Many developers are either unaware of or prefer not to use JVM flags when running their Java applications, especially those requiring complex numerical inputs, such as memory-related flags. Despite the abundance of these flags, tuning even just one or two can significantly enhance application performance, which is critical in performance-sensitive scenarios.

In this talk, I am going to explain an approach for auto tuning JVM flags that I studied in my master’s thesis which was in collaboration with Oracle. In my master’s thesis, I explored five different machine learning models trained on data that was extracted from the G1 GC logs of a Java application, with the goal of improving its throughput and latency. The results were promising.

I believe that by sharing these tools and ideas, developers struggling to optimize memory-related JVM flags could not only benefit from this approach but also find it an enjoyable process.

Recorded at Jfokus 2025.