RFR (S): 8136679: JFR Event for adaptive IHOP

Thomas Schatzl thomas.schatzl at oracle.com
Thu Nov 19 11:02:33 UTC 2015


Hi Tom,

On Wed, 2015-11-18 at 17:34 -0500, Tom Benson wrote:
> Hi Thomas,
> In g1IHOPControl.cpp, the call to report_basic_ihop_statistics at line 
> 198 in G1AdaptiveIHOPControl::send_event passes 
> _allocation_rate_s.last() for the last_allocation_duration argument.  
> Looking at your other IHOP change, I think it actually is a rate, rather 
> than a duration, so perhaps not what you want here. ?
> 
> I also noticed an extra-trivial typo in trace.xml: "since last GC in 
> bytes/seconds"should be /second.
> 
> Looks fine otherwise.

It contains much more changes than you'd expect because the first issue
nudged me to move some duplicated code into the base class.

It also contains changes suggested by the perf team that improve the JFR
event specific to adaptive IHOP to also pass a IHOP threshold percentage
relative to the internal IHOP target occupancy, not only relative to the
total heap size.

Fixed in
http://cr.openjdk.java.net/~tschatzl/8136679/webrev.1_to_2
http://cr.openjdk.java.net/~tschatzl/8136679/webrev.2 (full)

Thanks,
  Thomas





More information about the hotspot-gc-dev mailing list