RFR: Make adaptive default heuristics
Roman Kennke
rkennke at redhat.com
Tue Feb 7 09:59:33 UTC 2017
Hi,
Since introducing degenerated marking and tweaking the heuristics a
little bit for the double-marking issue, adaptive heuristics have
become much better. They no longer tend to go into full-gc (I couldn't
push them into full-GC even under heavy load), and it makes better use
of heap headroom (dynamic tends to underuse the heap and leave ~20%
empty at the top).
I propose to make adaptive the default. In all benchmarks I tested it
with I've seen some increase in performance, or at least no regression
(doesn't make a difference on apps that are not GC intensive).
http://cr.openjdk.java.net/~rkennke/adaptive/webrev.00/
Ok?
Roman
More information about the shenandoah-dev
mailing list