RFR (M): 8141141: Young and Old gen PLAB stats are similar in output with -XX:+PrintPLAB
Thomas Schatzl
thomas.schatzl at oracle.com
Tue Mar 1 17:35:37 UTC 2016
Hi all,
can I get reviews for the following change that tries to improve PLAB
logging output.
In particular, instead of:
[2,245s][debug ][gc,plab] GC(33) (allocated = 1 wasted = 0 unused = 0
used = 1 undo_waste = 0 region_end_waste = 0 regions filled = 0
direct_allocated = 0 failure_used = 0 failure_waste = 0) (plab_sz = 0
desired_plab_sz = 258)
[2,245s][debug ][gc,plab] GC(33) (allocated = 1 wasted = 0 unused = 0
used = 1 undo_waste = 0 region_end_waste = 0 regions filled = 0
direct_allocated = 0 failure_used = 0 failure_waste = 0) (plab_sz = 0
desired_plab_sz = 258)
The logging looks like:
[0.330s][debug][gc,plab ] GC(0) Young PLAB allocation: allocated:
1825632B, wasted: 29424B, unused: 2320B, used: 1793888B, undo waste:
0B,
[0.330s][debug][gc,plab ] GC(0) Young other allocation: region end
waste: 0B, regions filled: 2, direct allocated: 271520B, failure used:
0B, failure wasted: 0B
[0.330s][debug][gc,plab ] GC(0) Young sizing: calculated: 358776B,
actual: 358776B
[0.330s][debug][gc,plab ] GC(0) Old PLAB allocation: allocated:
427248B, wasted: 592B, unused: 368584B, used: 58072B, undo waste: 0B,
[0.330s][debug][gc,plab ] GC(0) Old other allocation: region end
waste: 0B, regions filled: 1, direct allocated: 41704B, failure used:
0B, failure wasted: 0B
[0.330s][debug][gc,plab ] GC(0) Old sizing: calculated: 11608B,
actual: 11608B
I.e. with the following changes:
- add "Young"/"Old" identifier to show "generation"
- split information into multiple lines; the "other allocation" and
"sizing" are optional. "other allocation" is only printed when using
G1, "sizing" only when ResizePLAB is enabled
- values are in bytes now (with units), not heapwords
I think it is a big improvement.
CR:
https://bugs.openjdk.java.net/browse/JDK-8141141
Webrev:
http://cr.openjdk.java.net/~tschatzl/8141141/webrev/
Testing:
jprt, existing plab logging test cases
More information about the hotspot-gc-dev
mailing list