RFR: 8078901: Add trace event for G1 MMU information
erik.helin at oracle.com
Tue Jun 30 08:58:12 UTC 2015
On 2015-06-26, Stefan Johansson wrote:
> Thanks Erik for looking at this,
> On 2015-06-18 10:44, Erik Helin wrote:
> >Hi Stefan,
> >sorry for not getting to this earlier, thanks a lot for this patch!
> >On 2015-04-30, Stefan Johansson wrote:
> >>Please review this change to add a trace event for MMU tracking:
> >I have two comments:
> >- instead of using multiple inheritance, can you instead have the
> > G1MMUTracer as a field in G1NewTracer and G1OldTracer? Then you can
> > have a getter to the field.
> >- in G1MMUTracer, can you have the function report_mmu not take a
> > G1MMUTracker* but instead have two additional double parameters, one
> > for time_slice and one for max_gc_time? This way the tracers don't
> > need to know about G1MMUTracker.
> Erik and I have discussed this a little more off-line and come to the
> conclusion to have an all-static class handling the the reporting of the MMU
> event. This avoids the multiple inheritance and other problems of having a
> G1MMUTracer as part of the other tracers. This solution slightly differ from
> previously uses of the "tracers" but still keeps the code separated in a
> good way.
> New webrev:
Looks good, Reviewed. As we discussed offline, please revert the changes
touching G1OldTracer and CMSTracer (no need to re-review that change).
> >>When adding pause information to the G1MMUTracker, it now reports a
> >>trace-event with the latest MMU information.
> >>add_pause() previously took a bool argument that was unused, but said
> >>whether or not it was a concurrent-cycle pause. I removed the bool argument
> >>and added G1MMUTracer* argument. The G1MMUTracer is an interface that both
> >>the stw-tracer and the cm-tracer now implement to make sure the correct GC
> >>id is provided with the event.
More information about the hotspot-gc-dev