Feedback from experiments on production application

[Aleksey Shipilev]

On 05/09/2018 07:53 AM, Roman Kennke wrote:
> Would it be possible to run the same experiments with a newer build,
> e.g. one from here:
> 
> https://builds.shipilev.net/

More precisely, this one for 8u:
  https://builds.shipilev.net/openjdk-shenandoah-jdk8/


>> I would suggest mentioning somewhere in the user guide the fact, that the class unloading
>> feature can significantly prolong GC pauses. Also the GC logging could mention that the class
>> unloading is going on (yes, you can see it with class unloading logging on, but most people
>> usually don’t have it on, when trying to fix GC issues).
It does mention it in recent builds, like this:

[Pause Init Mark (unload classes), 3.463 ms]
[Concurrent marking (unload classes) 557M->557M(8192M), 53.294 ms]
[Pause Final Mark (unload classes), 15.750 ms]
[Concurrent cleanup 557M->543M(8192M), 0.288 ms]

Thanks,
-Aleksey