RFR (XS) 8240511: Shenandoah: parallel safepoint workers count should be ParallelGCThreads

Roman Kennke rkennke at redhat.com
Wed Mar 4 10:41:17 UTC 2020


Ok yes, that makes sense.

That flag predates upstream integration of that code, and it wasn't
quite clear how many threads are useful for safepoint cleanup. IIRC, I
found that hammering it with ParallelGCThreads was overkill - on my
machine. But you are right, hard-wiring it to 4 is certainly overkill on
smaller machines than mine ;-)

Roman

> RFE:
>   https://bugs.openjdk.java.net/browse/JDK-8240511
> 
> See bug for rationale. The best fix seems to be just using ParallelGCThreads and ditching
> Shenandoah-specific option altogether:
>   https://cr.openjdk.java.net/~shade/8240511/webrev.01/
> 
> Testing: hotspot_gc_shenandoah, eyeballing gross pause times
> 




More information about the hotspot-gc-dev mailing list