RFR JDK-8060017 G1: Report heap sizing time

[Thomas Schatzl]

Hi Erik,

On Wed, 2015-10-14 at 10:00 -0400, Eric Caspole wrote:
> Hi Thomas, I have a couple questions below on your comments so I can see 
> how to proceed -
> Thanks,
> Eric
> 
> 
> On 10/14/2015 05:17 AM, Thomas Schatzl wrote:
> > Hi Eric,
> >
> >
> > some comments:
> >   - the change only covers gathering timing numbers for expansion at the
> > end of GC. Expansion may also occur during GC when the space used for
> > evacuation is not sufficient any more.
> > Not sure if that is intended, but the CR itself does not indicate that.
> >
> > If you think gathering information about time to expand during gc is out
> > of scope of this CR (that's fine, it was me who suggested that), please
> > consider adapting the CR and create a new one for tracking the expansion
> > time during Object Copy.
> 
> I see what you mean here but I was not intending to add parallel Object 
> Copy expand time here with the way I understood this CR. I'll create a 
> new CR for recording the parallel Object Copy expand time indicated as a 
> sort of sub-timing of Object Copy.

That's fine.

> >   - please add the verification for this log message to the
> > TestGCLogMessages test.
> 
> OK.

Thanks.

> >   - maybe a better name for this phase that indicates that it is only
> > about expansion at the end of the GC could be found?
> 
> That is a good idea if later on Object Copy expand time is shown too.
>
> >   - I do not think it is a good idea to record heap expansion time during
> > full GC. The PhaseTimes data structure is otherwise only filled in
> > during young gc, so it's just confusing. (In G1CollectedHeap::
> > resize_if_necessary_after_full_collection()).
> 
> Do you mean you don't want to record/display the expand time at all from 
> there or just not record it using the PhaseTimes at that point? It seems 
> like you would still want to show the expand time even for full gc, why not?

While I would like to improve logging for Full GC, at the moment
G1GCPhaseTimes is only used for measuring young gc pauses, and is
heavily geared towards it.

So I think it would be better to not measure heap expansion/shrinking
time for Full GC at this point. Improving logging for Full GC is imo a
whole different set of changes.

Thanks,
  Thomas