RFR: Adaptive accounts trashed cset twice

[Roman Kennke]

Looks good

Am 23. September 2017 12:35:19 MESZ schrieb Aleksey Shipilev <shade at redhat.com>:
>http://cr.openjdk.java.net/~shade/shenandoah/heuristics-adaptive-twice-cset/webrev.01/
>
>Recent improvement in adaptive heuristics was incomplete due to a
>simple overlook: we account the
>cset from the previous run twice. We do it "normally" with trashing the
>cset, and accounting it in
>trash/free. We do it second time via _bytes_in_cset. _bytes_in_cset is
>needed for polling potential
>free before actual trashing happens in should_start_conc_mark, but not
>in the actual cset construction.
>
>This fix makes LRUFragger survive with even further LDSes, when current
>(bad) code overshoots free
>space and runs into OOME-during-evac. This is not a regression per se,
>because on non-overloaded
>heap this matters much less, if at all.
>
>Testing: hotspot_gc_shenandoah, benchmarks
>
>Thanks,
>-Aleksey

-- 
Diese Nachricht wurde von meinem Android-Gerät mit K-9 Mail gesendet.