RFR: 8246436: JFR: Avoid parsing metadata.xml during startup
Erik Gahlin
erik.gahlin at oracle.com
Mon Jun 22 07:49:22 UTC 2020
Thanks for all the reviews and the measurements!
Erik
> On 18 Jun 2020, at 13:56, Claes Redestad <claes.redestad at oracle.com> wrote:
>
> Hi Erik,
>
> looks good - and a great improvement!
>
> By my measures this removes about 12% of the total JVM cpu use when
> running a Hello World with -XX:StartFlightRecording - or roughly 35ms
> off the total runtime[1]
>
> /Claes
>
> [1]
> perf stat -r 25 java -XX:StartFlightRecording HelloWorld
>
> Before:
> 2622.928325 task-clock (msec) # 5.092 CPUs utilized ( +- 1.13% )
> 6,654 context-switches # 0.003 M/sec ( +- 2.36% )
> 30 cpu-migrations # 0.012 K/sec ( +- 5.10% )
> 29,573 page-faults # 0.011 M/sec ( +- 1.40% )
> 5,952,854,795 cycles # 2.270 GHz ( +- 1.10% )
> 6,431,402,332 instructions # 1.08 insn per cycle ( +- 0.87% )
> 1,285,805,882 branches # 490.218 M/sec ( +- 0.87% )
> 50,742,706 branch-misses # 3.95% of all branches ( +- 0.81% )
>
> 0.515076132 seconds time elapsed ( +- 1.41% )
>
> After:
> 2300.679150 task-clock (msec) # 4.778 CPUs utilized ( +- 1.28% )
> 6,509 context-switches # 0.003 M/sec ( +- 2.19% )
> 29 cpu-migrations # 0.012 K/sec ( +- 5.23% )
> 28,050 page-faults # 0.012 M/sec ( +- 1.78% )
> 5,177,014,775 cycles # 2.250 GHz ( +- 1.32% )
> 5,573,218,149 instructions # 1.08 insn per cycle ( +- 1.07% )
> 1,116,169,965 branches # 485.148 M/sec ( +- 1.08% )
> 44,986,319 branch-misses # 4.03% of all branches ( +- 0.85% )
>
> 0.481467615 seconds time elapsed ( +- 1.27% )
>
>
> On 2020-06-18 01:16, Erik Gahlin wrote:
>> Hi,
>> Could I have review of an enhancement that will avoid parsing the XML metadata for the JVM events during startup/runtime.
>> Thanks to Magnus for helping out with changes to the make files.
>> Bug:
>> https://bugs.openjdk.java.net/browse/JDK-8246436
>> Webrev:
>> http://cr.openjdk.java.net/~egahlin/8246436/
>> Testing: tier1+tier2 + jdk/jdk/jfr
>> Thanks
>> Erik
More information about the build-dev
mailing list