RFR: 8277814: ConcurrentRefineThread should report rate when deactivating

[Kim Barrett]

On Fri, 26 Nov 2021 11:31:25 GMT, Thomas Schatzl <tschatzl at openjdk.org> wrote:

>> Please review this change to the logging by concurrent refinement threads.
>> This change adds the cards processed per ms to the deactivation log message.
>> 
>> Testing:
>> mach5 tier1
>> locally ran a test with gc+refine=debug logging enabled and verified the
>> expected change to the deactivation log message.
>
> src/hotspot/share/gc/g1/g1ConcurrentRefineStats.cpp line 38:
> 
>> 36:   // Report 0 when no time recorded because no refinement performed.
>> 37:   double secs = refinement_time().seconds();
>> 38:   return (secs > 0) ? (refined_cards() / (secs * MILLIUNITS)) : 0.0;
> 
> Not sure how this is used, but in similar calculations we only use and return a value if the elapsed time is large enough (something like a microsecond or so) to remove extreme outliers. Not sure if that is needed here too.

This change only uses it for logging.  I could just drop this function and do the equivalent calculation at the point where it's used.  I don't think a made up lower limit really makes that much sense anyway; I remember some places like you describe, but I thought many had been eliminated.  I could only find a couple.  The expectation here is that refined_cards() is probably also zero, unless the time granularity is rather coarse.

-------------

PR: https://git.openjdk.java.net/jdk/pull/6550