-
Enhancement
-
Resolution: Fixed
-
P4
-
23
-
b25
Parallel full-GC employs two bitmaps to track object start and end points. This configuration necessitates a larger memory footprint, with each bitmap consuming approximately 1.5% of the heap size.
Transitioning to the mark-compact algorithm (the same one used in Serial/G1 full-GC) can mitigate the memory footprint and bring it more in line with other garbage collectors.
Under the new algorithm, after marking, three phases ensue: calculating new addresses for objects, forwarding them, and relocating objects to their new addresses.
The relocation phase will largely reuse the logic and implementation from the existing algorithm.
Transitioning to the mark-compact algorithm (the same one used in Serial/G1 full-GC) can mitigate the memory footprint and bring it more in line with other garbage collectors.
Under the new algorithm, after marking, three phases ensue: calculating new addresses for objects, forwarding them, and relocating objects to their new addresses.
The relocation phase will largely reuse the logic and implementation from the existing algorithm.
- duplicates
-
JDK-8320165 Parallel: Full GC code is very slow due to quadratic calc_new_address
- Closed
- relates to
-
JDK-8332961 Parallel: Limit PSParallelCompact::verify_complete range
- Resolved
-
JDK-8263731 [REDO] MoveAndUpdateClosure::do_addr calls function with side-effects in an assert
- Closed