The heuristics currently do not seem to take the size of the heap into account when calculating the default number of worker threads. This can often lead to a huge amount of thread overhead for small heaps. As an extreme example, on a on a 40 HW thread machine, 9-ea+144 will start up over 100 GC threads (G1) OOTB for a 16m heap.

This has caused performance issues in environments where many small processes are run in parallel.

This ER is to document the issue and prompt a discussion on whether we can improve the ergonomics for smaller heaps.