The pause time behavior depends a lot on what is going on in application itself. Shenandoah wiki has a section on that: https://wiki.openjdk.java.net/display/sh

The pause time behavior depends a lot on what is going on in application itself. Shenandoah wiki has a section on that: https://wiki.openjdk.java.net/display/shenandoah/Main#Main-GeneralIdeas https://wiki.openjdk.java.net/display/shenandoah/Main#Main-G.... It also depends on the hardware, because the same amount of GC work instruction-wise may translate to drastically different cycle/wall-clock time. Try the same 1 GB heap on large x86 desktop and then on Raspberry Pi! This, unfortunately, makes point benchmark numbers not very relevant, and you would want to see how the particular deployment works. With many applications you can push concurrent GCs down to sub-millisecond range without even trying. With some applications, adjusting the application to avoid pitfalls that inflate pauses helps -- the report mentions a few adjustments like that. Edit: Ah, I should have mentioned I am one of those Shenandoah devs, and I have never said "your mileage may vary" that often until diving into GC development work :D