Thanks for the write-up with some actual numbers (though a contrived workload). Crazy how low the pause times for ZGC are now. I hope Valhalla can help alleviate some of the GC pressure in general as libraries/frameworks/programs migrate their data structures to use it.

The benchmark doesn't seem particular relevant for real world usage. I've always seen sub 100 microsecond pause times with ZGC (for a variety of different workloads), and that was with large heaps (10s of GBs). Shenandoah also delivers sub ms pause times for large heaps