RFR(S): 8151808: Factor G1 heap sizing code out of the G1CollectorPolicy

Jesper Wilhelmsson jesper.wilhelmsson at oracle.com
Wed Mar 16 19:29:21 UTC 2016

Looks good!

Den 16/3/16 kl. 16:41, skrev Mikael Gerdin:
> Hi Tom,
> On 2016-03-15 19:18, Tom Benson wrote:
>> Hi Mikael,
>> I noticed that g1HeapSizingPolicy.hpp defines
>> NumPrevPausesForHeuristics, which is also defined in your new
>> g1Measurements.hpp header.  I think it should only be in the latter. And
>> as long as you're in g1HeapSizingPolicy.hpp, could you stick a blank
>> line before line 42, "// Ratio check data..."?  8^)
> Both use NumPrevPausesForHeuristics it's not clear to me that it's semantically
> the same value as the one in what is now called G1Analytics.
>  From my understanding the heap sizing wants to wait until "enough" gcs have
> occurred so that it makes sense to do the resizing but I don't see the
> connection with the lenghts of _recent_gc_times_ms,
> _concurrent_mark_remark_times_ms, etc.
> I've created a new webrev based on the rename to G1Analytics:
> http://cr.openjdk.java.net/~mgerdin/8151808/webrev.1/
>> Aside from that, looks good.
>> Tom
>> On 3/15/2016 9:40 AM, Mikael Gerdin wrote:
>>> Hi all,
>>> The code responsible for determining the heap expansion amount is
>>> currently part of the class G1CollectorPolicy. The computations and
>>> data required by this functionality is completely independent of the
>>> other pieces of the class and could be moved to class of its own.
>>> One could of course imagine that there could be situations where the
>>> sizing of the heap is influenced by the more general collector policy
>>> but there is nothing preventing that with the new model. The only
>>> difference would be that there would be a need for clearer API for
>>> such influences.
>>> The patch is based on 8151711 and 8151637 but can be delivered
>>> separately, I just have them all in the same workspace.
>>> Bug: https://bugs.openjdk.java.net/browse/JDK-8151808
>>> Webrev: http://cr.openjdk.java.net/~mgerdin/8151808/webrev.0/
>>> Testing: JPRT, RBT GC Testing
>>> Perf testing in aggregate with some other policy changes shows no
>>> regressions or improvements.
>>> /Mikael

More information about the hotspot-gc-dev mailing list