RFR (S/M): 8136678: Implement adaptive sizing algorithm for IHOP

Thomas Schatzl thomas.schatzl at oracle.com
Mon Nov 23 10:28:51 UTC 2015


Hi Mikael,

  thanks for looking at the change.

On Mon, 2015-11-23 at 10:26 +0100, Mikael Gerdin wrote:
> On 2015-11-20 19:54, Thomas Schatzl wrote:
> > Hi,
> >
> > On Fri, 2015-11-20 at 16:56 +0100, Mikael Gerdin wrote:
> >> On 2015-11-13 13:16, Thomas Schatzl wrote:
> >>> Hi all,
> >>>
[...]
> > I will move the const.
> 
> Thanks.

Done.

> >
> >> g1IHOPControl.cpp
> >>
> >>    119   _marking_times_s(10, 0.95),
> >>    120   _allocation_rate_s(10, 0.95),
> >> Would you mind making these named constants? The 10 is the same as the
> >> truncated seq length in G1CollectorPolicy but the 0.95 is not the
> >> standard "alpha" value, would it make sense to describe what 0.95 comes
> >> from?
> >
> > Internal testing showed that it is good to strongly favor most recent
> > samples. I will make this an experimental flag.
> 
> I think a comment would suffice instead of a flag.
> 
> // 0.95 favors more recent samples
>  >>    119   _marking_times_s(10, 0.95),
>  >>    120   _allocation_rate_s(10, 0.95),

Done.

New webrev at
http://cr.openjdk.java.net/~tschatzl/8136678/webrev.2_to_3/ (diff)
http://cr.openjdk.java.net/~tschatzl/8136678/webrev.3/ (full)

Thanks,
  Thomas




More information about the hotspot-gc-dev mailing list