GC duration and pause time grow linearly

[Aleksey Shipilev]

On 6/1/20 9:16 AM, Kirill Ilyukhin wrote:
> Understood about JDK 13. Will try updating to 14u, but it is going to take a month or two to get the
> data because CG/Pause time grows very slowly.

If you have a staging environment for experiments, forcing back-to-back GCs with
-XX:ShenandoahGCHeuristics=aggressive might be interesting to stress out the workload without
waiting for months.

14u nightlies (targeting 14.0.2 to be released in July) have the verbose per-cycle logging (enabled
with -Xlog:gc+stats), and that should point to the which work actually goes up and up:
  https://builds.shipilev.net/openjdk-jdk14/

> Please try the URL below, this is gceasy.io <http://gceasy.io> report I was referring to.
> https://gceasy.io/my-gc-report.jsp?p=c2hhcmVkLzIwMjAvMDYvMS8tLWdjLnN3MTIubG9nLmd6LS02LTU5LTI3&channel=WEB

Thanks! It is weird to see most pauses growing. Something is definitely off.

-- 
-Aleksey