If we need to have a small footprint and high throughput, then the best way to achieve that is to use a parallel collector and a small heap. In this case, what we have is a full garbage collection, which pauses the application for slightly longer in comparison to using G1 GC. This will allow the application to perform GC to claim the memory allocated to objects, but the application will pause while it performs GC. The pauses will be longer, but not as frequent. Consequently, we can still achieve more throughput while keeping the footprint low.

We use the -XX:+UseParallelGC JVM option to indicate that the JVM should use the parallel GC type.