to study the behavior of benchmarks with thousands of configurations. As a result of this study, we propose a practically usable strategy for finding a “good” parameter combination with a small number of experiments. Using this strategy, only 10 configurations need to be tested in order to reduce the garbage collection time down to 21% (56% on average). For the studied benchmarks, this results in an overall performance gain of up to 13%.