JDK-8255851 : Footprint regression 13% (increase) in JFR-Footprint3-Noop-Base-G1 after JDK-8245194
  • Type: Bug
  • Component: core-libs
  • Sub-Component: java.nio
  • Affected Version: 16
  • Priority: P4
  • Status: Resolved
  • Resolution: Cannot Reproduce
  • Submitted: 2020-11-03
  • Updated: 2020-12-01
  • Resolved: 2020-11-30
Related Reports
Relates :  
JDK-8245194 - Unix domain socket channel implementation
Description
This regression is only in the JFR version, not in the plain Footprint3-Noop-Base-ParGC.
Comments
AFAICT this regression is happening because this change randomly causes a cocnsitent but big change in the complation order and there seems to be at least 1 more method getting compiled with a lot of inlining. This activity s what causes the increase, which is Arena memory according to NMT. I will update this more when I have more time to work on it. But it is clear to me that this code change of jdk-16+22-1323 is not at fault, its just bad luck.
01-12-2020
Have not been able to reproduce the regression when running the test comparing the two builds b21 and b22. If any further evidence of a problem can be found, the bug can be re-opened.
30-11-2020