RFR: Trim down native GC footprint

Roman Kennke rkennke at redhat.com
Mon May 22 09:22:52 UTC 2017


Am 22.05.2017 um 10:36 schrieb Aleksey Shipilev:
> Hi,
>
> I have reviewed Shenandoah memory footprint with NMT, and suggest we do these
> things:
>
>  a) Avoid instantiating jvmstat counters when sampling is disabled;
>  b) Avoid creating matrix when UseShenandoahMatrix is disabled;
>  c) Avoid creating partialGC task queues when policy does not enable partial gc;
>  d) Avoid conservative heap alignment, but instead poll the actual one -- thus
> bitmaps are more compact for small heaps.
>
> All implemented here:
>   http://cr.openjdk.java.net/~shade/shenandoah/footprint-1/webrev.01/
>
> Before:
>
> - GC (reserved=17454KB, committed=17454KB)
>      (malloc=16430KB #257)
>      (mmap: reserved=1024KB, committed=1024KB)
>
> After:
>
> - GC (reserved=8742KB, committed=8742KB)
>      (malloc=8230KB #231)
>      (mmap: reserved=512KB, committed=512KB)
>
> Testing: hotspot_gc_shenandoah
>
> Thanks,
> -Aleksey
>
Very cool! Go!


Roman



More information about the shenandoah-dev mailing list